refactor : metric_logger 改用事件类型 (type=step/validation/epoch)

- 每种事件独立 schema，不再混入 null 字段
- 回调顺序 validation 移到 metric_logger 之前，确保 on_optimizer_step 先跑
- 用内部 _last_val_loss 代替 TrainContext.last_val_iter 判断新验证
- 修复 factory.py 未使用导入、evaluate_ifeval.py 多余 f 前缀

CONTRIBUTING.md

@@ -5,7 +5,7 @@ Thank you for your interest in contributing! This document provides step-by-step
 ## Quick Start
 
 ```bash
-git clone https://github.com/your-username/AstrAI.git
+git clone https://github.com/ViperEkura/AstrAI.git
 cd AstrAI
 pip install -e ".[dev]"     # install with dev dependencies (pytest, ruff)
 ```

README.md

@@ -9,9 +9,9 @@
 <div align="center">
   <img src="https://img.shields.io/badge/python-3.12+-blue.svg" alt="python">
   <img src="https://img.shields.io/badge/license-GPL--3.0-blue.svg" alt="license">
-  <img src="https://img.shields.io/github/v/release/ViperEkura/AstrAI?color=76bad9" alt="release">
+  <img src="https://img.shields.io/github/v/release/ViperEkura/AstrAI?label=Release&color=76bad9" alt="release">
-  <img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fapi.github.com%2Frepos%2FViperEkura%2FAstrAI&query=%24.stargazers_count&label=stars&suffix=%20stars&color=76bad9" alt="stars">
+  <img src="https://img.shields.io/github/stars/ViperEkura/AstrAI?style=flat&label=Stars&color=76bad9" alt="stars">
-  <img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fapi.github.com%2Frepos%2FViperEkura%2FAstrAI&query=%24.forks_count&label=forks&suffix=%20forks&color=76bad9" alt="forks">
+  <img src="https://img.shields.io/github/forks/ViperEkura/AstrAI?style=flat&label=Forks&color=76bad9" alt="forks">
 </div>
 <br>
 
@@ -28,7 +28,8 @@
 ## 📖 Table of Contents
 
 - [Features](#features)
-- [Quick Start](#quick-start)
+- [Getting Started](#getting-started)
+- [Demo](#demo)
 - [Documentation](#documentation)
 - [Contributing](#contributing)
 - [Community](#community)
@@ -49,39 +50,50 @@
 - 🤗 **HuggingFace-Style API**: AutoModel/AutoTokenizer APIs inspired by HuggingFace for easy model and tokenizer loading.
 - 🔌 **Dual API Compatibility**: Supports both OpenAI and Anthropic chat completion APIs out of the box.
 
-### Quick Start
+### Getting Started
 
-#### Installation
+End-to-end walkthrough in 5 steps:
+
+**1. Install**
 
 ```bash
 git clone https://github.com/ViperEkura/AstrAI.git
 cd AstrAI
 pip install -e .
+# pip install -e ".[dev]"    # optional: dev dependencies (pytest, ruff)
 ```
 
-For development dependencies:
+**2. Download model**
 
 ```bash
-pip install -e ".[dev]"
+python scripts/demo/download.py    # downloads 1B checkpoint to params/
 ```
 
-#### Download Pre-trained Model
+**3. Preprocess data**
 
-Download pre-trained model weights (1B bilingual checkpoint) to `params/`:
+Create `pretrain.json` (preprocessing config for `seq` strategy):
+
+```json
+{
+    "version": 1,
+    "input": {"sections": [{"field": "text", "action": "train"}]},
+    "preprocessing": {"max_seq_len": 2048},
+    "output": {"storage_format": "bin"}
+}
+```
 
 ```bash
-python scripts/demo/download.py
+python scripts/tools/preprocess.py data/*.jsonl -o output/ -c pretrain.json
 ```
 
-Or download manually from [HuggingFace](https://huggingface.co/ViperEk/KHAOSZ) into `params/`.
+**4. Train**
-
-#### Train a Model
 
 ```bash
 export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 nohup python scripts/tools/train.py \
     --nprocs=4 \
+    --parallel_mode=ddp \
     --train_type=seq \
     --data_root_path=/path/to/dataset \
     --param_path=/path/to/model \
@@ -101,15 +113,54 @@ nohup python scripts/tools/train.py \
     > out.log 2> err.log &
 ```
 
-Full reference at [Parameter Guide](assets/docs/params.md).
+**5. Serve & query**
 
-#### Generate Text
+```bash
+# Terminal 1: start server
+python scripts/tools/server.py --param_path ./params --device cuda
+
+# Terminal 2: query
+curl http://localhost:8000/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"messages":[{"role":"user","content":"Hello"}],"max_tokens":512}'
+```
+
+### Demo
+
+Check out the demos in the `scripts/demo/` folder:
+
+```bash
+# Download model weights (required before running demos)
+python scripts/demo/download.py                      # model → params/
+
+# Interactive streaming chat (multi-turn, maintains history)
+python scripts/demo/stream_chat.py
+# Type your message after >>, type !exit to quit
+
+# Batch generation (5 hardcoded prompts, non-streaming)
+python scripts/demo/generate_batch.py
+
+# Single-prompt autoregressive streaming
+python scripts/demo/generate_ar.py
+```
+
+All generation demos use `temperature=0.8`, `top_p=0.95`, `top_k=50`, `max_tokens=2048` by default and require `params/` to contain model weights (run `download.py` first).
+
+Watch a video walkthrough on [bilibili](https://www.bilibili.com/video/BV1fuLB6yEj6).
+
+---
+
+See [Documentation](#documentation) for full references beyond the examples above.
+
+#### Text Generation
+
+Batch generation from a JSONL file:
 
 ```bash
 python scripts/tools/generate.py \
-    --param_path /path/to/model \
+    --param_path ./params \
-    --input_json_file /path/to/input.json \
+    --input_json_file input.jsonl \
-    --output_json_file /path/to/output.json
+    --output_json_file output.jsonl
 ```
 
 #### Docker
@@ -123,9 +174,6 @@ docker build -t astrai:latest .
 # Run with GPU support
 docker run --gpus all -it astrai:latest
 
-# Run with specific GPUs
-docker run --gpus '"device=0,1"' -it astrai:latest
-
 # Run inference server
 docker run --gpus all -p 8000:8000 astrai:latest \
   python -m scripts.tools.server --port 8000 --device cuda
@@ -142,88 +190,42 @@ docker compose --profile cpu up -d
 
 > **Note**: `--gpus all` is required for CUDA support. Without it, `torch.cuda.is_available()` will return `False`.
 
-#### Start HTTP Server
+#### HTTP API Examples
 
-Start the inference server with OpenAI and Anthropic-compatible HTTP API:
+Additional request examples beyond the [Getting Started](#getting-started) flow:
 
 ```bash
-python -m scripts.tools.server --port 8000 --device cuda
-```
-
-Make requests:
-
-```bash
-# OpenAI-compatible
-curl -X POST http://localhost:8000/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d '{
-    "messages": [{"role": "user", "content": "Hello"}],
-    "max_tokens": 512
-  }'
-
 # OpenAI-compatible streaming
 curl -X POST http://localhost:8000/v1/chat/completions \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"messages":[{"role":"user","content":"Tell a story"}],"stream":true,"max_tokens":500}'
-    "messages": [{"role": "user", "content": "Tell a story"}],
-    "stream": true,
-    "max_tokens": 500
-  }'
 
 # Anthropic-compatible
 curl -X POST http://localhost:8000/v1/messages \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"model":"astrai","system":"You are a helpful assistant.","messages":[{"role":"user","content":"Hello"}],"max_tokens":512}'
-    "model": "astrai",
-    "system": "You are a helpful assistant.",
-    "messages": [{"role": "user", "content": "Hello"}],
-    "max_tokens": 512
-  }'
 
 # Anthropic-compatible streaming with stop sequences
 curl -X POST http://localhost:8000/v1/messages \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"model":"astrai","messages":[{"role":"user","content":"Write a story"}],"max_tokens":500,"stream":true,"stop_sequences":["The end"]}'
-    "model": "astrai",
-    "messages": [{"role": "user", "content": "Write a story"}],
-    "max_tokens": 500,
-    "stream": true,
-    "stop_sequences": ["The end"]
-  }'
 
 # Health check
 curl http://localhost:8000/health
 ```
 
-#### Demo
+See [Inference Guide](assets/docs/inference.md) for SSE streaming format, error codes, and stats endpoint.
-
-Check out the demos in the `scripts/demo/` folder:
-
-```bash
-# Download pre‑processed data (required before running demos)
-python scripts/demo/download.py
-
-# Interactive streaming chat
-python scripts/demo/stream_chat.py
-
-# Batch generation
-python scripts/demo/generate_batch.py
-
-# Auto‑regressive generation
-python scripts/demo/generate_ar.py
-```
-
-Watch a video walkthrough on [bilibili](https://www.bilibili.com/video/BV1fuLB6yEj6).
 
 ### Documentation
 
 | Document | Description |
 |----------|-------------|
-| [Parameter Guide](./assets/docs/params.md) | Training & inference parameters |
+| [CLI Reference](./assets/docs/params.md) | Parameters for all CLI tools (train, server, generate, preprocess) |
 | [Architecture](./assets/docs/architecture.md) | System architecture, class diagram & design patterns |
 | [Training](./assets/docs/training.md) | Training loop, strategies & formulas |
 | [Inference](./assets/docs/inference.md) | KVCache, continuous batching, sampling & HTTP API |
 | [Data Flow](./assets/docs/dataflow.md) | Data pipeline, storage backends & dataset architecture |
+| [Preprocessing](./assets/docs/preprocessing.md) | Declarative JSON-driven data preprocessing |
 
 ### Contributing
 

assets/docs/README-zh-CN.md

@@ -15,9 +15,9 @@
 <div align="center">
   <img src="https://img.shields.io/badge/python-3.12+-blue.svg" alt="python">
   <img src="https://img.shields.io/badge/license-GPL--3.0-blue.svg" alt="license">
-  <img src="https://img.shields.io/github/v/release/ViperEkura/AstrAI?color=76bad9" alt="release">
+  <img src="https://img.shields.io/github/v/release/ViperEkura/AstrAI?label=Release&color=76bad9" alt="release">
-  <img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fapi.github.com%2Frepos%2FViperEkura%2FAstrAI&query=%24.stargazers_count&label=stars&suffix=%20stars&color=76bad9" alt="stars">
+  <img src="https://img.shields.io/github/stars/ViperEkura/AstrAI?style=flat&label=Stars&color=76bad9" alt="stars">
-  <img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fapi.github.com%2Frepos%2FViperEkura%2FAstrAI&query=%24.forks_count&label=forks&suffix=%20forks&color=76bad9" alt="forks">
+  <img src="https://img.shields.io/github/forks/ViperEkura/AstrAI?style=flat&label=Forks&color=76bad9" alt="forks">
 </div>
 
 <br>
@@ -34,7 +34,8 @@
 ## 📖 目录
 
 - [特性](#特性)
-- [快速开始](#快速开始)
+- [快速上手](#快速上手)
+- [演示](#演示)
 - [文档](#文档)
 - [贡献](#贡献)
 - [社区](#社区)
@@ -55,39 +56,50 @@
 - 🤗 **HuggingFace 风格 API**: 类 HuggingFace 的 AutoModel/AutoTokenizer 接口，方便加载模型和分词器。
 - 🔌 **双 API 兼容**: 同时支持 OpenAI 和 Anthropic 聊天补全 API，开箱即用。
 
-### 快速开始
+### 快速上手
 
-#### 安装
+端到端演示，只需 5 步：
+
+**1. 安装**
 
 ```bash
 git clone https://github.com/ViperEkura/AstrAI.git
 cd AstrAI
 pip install -e .
+# pip install -e ".[dev]"    # 可选：开发依赖（pytest, ruff）
 ```
 
-安装开发依赖：
+**2. 下载模型**
 
 ```bash
-pip install -e ".[dev]"
+python scripts/demo/download.py    # 下载 1B 检查点到 params/
 ```
 
-#### 下载预训练模型
+**3. 预处理数据**
 
-下载预训练模型权重（1B 双语检查点）到 `params/` 目录：
+创建 `pretrain.json`（`seq` 策略的预处理配置）：
+
+```json
+{
+    "version": 1,
+    "input": {"sections": [{"field": "text", "action": "train"}]},
+    "preprocessing": {"max_seq_len": 2048},
+    "output": {"storage_format": "bin"}
+}
+```
 
 ```bash
-python scripts/demo/download.py
+python scripts/tools/preprocess.py data/*.jsonl -o output/ -c pretrain.json
 ```
 
-或从 [HuggingFace](https://huggingface.co/ViperEk/KHAOSZ) 手动下载放入 `params/`。
+**4. 训练**
-
-#### 训练模型
 
 ```bash
 export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 nohup python scripts/tools/train.py \
     --nprocs=4 \
+    --parallel_mode=ddp \
     --train_type=seq \
     --data_root_path=/path/to/dataset \
     --param_path=/path/to/model \
@@ -107,15 +119,54 @@ nohup python scripts/tools/train.py \
     > out.log 2> err.log &
 ```
 
-完整参数列表见[参数说明](./params.md)。
+**5. 启动服务并调用**
+
+```bash
+# 终端 1：启动服务
+python scripts/tools/server.py --param_path ./params --device cuda
+
+# 终端 2：发起请求
+curl http://localhost:8000/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{"messages":[{"role":"user","content":"你好"}],"max_tokens":512}'
+```
+
+### 演示
+
+查看 `scripts/demo/` 文件夹中的演示：
+
+```bash
+# 下载模型权重（运行演示前必需）
+python scripts/demo/download.py                      # model → params/
+
+# 交互式流式聊天（多轮对话，保持历史记录）
+python scripts/demo/stream_chat.py
+# 在 >> 后输入消息，输入 !exit 退出
+
+# 批量生成（5 条硬编码提示词，非流式）
+python scripts/demo/generate_batch.py
+
+# 单条提示词自回归流式生成
+python scripts/demo/generate_ar.py
+```
+
+所有生成演示默认使用 `temperature=0.8`、`top_p=0.95`、`top_k=50`、`max_tokens=2048`，需要 `params/` 目录包含模型权重（请先运行 `download.py`）。
+
+观看 [bilibili](https://www.bilibili.com/video/BV1fuLB6yEj6) 上的视频演示。
+
+---
+
+更多选项请参考[文档](#文档)。
 
 #### 文本生成
 
+从 JSONL 文件批量生成：
+
 ```bash
 python scripts/tools/generate.py \
-    --param_path /path/to/model \
+    --param_path ./params \
-    --input_json_file /path/to/input.json \
+    --input_json_file input.jsonl \
-    --output_json_file /path/to/output.json
+    --output_json_file output.jsonl
 ```
 
 #### Docker
@@ -129,9 +180,6 @@ docker build -t astrai:latest .
 # 启用 GPU 运行
 docker run --gpus all -it astrai:latest
 
-# 指定特定 GPU
-docker run --gpus '"device=0,1"' -it astrai:latest
-
 # 运行推理服务
 docker run --gpus all -p 8000:8000 astrai:latest \
   python -m scripts.tools.server --port 8000 --device cuda
@@ -148,88 +196,42 @@ docker compose --profile cpu up -d
 
 > **注意**: 必须使用 `--gpus all` 才能启用 CUDA 支持，否则 `torch.cuda.is_available()` 将返回 `False`。
 
-#### 启动 HTTP 服务
+#### HTTP API 示例
 
-启动推理服务器，支持 OpenAI 和 Anthropic 兼容的 HTTP API：
+除[快速上手](#快速上手)流程外，更多请求示例：
 
 ```bash
-python -m scripts.tools.server --port 8000 --device cuda
-```
-
-发起请求：
-
-```bash
-# OpenAI 兼容
-curl -X POST http://localhost:8000/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d '{
-    "messages": [{"role": "user", "content": "你好"}],
-    "max_tokens": 512
-  }'
-
 # OpenAI 兼容流式
 curl -X POST http://localhost:8000/v1/chat/completions \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"messages":[{"role":"user","content":"讲个故事"}],"stream":true,"max_tokens":500}'
-    "messages": [{"role": "user", "content": "讲个故事"}],
-    "stream": true,
-    "max_tokens": 500
-  }'
 
 # Anthropic 兼容
 curl -X POST http://localhost:8000/v1/messages \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"model":"astrai","system":"你是一个乐于助人的助手。","messages":[{"role":"user","content":"你好"}],"max_tokens":512}'
-    "model": "astrai",
-    "system": "你是一个乐于助人的助手。",
-    "messages": [{"role": "user", "content": "你好"}],
-    "max_tokens": 512
-  }'
 
 # Anthropic 兼容流式并设置停止序列
 curl -X POST http://localhost:8000/v1/messages \
   -H "Content-Type: application/json" \
-  -d '{
+  -d '{"model":"astrai","messages":[{"role":"user","content":"写个故事"}],"max_tokens":500,"stream":true,"stop_sequences":["结束"]}'
-    "model": "astrai",
-    "messages": [{"role": "user", "content": "写个故事"}],
-    "max_tokens": 500,
-    "stream": true,
-    "stop_sequences": ["结束"]
-  }'
 
 # 健康检查
 curl http://localhost:8000/health
 ```
 
-#### 演示
+SSE 流式格式、错误码和统计端点详见[推理文档](./inference.md)。
-
-查看 `scripts/demo/` 文件夹中的演示：
-
-```bash
-# 下载预处理数据（运行演示前必需）
-python scripts/demo/download.py
-
-# 交互式流式聊天
-python scripts/demo/stream_chat.py
-
-# 批量生成
-python scripts/demo/generate_batch.py
-
-# 自回归生成
-python scripts/demo/generate_ar.py
-```
-
-观看 [bilibili](https://www.bilibili.com/video/BV1fuLB6yEj6) 上的视频演示。
 
 ### 文档
 
 | 文档 | 说明 |
 |------|------|
-| [参数说明](./params.md) | 训练与推理参数配置 |
+| [CLI 参考](./params.md) | 所有 CLI 工具参数（训练、服务、生成、预处理） |
 | [架构文档](./architecture.md) | 系统架构、类图与设计模式 |
 | [训练文档](./training.md) | 训练循环、策略与公式 |
 | [推理文档](./inference.md) | KVCache、连续批处理、采样与 HTTP API |
 | [数据流程](./dataflow.md) | 数据管道、存储后端与数据集架构 |
+| [数据预处理](./preprocessing.md) | 声明式 JSON 驱动数据预处理 |
 
 ### 贡献
 

assets/docs/architecture.md

@@ -1,5 +1,12 @@
 # AstrAI Architecture
 
+## Contents
+
+- [Class Diagram](#class-diagram) — Full Mermaid class diagram across 10+ namespaces
+- [Module Overview](#module-overview) — Component inventory per module
+- [Design Patterns](#design-patterns) — 13 documented patterns with classes
+- [Core Relationships](#core-relationships) — 11 key inter-component relationships
+
 ## Class Diagram
 
 ```mermaid
@@ -8,6 +15,8 @@ classDiagram
         class BaseConfig {
             +to_dict() Dict
             +from_dict(d) Self
+            +from_file(path) Self
+            +to_file(path)
         }
 
         class BaseModelConfig {
@@ -17,53 +26,86 @@ classDiagram
         }
 
         class AutoRegressiveLMConfig {
-            +int vocab_size
+            +Optional[int] vocab_size
-            +int dim
+            +Optional[int] dim
-            +int n_layers
+            +Optional[int] n_layers
-            +float norm_eps
+            +Optional[float] norm_eps
-            +int dim_ffn
+            +Optional[int] dim_ffn
             +Optional[bool] tie_weight
             +Optional[dict] rope_scaling
-            +int max_len
+            +Optional[int] max_len
-            +float rope_theta
+            +Optional[float] rope_theta
             +str attn_type
-            +int n_heads
+            +Optional[int] n_heads
-            +int n_kv_heads
+            +Optional[int] n_kv_heads
-            +bool use_qk_norm
+            +Optional[bool] use_qk_norm
-            +bool use_gated_attention
+            +Optional[bool] use_gated_attention
             +Optional[int] kv_lora_rank
             +Optional[int] qk_nope_head_dim
             +Optional[int] qk_rope_head_dim
             +str ffn_type
-            +int n_routed_experts
+            +Optional[int] n_routed_experts
-            +int n_shared_experts
+            +Optional[int] n_shared_experts
-            +int n_activated_experts
+            +Optional[int] n_activated_experts
             +Optional[str] topk_method
         }
 
         class EncoderConfig {
-            +int vocab_size
+            +Optional[int] vocab_size
-            +int dim
+            +Optional[int] dim
-            +int n_layers
+            +Optional[int] n_layers
-            +float norm_eps
+            +Optional[float] norm_eps
-            +int dim_ffn
+            +Optional[int] dim_ffn
-            +int max_len
+            +Optional[int] max_len
-            +float rope_theta
+            +Optional[float] rope_theta
-            +int n_heads
+            +Optional[int] n_heads
-            +int n_kv_heads
+            +Optional[int] n_kv_heads
-            +bool use_qk_norm
+            +Optional[bool] use_qk_norm
-            +bool use_gated_attention
+            +Optional[bool] use_gated_attention
             +Optional[dict] rope_scaling
             +Optional[str] pooling_type
             +Optional[bool] normalize_embeddings
         }
 
         class ConfigFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +load(raw) BaseConfig
         }
 
+        class InputConfig {
+            +Optional[List[Dict]] sections
+            +Optional[Dict[str, Dict]] sources
+        }
+
+        class ProcessingConfig {
+            +int max_seq_len
+            +int min_chars
+            +int max_chars
+            +Optional[int] max_items
+            +str packing_strategy
+            +int max_packed_len
+            +str truncation_mode
+        }
+
+        class OutputConfig {
+            +Optional[str] domain_key
+            +str storage_format
+            +int max_tokens_per_shard
+            +Dict[str, str] dtype
+            +str position_ids_mode
+        }
+
+        class PipelineConfig {
+            +int version
+            +InputConfig input
+            +dict mask
+            +str mask_default
+            +ProcessingConfig preprocessing
+            +OutputConfig output
+            +from_dict(d) Self
+        }
+
         class TrainConfig {
             +Callable[[], nn.Module] model_fn
             +str strategy
@@ -156,13 +198,13 @@ classDiagram
         }
 
         class StoreFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +create(storage_type) Store
         }
 
         class DatasetFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +create(train_type, window_size, stride) BaseDataset
             +load(train_type, load_path, window_size, stride, storage_type) BaseDataset
@@ -185,7 +227,7 @@ classDiagram
     namespace model {
         class AutoModel {
             +BaseModelConfig config
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +get_component_class(name) Type
             +from_pretrained(path, disable_random_init, strict) nn.Module
@@ -312,10 +354,38 @@ classDiagram
         }
     }
 
+    namespace preprocessing {
+        class BaseMaskBuilder {
+            <<abstract>>
+            +build(item, config, tokenizer) Optional[dict]
+        }
+
+        class SectionedMaskBuilder {
+            +SectionRenderer renderer
+            +build(item, config, tokenizer) Optional[dict]
+            +_build_single(item, config, tokenizer) Optional[dict]
+            +_build_multi(item, sources_spec, config, tokenizer) Optional[dict]
+        }
+
+        class Pipeline {
+            +PipelineConfig config
+            +List[str] paths
+            +str output_dir
+            +str tokenizer_path
+            +BaseMaskBuilder mask_builder
+            +PackingStrategy _packer
+            +PositionIdStrategy _position_id
+            +StoreWriter _writer
+            +transform(item) Optional[dict]
+            +run()
+            +_flush(domains, shard_idx)
+        }
+    }
+
     namespace tokenize {
         class AutoTokenizer {
             +vocab_size int
-            +encode(tokens, out_ids, is_pretokenized, add_special_tokens) List[int]
+            +encode(tokens, out_ids, is_pretokenized, add_special_tokens) List
             +decode(tokens, skip_special_tokens) str
             +__getattr__(name) Any (bos_id, eos_id, pad_id, stop_ids)
             +apply_chat_template(messages, system_prompt, tokenize, add_generation_prompt) Union[str, List[int]]
@@ -333,27 +403,26 @@ classDiagram
     }
 
     namespace factory {
-        class Registry {
-            +Dict _entries
-            +register(name, component_cls, category, priority)
-            +get(name) Type
-            +list_names() List[str]
-        }
-
         class BaseFactory {
-            +Registry _registry
+            +Dict _entries
-            +register(name, category, priority) decorator
+            +register(name) decorator
             +create(name, *args, **kwargs) T
             +list_registered() list
         }
+
+        class MaskBuilderFactory {
+            +Dict _entries
+            +register(name) decorator
+            +create(name, *args, **kwargs) BaseMaskBuilder
+        }
     }
 
     namespace trainer {
         class Trainer {
             +TrainConfig train_config
             +List[TrainCallback] callbacks
-            +train(checkpoint)
+            +train(resume_dir)
-            +_get_default_callbacks() List[TrainCallback]
+            -_get_default_callbacks() List[TrainCallback]
         }
 
         class TrainContext {
@@ -383,13 +452,17 @@ classDiagram
         }
 
         class BaseStrategy {
-            +Union[Callable, nn.Module] model
+            +Callable model
+            +Optional[BaseExecutor] executor
+            +Optional[Callable] model_fn
+            +dict extra_kwargs
             +str device
+            +__call__(batch) Tensor
             +compute_loss(batch) Tensor
         }
 
         class StrategyFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +create(train_type, model, device, **kwargs) BaseStrategy
         }
@@ -425,17 +498,20 @@ classDiagram
         class BaseScheduler {
             +get_lr() List[float]
             +step()
+            +state_dict() dict
+            +load_state_dict(d)
         }
 
         class SchedulerFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
-            +create(optimizer, schedule_type, **kwargs) BaseScheduler
+            +create(name, *args, **kwargs) BaseScheduler
         }
 
         class CosineScheduler {
             +int warmup_steps
             +int lr_decay_steps
+            +int total_steps
             +float min_rate
         }
 
@@ -446,6 +522,13 @@ classDiagram
             +int t_mult
         }
 
+        class WSDScheduler {
+            +int warmup_steps
+            +int stable_steps
+            +int decay_steps
+            +float min_rate
+        }
+
         class TrainCallback {
             <<protocol>>
             +on_train_begin(context)
@@ -474,11 +557,11 @@ classDiagram
             +int interval
             +bool weight_only
             +Callable save_extra_fn
-            +_save_checkpoint(context)
+            -_save_checkpoint(context)
             +on_batch_end(context)
             +on_train_end(context)
             +on_error(context)
-            +save_extra(context)$
+            +save_extra(context) dict$
         }
 
         class ProgressBarCallback {
@@ -491,7 +574,7 @@ classDiagram
         }
 
         class MetricLoggerCallback {
-            +str log_dir
+            +Path log_dir
             +int save_interval
             +int log_interval
             +List[str] metrics
@@ -501,12 +584,12 @@ classDiagram
         }
 
         class ValidationCallback {
-            +_run_validation(context)
+            -_run_validation(context)
             +on_optimizer_step(context)
         }
 
         class CallbackFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +create(name, **kwargs) TrainCallback
         }
@@ -517,7 +600,7 @@ classDiagram
             +float weight_decay
             +bool nesterov
             +int ns_steps
-            +float adamw_lr
+            +Optional[float] adamw_lr
             +tuple adamw_betas
             +float adamw_eps
             +float adamw_wd
@@ -634,7 +717,7 @@ classDiagram
         class Task {
             +str task_id
             +List prompt_ids
-            +int max_tokens
+            +Optional[int] max_tokens
             +float temperature
             +float top_p
             +int top_k
@@ -643,8 +726,8 @@ classDiagram
             +int input_tokens
             +int output_tokens
             +float arrival_time
-            +float finish_time
+            +Optional[float] finish_time
-            +Callable stream_callback
+            +Optional[Callable] stream_callback
             +int next_pos
             +is_finished(stop_ids) bool
         }
@@ -671,6 +754,11 @@ classDiagram
             +activate(task)
             +return_to_waiting(tasks)
             +get_active_tasks() List[Task]
+            +has_work() bool
+            +wait_for_tasks(timeout)
+            +get_waiting_tasks() List[Task]
+            +clear_queues()
+            +wake()
             +get_stats() Dict
         }
 
@@ -787,12 +875,13 @@ classDiagram
             +request
             +engine
             +builder: ResponseBuilder
-            +handle() Union[StreamingResponse, Dict]
+            +async handle() Union[StreamingResponse, Dict]
-            -_handle_stream(agen, ctx, stops) StreamingResponse
+            -_handle_stream(agen, ctx, stop_sequences) StreamingResponse
-            -_handle_non_stream(agen, ctx, stops) Dict
+            -async _handle_non_stream(agen, ctx, stop_sequences) Dict
         }
 
         class StopChecker {
+            +__init__(sequences)
             +check(text) Optional[str]
         }
 
@@ -804,9 +893,15 @@ classDiagram
             +int completion_tokens
         }
 
-        class app {
+        class StopInfo {
-            <<singleton>>
+            +Optional[str] matched
-            +FastAPI app
+            +str body
+            +str yielded
+        }
+
+        class get_app {
+            <<module>>
+            +get_app() FastAPI
         }
     }
 
@@ -829,14 +924,14 @@ classDiagram
     }
 
     namespace parallel {
-        class Functions {
+        class setup {
             <<module>>
             +spawn_parallel_fn(func, world_size, backend, master_addr, master_port, device_type, start_method, **kwargs)
-            +setup_parallel(rank, world_size, backend, master_addr, master_port, device_type)
+            +setup_parallel(rank, world_size, backend, master_addr, master_port, device_type) contextmanager
             +get_current_device() str
             +get_world_size() int
             +get_rank() int
-            +only_on_rank(rank, sync) decorator
+            +only_on_rank(rank, sync=False) decorator
         }
 
         class GradientState {
@@ -847,6 +942,7 @@ classDiagram
         class AccumOptimizer {
             +Optimizer optimizer
             +GradientState gradient_state
+            +param_groups (property)
             +step(closure)
             +zero_grad()
             +state_dict() dict
@@ -867,7 +963,7 @@ classDiagram
             +prepare(model, optimizer, dataloader, scheduler) tuple
             +accumulate(model) context manager
             +backward(loss)
-            +unwrap_model(model) nn.Module
+            +unwrap_model(model) dict
             +sync_gradients (property) bool
             +grad_accum_steps (property) int
         }
@@ -876,18 +972,18 @@ classDiagram
         }
 
         class DDPExecutor {
-            +_prepare_model(model) nn.Module
+            -_prepare_model(model) nn.Module
-            +_no_sync(model) context manager
+            -_no_sync(model) context manager
-            +unwrap_model(model) nn.Module
+            +unwrap_model(model) dict
         }
 
         class FSDPExecutor {
-            +_prepare_model(model) nn.Module
+            -_prepare_model(model) nn.Module
-            +unwrap_model(model) nn.Module
+            +unwrap_model(model) dict
         }
 
         class ExecutorFactory {
-            +Registry _registry
+            +Dict _entries
             +register(name) decorator
             +create(parallel_mode, **kwargs) BaseExecutor
         }
@@ -899,11 +995,25 @@ classDiagram
         }
 
         class ColumnParallelLinear {
+            +int in_features
+            +int out_features
+            +int out_features_per_rank
+            +bool gather_results
+            +Parameter weight
+            +Optional[Parameter] bias
             +forward(x) Tensor
+            +load_state_dict(state_dict)
         }
 
         class RowParallelLinear {
+            +int in_features
+            +int out_features
+            +int in_features_per_rank
+            +bool reduce_results
+            +Parameter weight
+            +Optional[Parameter] bias
             +forward(x) Tensor
+            +load_state_dict(state_dict)
         }
     }
 
@@ -916,6 +1026,7 @@ classDiagram
     BaseStrategy <|-- GRPOStrategy
     BaseScheduler <|-- CosineScheduler
     BaseScheduler <|-- SGDRScheduler
+    BaseScheduler <|-- WSDScheduler
     TrainCallback <|-- GradientClippingCallback
     TrainCallback <|-- GradientCheckpointingCallback
     TrainCallback <|-- CheckpointCallback
@@ -931,13 +1042,16 @@ classDiagram
     BaseSamplingStrategy <|-- TemperatureStrategy
     BaseSamplingStrategy <|-- TopKStrategy
     BaseSamplingStrategy <|-- TopPStrategy
-    BaseSamplingStrategy <|-- SamplingPipeline
     ParallelModel <|-- RowParallelLinear
     ParallelModel <|-- ColumnParallelLinear
     AutoModel <|-- AutoRegressiveLM
     AutoModel <|-- EmbeddingEncoder
     BaseConfig <|-- BaseModelConfig
     BaseConfig <|-- TrainConfig
+    BaseConfig <|-- InputConfig
+    BaseConfig <|-- ProcessingConfig
+    BaseConfig <|-- OutputConfig
+    BaseConfig <|-- PipelineConfig
     BaseModelConfig <|-- AutoRegressiveLMConfig
     BaseModelConfig <|-- EncoderConfig
     BaseFactory <|-- AutoModel
@@ -950,11 +1064,13 @@ classDiagram
     BaseFactory <|-- StoreFactory
     BaseFactory <|-- ExecutorFactory
     BaseFactory <|-- ConfigFactory
+    BaseFactory <|-- MaskBuilderFactory
     BaseExecutor <|-- NoneExecutor
     BaseExecutor <|-- DDPExecutor
     BaseExecutor <|-- FSDPExecutor
     ResponseBuilder <|-- OpenAIResponseBuilder
     ResponseBuilder <|-- AnthropicResponseBuilder
+    BaseMaskBuilder <|-- SectionedMaskBuilder
 
     %% --- Composition (strong ownership, part destroyed with whole) ---
     KVCache *-- PagePool
@@ -973,7 +1089,6 @@ classDiagram
     DecoderBlock *-- RMSNorm
     ChatCompletionRequest *-- ChatMessage
     MessagesRequest *-- AnthropicMessage
-    BaseFactory *-- Registry
     BaseExecutor *-- GradientState
     AccumOptimizer o-- GradientState
     AccumScheduler o-- GradientState
@@ -991,9 +1106,13 @@ classDiagram
     KvcacheView o-- Storage
     SamplingPipeline o-- BaseSamplingStrategy
     BaseDataset o-- Store
+    Pipeline o-- PipelineConfig
+    Pipeline o-- BaseMaskBuilder
 
     %% --- Dependency (uses temporarily) ---
     TrainConfig ..> BaseStrategy : selects
+    PipelineConfig ..> MaskBuilderFactory : selects
+    MaskBuilderFactory ..> BaseMaskBuilder : creates
     StrategyFactory ..> BaseStrategy : creates
     SchedulerFactory ..> BaseScheduler : creates
     DatasetFactory ..> BaseDataset : creates
@@ -1046,12 +1165,13 @@ classDiagram
 
 | Module | Components | Description |
 |--------|------------|-------------|
-| **astrai.config** | BaseConfig, BaseModelConfig, AutoRegressiveLMConfig, EncoderConfig, ConfigFactory, TrainConfig | Configuration management (to_dict/from_dict, to_file/from_file) |
+| **astrai.config** | BaseConfig, BaseModelConfig, AutoRegressiveLMConfig, EncoderConfig, ConfigFactory, TrainConfig, PipelineConfig, InputConfig, ProcessingConfig, OutputConfig | Configuration management (to_dict/from_dict, to_file/from_file) |
+| **astrai.preprocessing** | BaseMaskBuilder, MaskBuilderFactory, SectionedMaskBuilder, Pipeline, filter_by_length, PackingStrategy, PackingStrategyFactory, PositionIdStrategy, PositionIdStrategyFactory, StoreWriter, StoreWriterFactory | Declarative JSON-driven data preprocessing |
 | **astrai.dataset** | BaseDataset–GRPODataset, Store–MmapStore, StoreFactory, ResumableDistributedSampler, DatasetFactory | Dataset loading and management |
 | **astrai.serialization** | Checkpoint | Model serialization |
 | **astrai.model** | AutoModel, AutoRegressiveLM, EmbeddingEncoder, DecoderBlock, GQA, MLA, MLP, DeepSeekMoE, AttnFactory, FFNFactory, RMSNorm, Linear, RotaryEmbedding, Embedding | Neural network model |
 | **astrai.tokenize** | AutoTokenizer, ChatTemplate | Tokenizer and chat template |
-| **astrai.trainer** | Trainer, TrainContext, TrainContextBuilder, BaseStrategy–GRPOStrategy, StrategyFactory, BaseScheduler–SGDRScheduler, SchedulerFactory, TrainCallback(Protocol)–ValidationCallback, CallbackFactory, Muon | Training workflow |
+| **astrai.trainer** | Trainer, TrainContext, TrainContextBuilder, BaseStrategy–GRPOStrategy, StrategyFactory, BaseScheduler–WSDScheduler, SchedulerFactory, TrainCallback(Protocol)–ValidationCallback, CallbackFactory, Muon | Training workflow |
 | **astrai.inference** | InferenceEngine, InferenceScheduler, Executor, KVCache–KvcacheView, Allocator–Storage, Task, TaskManager, TaskStatus, GenerationRequest, GenerateResult, BaseSamplingStrategy–SamplingPipeline, ProtocolHandler, ResponseBuilder, OpenAIResponseBuilder, AnthropicResponseBuilder, StopChecker, GenContext, ChatMessage–MessagesRequest, app | Inference service |
 | **astrai.parallel** | spawn_parallel_fn, setup_parallel, get_rank/get_world_size/get_current_device, only_on_rank, BaseExecutor, ExecutorFactory, NoneExecutor, DDPExecutor, FSDPExecutor, GradientState, AccumOptimizer, AccumScheduler, ParallelModel, RowParallelLinear, ColumnParallelLinear | Distributed parallel & gradient accumulation |
 | **astrai.factory** | Registry, BaseFactory[T] | Component registration |
@@ -1062,7 +1182,7 @@ classDiagram
 | Pattern | Classes | Purpose |
 |---------|---------|---------|
 | **Factory** | `AttnFactory`, `FFNFactory`, `StrategyFactory`, `DatasetFactory`, `SchedulerFactory`, `CallbackFactory`, `StoreFactory`, `ConfigFactory`, `ExecutorFactory` | Decorator-based component creation |
-| **Registry** | `BaseFactory`, `Registry` | Component registration with category/priority |
+| **Registry** | `BaseFactory` | Component registration |
 | **Strategy** | `SEQStrategy`, `SFTStrategy`, `DPOStrategy`, `GRPOStrategy` | Training strategy switching |
 | **Strategy (Sampling)** | `TemperatureStrategy`, `TopKStrategy`, `TopPStrategy`, `SamplingPipeline` | Composable logit transformations |
 | **Strategy (API)** | `ResponseBuilder`, `OpenAIResponseBuilder`, `AnthropicResponseBuilder` | HTTP API handler with format hooks |
@@ -1070,14 +1190,14 @@ classDiagram
 | **Observer** | `TrainCallback`, callback implementations | Training process monitoring |
 | **Context** | `TrainContext` | Unified training state bag |
 | **Object Pool** | `Allocator`, `PagePool` | Page-based KV cache with LRU eviction |
-| **Executor** | `BaseExecutor`, `NoneExecutor`, `DDPExecutor` | Gradient accumulation & model distribution |
+| **Executor** | `BaseExecutor`, `NoneExecutor`, `DDPExecutor`, `FSDPExecutor` | Gradient accumulation & model distribution |
 | **Storage** | `Store`, `H5Store`, `MmapStore` | Format-agnostic data access with multi-segment support |
 | **Producer-Consumer** | `InferenceScheduler`, `Task`, queues | Continuous batching |
 | **AutoModel Registry** | `AutoModel`, `AutoRegressiveLM`, `EmbeddingEncoder` | Model-type dynamic loading |
 
 ## Core Relationships
 
-1. **Config → Training**: `TrainConfig` holds model, dataset, optimizer_fn, scheduler_fn, `parallel_mode`, `executor_kwargs`
+1. **Config → Training**: `TrainConfig` holds `model_fn`, `dataset`, `optimizer_fn`, `scheduler_fn`, `parallel_mode`, `executor_kwargs`
 2. **Training Flow**: `Trainer` → `TrainContextBuilder` → `TrainContext`, uses `BaseStrategy` for loss, `BaseExecutor` for gradient accumulation + model distribution
 3. **Strategy Selection**: `StrategyFactory` creates strategy by `train_type`
 4. **Executor Selection**: `ExecutorFactory.create(cfg.parallel_mode, grad_accum_steps=cfg.grad_accum_steps, **cfg.executor_kwargs)` → `NoneExecutor` / `DDPExecutor` / `FSDPExecutor`
@@ -1085,8 +1205,8 @@ classDiagram
 6. **Distributed**: `spawn_parallel_fn` + `setup_parallel` for multi-process DDP
 7. **Dataset Loading**: `DatasetFactory` creates datasets, `Store` (H5Store/MmapStore) loads data with explicit `_length` and multi-segment `_data`
 8. **Checkpoint**: `Checkpoint` saves/loads safetensors + metadata (rank-0 only), extra state saved as `{key}.pt`
-9. **Scheduler**: `SchedulerFactory` creates `CosineScheduler`/`SGDRScheduler`
+9. **Scheduler**: `SchedulerFactory` creates `CosineScheduler`/`SGDRScheduler`/`WSDScheduler`
 10. **AutoModel**: `from_pretrained()` loads `config.json` + `model.safetensors`, `_disable_random_init` replaces `nn.init.*` with no-ops
 11. **Protocols**: `OptimizerProtocol` / `SchedulerProtocol` — structural subtyping for `AccumOptimizer` / `AccumScheduler` wrappers
 
-> Document Update Time: 2026-05-28
+> Document Update Time: 2026-05-30

assets/docs/dataflow.md

@@ -1,17 +1,58 @@
 # Data Flow
 
-This document describes the data pipeline: from raw text to model input tensors.
+This document describes the data pipeline: from raw text to model input tensors. For creating preprocessing configs, see [Preprocessing Guide](preprocessing.md).
+
+## Contents
+
+- [Overview](#overview)
+- [Data Preparation](#data-preparation) — tokenization, format detection, backends
+- [Data Keys by Training Type](#data-keys-by-training-type)
+- [Dataset Architecture](#dataset-architecture)
+- [Sampler](#sampler)
+- [DataLoader](#dataloader)
 
 ## Overview
 
 ```
-Raw Text → AutoTokenizer → Token IDs → .h5/.bin → Dataset → Sampler → DataLoader → Training/Inference
+JSONL Lines → Pipeline (mask builder) → Tokenized Tensors
+                                              ↓
+                                      .h5 or .bin storage
+                                              ↓
+                                      Store.load()
+                                              ↓
+                                      Store.fetch(begin, end, keys)
+                                              ↓
+                                      BaseDataset.__getitem__(idx)
+                                              ↓
+                                      Sampler → DataLoader → Training / Inference
 ```
 
 ## Data Preparation
 
 Raw text is tokenized via `AutoTokenizer.encode()` and saved as HDF5 (`.h5`) or binary (`.bin` + `meta.json`) files with keyed tensor groups.
 
+### Tokenization
+
+The `Pipeline` reads JSONL lines, applies the mask builder (see [Preprocessing](preprocessing.md)), and produces flat token sequences:
+
+```python
+# Per JSONL line: messages → chat template → token IDs + loss mask
+tokens = tokenizer.encode(rendered_text)        # List[int]
+loss_mask = [0, 0, 0, 1, 1, 1, 1, 1, 1]        # 0=masked, 1=train
+# Stored as flat tensors, packed with other lines by packing strategy
+```
+
+The output `meta.json` records the storage format, key names, dtype, total token count, and tensor shapes for each shard.
+
+### Format Detection
+
+`detect_format(load_path)` inspects the directory:
+
+- If `*.h5` files exist → `"h5"` (HDF5 backend)
+- If `*.bin` + `meta.json` files exist → `"bin"` (memory-mapped backend)
+
+### Store Backends
+
 Storage format is auto-detected by `detect_format()`; backends are dispatched via registry:
 
 ```
@@ -19,28 +60,39 @@ StoreFactory.create("h5")  → H5Store
 StoreFactory.create("bin") → MmapStore
 ```
 
-H5 backend supports shared memory via `.share_memory_()`. Bin (mmap) uses OS page-cache sharing natively.
+**H5Store**: Reads HDF5 files, supports `share_memory_()` for multi-process DataLoader workers (copies tensors to shared memory).
+
+**MmapStore**: Memory-maps `.bin` files. OS page cache sharing is native — no explicit `share_memory_()` needed. Uses `torch.from_numpy(np.memmap(...))`.
+
+Both backends normalise tensors into `Store._data[Dict[str, List[Tensor]]]` + `Store._cum[Dict[str, List[int]]]` (cumulative lengths for bisect-based indexing).
 
 ## Data Keys by Training Type
 
 | Type | Storage Keys |
 |------|-------------|
 | `seq` | `sequence` (→ input_ids, target_ids via offset-by-1) |
-| `sft` | `sequence`, `loss_mask` |
+| `sft` | `sequence`, `loss_mask`, `position_ids` |
 | `dpo` | `chosen`, `rejected`, `chosen_mask`, `rejected_mask` |
 | `grpo` | `prompts`, `responses`, `masks`, `rewards` |
 
 ## Dataset Architecture
 
 ```
-DatasetFactory.load(train_type, load_path, window_size, stride, storage_type)
+DatasetFactory.load(train_type, load_path, window_size, stride=None, storage_type=None)
-  → StoreFactory.create(detect_format(path))
+  → BaseDataset.load(load_path, storage_type=None)
-    → Store._data[Dict[str, List[Tensor]]] + _cum[Dict[str, List[int]]]
+    → detect_format(load_path)
-      → BaseDataset.__getitem__(idx)
+    → StoreFactory.create(storage_type)
-        → sliding window [begin, end) via get_index(idx)
+    → Store.load(load_path)
+      → H5Store._normalize() / MmapStore._normalize()
+        → Store._data[Dict[str, List[Tensor]]] + _cum[Dict[str, List[int]]]
+          → BaseDataset.__getitem__(idx)
+            → get_index(idx) → [begin, end)
+            → Store.fetch(begin, end, keys) → Tensor / Dict[str, Tensor]
 ```
 
-`window_size` = max input length, `stride` = step between consecutive samples (defaults to `window_size`).
+`window_size` = max input length, `stride` = step between consecutive samples (defaults to `window_size`, optional). `storage_type` defaults to `None` (auto-detect via `detect_format`).
+
+`Store.fetch(begin, end, keys)` accepts a single key (`str`) returning a `Tensor`, or a list of keys returning `Dict[str, Tensor]`. Internally uses `bisect` across multi-segment tensors. Raises `RuntimeError("Store not loaded")` if called before `load()`.
 
 ## Sampler
 
@@ -54,4 +106,4 @@ DatasetFactory.load(train_type, load_path, window_size, stride, storage_type)
 
 Standard PyTorch `DataLoader` with configurable `batch_size`, `num_workers`, `pin_memory`, `prefetch_factor`. Sampler produces indices; dataloader fetches tensor batches via `__getitem__`.
 
-> Document Update Time: 2026-05-28
+> Document Update Time: 2026-06-19

assets/docs/inference.md

@@ -1,5 +1,16 @@
 # Inference
 
+## Contents
+
+- [KV Cache](#kv-cache)
+- [KVCache System](#kvcache-system)
+- [Continuous Batching](#continuous-batching)
+- [Sampling](#sampling-strategy-pattern)
+- [Protocol Handlers](#protocol-handlers-strategy-pattern)
+- [Engine & GenerateResult](#engine--generateresult)
+- [HTTP API](#http-api) — endpoints, SSE, errors, stats
+- [Engine API](#engine-api)
+
 ## KV Cache
 
 At decode time, only the last query token matters. All previous K/V are cached to avoid recomputation:
@@ -12,7 +23,7 @@ RoPE is applied **before** KV cache write, not after — otherwise position enco
 
 ## KVCache System
 
-Six classes working together:
+Six classes (plus two helpers) working together:
 
 ```
 KVCache (facade)
@@ -43,7 +54,8 @@ KVCache (facade)
 BaseSamplingStrategy (ABC)
   ├── TemperatureStrategy
   ├── TopKStrategy
-  └── TopPStrategy
+  ├── TopPStrategy
+  └── SamplingPipeline
 ```
 
 `SamplingPipeline` composes them: Temperature → Top-K → Top-P → softmax → multinomial.  
@@ -73,7 +85,9 @@ Adding a protocol = one builder file, no handler subclassing needed.
 InferenceEngine
   ├── generate(prompt, stream, ...) → str | List[str] | Generator
   ├── generate_with_request(req)    → same
-  └── generate_async(prompt, ...)   → AsyncGenerator
+  ├── generate_async(prompt, ...)   → AsyncGenerator
+  ├── get_stats()                   → Dict
+  └── shutdown()
 ```
 
 `GenerateResult` uses `Condition` for non-streaming (`wait_completion()`) and `Event` for streaming (`wait()`). Stream callback is `cb(token)`.
@@ -124,12 +138,98 @@ Supports `stop_sequences` and streaming via `event: content_block_delta`.
 | Param | Type | Default | Description |
 |-------|------|---------|-------------|
 | `messages` | List[dict] | required | Chat messages (role, content) |
-| `temperature` | float | 1.0 | Sampling temperature (>= 0.0) |
-| `top_p` | float | 1.0 | Nucleus threshold |
 | `top_k` | int | 50 | Top-k count |
+| `top_p` | float | 1.0 | Nucleus threshold |
+| `temperature` | float | 1.0 | Sampling temperature (> 0.0) |
 | `max_tokens` | Optional[int] | None | Max generation length |
 | `stream` | bool | False | Stream output |
 
+### SSE Streaming Format
+
+**OpenAI** (`/v1/chat/completions`, `stream=true`):
+
+```
+data: {"id":"chatcmpl-...","object":"chat.completion.chunk","created":...,"model":"astrai",
+       "choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}
+
+data: {"id":"chatcmpl-...","object":"chat.completion.chunk",...,
+       "choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}
+
+data: {"id":"chatcmpl-...","object":"chat.completion.chunk",...,
+       "choices":[{"index":0,"delta":{},"finish_reason":"stop"}],
+       "usage":{"prompt_tokens":5,"completion_tokens":1,"total_tokens":6}}
+
+data: [DONE]
+```
+
+**Anthropic** (`/v1/messages`, `stream=true`):
+
+```
+event: message_start
+data: {"type":"message_start","message":{"id":"msg_...","model":"astrai","role":"assistant",
+       "content":[],"stop_reason":null,...}}
+
+event: content_block_start
+data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}
+
+event: content_block_delta
+data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Hello"}}
+
+event: content_block_stop
+data: {"type":"content_block_stop","index":0}
+
+event: message_delta
+data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{...}}
+
+event: message_stop
+data: {"type":"message_stop"}
+```
+
+### Error Responses
+
+All endpoints use standard HTTP status codes:
+
+| Status | Meaning |
+|--------|---------|
+| 200 | Success |
+| 400 | Invalid request (bad JSON, missing fields, validation error) |
+| 405 | Method not allowed |
+| 422 | Unprocessable entity (Pydantic validation) |
+| 500 | Internal server error (model crash, OOM, scheduler failure) |
+| 503 | Service unavailable (model not loaded, engine not ready) |
+
+Error response body:
+
+```json
+{
+    "error": {
+        "message": "Invalid request: max_tokens must be > 0",
+        "type": "invalid_request_error",
+        "code": 400
+    }
+}
+```
+
+### Stats Endpoint
+
+```
+GET /stats
+```
+
+Response:
+
+```json
+{
+    "active_requests": 3,
+    "waiting_requests": 2,
+    "total_requests": 128,
+    "cache_usage": 0.45,
+    "tokens_generated": 10240
+}
+```
+
+`cache_usage` is the fraction of KV cache pages currently in use (0.0–1.0).
+
 ## Engine API
 
 ```python
@@ -142,7 +242,8 @@ engine.generate("Hello", stream=True)           # -> Generator[str]
 engine.generate(["A", "B"], stream=True)        # -> Generator[Tuple[int, str]]
 
 # Async
-await engine.generate_async("Hello", ...)       # -> AsyncGenerator[str]
+async for token in engine.generate_async("Hello", ...):    # -> AsyncGenerator[str]
+    print(token)
 ```
 
-> Document Update Time: 2026-05-28
+> Document Update Time: 2026-06-19

assets/docs/params.md

@@ -1,4 +1,11 @@
-# Parameter Documentation
+# CLI Parameter Reference
+
+## Contents
+
+- [Training Parameters](#training-parameters)
+- [Inference Server](#inference-server-serverpy)
+- [Generate](#generate-generatepy)
+- [Preprocess](#preprocess-preprocesspy)
 
 ## Training Parameters
 
@@ -48,6 +55,27 @@
 | `--start_epoch` | Resume from epoch (0 = from scratch) | 0 |
 | `--start_batch` | Resume from batch iteration | 0 |
 
+### Validation
+
+| Parameter | Description | Default |
+|-----------|-------------|---------|
+| `--val_split` | Ratio to split from training dataset for validation (e.g. 0.05) | None |
+| `--val_step` | Number of optimizer steps between validation runs | 1000 |
+
+### Logging
+
+| Parameter | Description | Default |
+|-----------|-------------|---------|
+| `--log_dir` | Directory for metric logs | checkpoint/logs |
+| `--log_interval` | Number of batch iterations between metric logs | 100 |
+| `--metrics` | Metrics to log (e.g. --metrics loss lr val_loss) | ["loss", "lr"] |
+
+### Gradient Checkpointing
+
+| Parameter | Description | Default |
+|-----------|-------------|---------|
+| `--gradient_checkpointing` | Enable activation checkpointing for DecoderBlock modules | False |
+
 ### Distributed Training
 
 | Parameter | Description | Default |
@@ -56,17 +84,21 @@
 | `--parallel_mode` | Parallel strategy (`none`, `ddp`, or `fsdp`) | none |
 | `--device_type` | Device type | cuda |
 | `--start_method` | Multiprocessing start method (`spawn`, `fork`, `forkserver`) | spawn |
+| `--backend` | Distributed training backend | nccl |
+| `--master_addr` | Master node address | localhost |
+| `--master_port` | Master node port | 29500 |
 
 ### Strategy-specific
 
 | Parameter | Description | Default | Used by |
 |-----------|-------------|---------|---------|
 | `--dpo_beta` | DPO beta value | 0.1 | `dpo` |
-| `--label_smoothing` | Label smoothing for cross-entropy loss | 0.05 | `seq`, `sft` |
+| `--label_smoothing` | Label smoothing for cross-entropy loss | 0.0 | `seq`, `sft` |
 | `--group_size` | GRPO group size | 4 | `grpo` |
 | `--grpo_clip_eps` | GRPO clipping epsilon | 0.2 | `grpo` |
 | `--grpo_kl_coef` | GRPO KL penalty coefficient | 0.01 | `grpo` |
 | `--grpo_sync_interval` | GRPO ref_model sync interval (steps) | 200 | `grpo` |
+| `--neftune_alpha` | NEFTune noise alpha (0=disabled, typical: 5.0) | 0.0 | `sft` |
 
 ### Usage Example
 
@@ -75,6 +107,7 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 nohup python scripts/tools/train.py \
     --nprocs=4 \
+    --parallel_mode=ddp \
     --train_type=seq \
     --data_root_path=/path/to/dataset \
     --param_path=/path/to/model \
@@ -96,4 +129,64 @@ nohup python scripts/tools/train.py \
 
 ---
 
-> Document Update Time: 2026-05-24
+## Inference Server (`server.py`)
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `--host` | str | `0.0.0.0` | Host address |
+| `--port` | int | `8000` | Port number |
+| `--param_path` | path | `project_root/params` | Path to model parameters |
+| `--device` | str | `cuda` | Device to load model on |
+| `--dtype` | str | `bfloat16` | Model weights dtype (`bfloat16`, `float16`, `float32`) |
+| `--max_batch_size` | int | `16` | Maximum batch size for continuous batching |
+| `--reload` | flag | `False` | Enable auto-reload for development |
+
+Usage:
+```bash
+python scripts/tools/server.py --param_path ./params --device cuda --dtype bfloat16
+```
+
+See [Inference Guide](inference.md) for HTTP API documentation.
+
+## Generate (`generate.py`)
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `--param_path` | str | required | Path to the model directory |
+| `--input_json_file` | str | required | Path to the input JSONL file |
+| `--output_json_file` | str | required | Path to the output JSONL file |
+| `--question_key` | str | `question` | Key for the question in input JSON |
+| `--response_key` | str | `response` | Key for the response in output JSON |
+| `--temperature` | float | `0.60` | Sampling temperature |
+| `--top_k` | int | `30` | Top-k filtering |
+| `--top_p` | float | `0.95` | Nucleus sampling threshold |
+| `--batch_size` | int | `1` | Batch size for generation |
+| `--max_tokens` | int | model config `max_len` | Maximum tokens to generate |
+
+Usage:
+```bash
+python scripts/tools/generate.py \
+    --param_path ./params \
+    --input_json_file input.jsonl \
+    --output_json_file output.jsonl
+```
+
+## Preprocess (`preprocess.py`)
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `input_files` | path(s) | required | Input JSONL file(s), supports glob (`data/*.jsonl`) |
+| `--output_dir`, `-o` | path | required | Output directory for processed data |
+| `--config`, `-c` | path | required | Preprocessing pipeline config (JSON) |
+| `--num_workers` | int | `4` | Number of parallel workers |
+
+Usage:
+```bash
+python scripts/tools/preprocess.py data/*.jsonl -o output/ -c sft.json
+```
+
+See [Preprocessing Guide](preprocessing.md) for config file format and examples.
+
+---
+
+> Document Update Time: 2026-06-19

assets/docs/preprocessing.md

@@ -0,0 +1,361 @@
+# Preprocessing Pipeline
+
+Declarative JSON-driven data preprocessing. One `SectionedMaskBuilder` handles all formats via `input.sections` (single-output) or `input.sources` (multi-output).
+
+## Contents
+
+- [Philosophy](#philosophy)
+- [Config Structure](#config-structure)
+- [Quick Start](#quick-start) — SFT Chat, SFT Instruction, Pretrain, DPO, GRPO examples
+- [Configuration Reference](#configuration-reference) — all fields
+- [Mask Algorithm](#mask-algorithm)
+- [Output Layout](#output-layout)
+- [CLI](#cli)
+- [Python API](#python-api)
+
+## Philosophy
+
+| Component | Responsibility |
+|-----------|---------------|
+| `tokenizer_config.json` (`chat_template`) | Formatting -- how roles become tokens |
+| `pipeline.json` (`mask`) | Masking -- which roles participate in training |
+
+A single config file captures the entire pipeline, reusable and version-controllable.
+
+## Config Structure
+
+```json
+{
+  "input":         {},   // sections (single) or sources (multi)
+  "mask":          {},   // role → "train" | "mask"
+  "mask_default":  "mask",
+  "preprocessing": {},
+  "output":        {}
+}
+```
+
+### Section Fields
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `field` | str | -- | JSONL key to read |
+| `action` | str | -- | `"train"` / `"mask"` / `"$role"` |
+| `template` | bool | `false` | Apply `chat_template` per message |
+| `add_special_tokens` | bool | `true` for first non-template section | Add special tokens during encode |
+
+### Source Fields (multi-output mode)
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `sections` | list[dict] | -- | Same as single-output section list |
+| `list_field` | bool | `false` | JSONL field holds a list; tokenise each element |
+| `mask_key` | str | `"{key}_mask"` | Explicit output key for loss mask |
+
+---
+
+## Quick Start
+
+### SFT Chat
+
+Input JSONL:
+
+```json
+{"messages": [{"role": "system", "content": "You are helpful."}, {"role": "user", "content": "Hi"}, {"role": "assistant", "content": "Hello!"}]}
+```
+
+Config:
+
+```json
+{
+  "input": {
+    "sections": [
+      {"field": "messages", "action": "$role", "template": true}
+    ]
+  },
+  "mask": {
+    "system": "mask",
+    "user": "mask",
+    "assistant": "train"
+  },
+  "mask_default": "mask",
+  "preprocessing": {
+    "max_seq_len": 2048
+  },
+  "output": {
+    "storage_format": "bin",
+    "dtype": {"loss_mask": "bool"}
+  }
+}
+```
+
+Output keys: `sequence` (int32), `loss_mask` (bool)
+
+### SFT Instruction
+
+Input JSONL:
+
+```json
+{"prompt": "Translate to French: Hello", "response": "Bonjour"}
+```
+
+Config:
+
+```json
+{
+  "input": {
+    "sections": [
+      {"field": "prompt",   "action": "mask", "add_special_tokens": true},
+      {"field": "response", "action": "train"}
+    ]
+  },
+  "mask_default": "mask",
+  "preprocessing": {
+    "max_seq_len": 2048
+  }
+}
+```
+
+Output keys: `sequence`, `loss_mask`
+
+### Pretrain
+
+Input JSONL:
+
+```json
+{"text": "Artificial Intelligence is a field of computer science..."}
+```
+
+Config:
+
+```json
+{
+  "input": {
+    "sections": [
+      {"field": "text", "action": "train"}
+    ]
+  },
+  "preprocessing": {
+    "max_seq_len": 8192,
+    "min_chars": 100
+  }
+}
+```
+
+Output keys: `sequence` (no `loss_mask` — all tokens trained)
+
+### DPO
+
+Input JSONL:
+
+```json
+{"chosen": [{"role": "user", "content": "What is 2+2?"}, {"role": "assistant", "content": "4"}], "rejected": [{"role": "user", "content": "What is 2+2?"}, {"role": "assistant", "content": "5"}]}
+```
+
+Config:
+
+```json
+{
+  "input": {
+    "sources": {
+      "chosen": {
+        "sections": [
+          {"field": "chosen", "action": "$role", "template": true}
+        ]
+      },
+      "rejected": {
+        "sections": [
+          {"field": "rejected", "action": "$role", "template": true}
+        ]
+      }
+    }
+  },
+  "mask": {
+    "user": "mask",
+    "assistant": "train"
+  },
+  "mask_default": "mask"
+}
+```
+
+Output keys: `chosen`, `chosen_mask`, `rejected`, `rejected_mask`
+
+### GRPO
+
+Input JSONL:
+
+```json
+{"prompt": [{"role": "user", "content": "What is 2+2?"}], "responses": ["4", "Five", "Four"], "rewards": [1.0, 0.3, 0.8]}
+```
+
+Config:
+
+```json
+{
+  "input": {
+    "sources": {
+      "prompts": {
+        "sections": [
+          {"field": "prompt", "action": "mask", "template": true}
+        ]
+      },
+      "responses": {
+        "sections": [
+          {"field": "responses", "action": "train"}
+        ],
+        "list_field": true,
+        "mask_key": "masks"
+      },
+      "rewards": {
+        "sections": [
+          {"field": "rewards", "action": "value"}
+        ]
+      }
+    }
+  },
+  "mask": {
+    "user": "mask",
+    "assistant": "train"
+  },
+  "mask_default": "mask"
+}
+```
+
+Output keys: `prompts`, `responses`, `masks`, `rewards` (float32)
+
+- `action: "value"` — extract raw values from JSONL without tokenisation
+- `list_field: true` — tokenise each list element independently, then concatenate
+- `mask_key: "masks"` — rename the auto-generated mask key (default: `responses_mask`)
+
+---
+
+## Configuration Reference
+
+### `input`
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `sections` | list[dict] or null | `null` | Section specs for single-output mode |
+| `sources` | dict[str, dict] or null | `null` | Source specs for multi-output mode (DPO/GRPO) |
+
+When `sources` is set, `sections` is ignored.
+
+### `mask`
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `mask` | dict | `{}` | `{role: "train" \| "mask"}` |
+| `mask_default` | str | `"mask"` | Default action for unlisted roles |
+
+### `preprocessing`
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `max_seq_len` | int | `2048` | Truncate sequences to this length |
+| `min_chars` | int | `50` | Skip text-mode items shorter than this |
+| `max_chars` | int | `2000000` | Skip text-mode items longer than this |
+| `max_items` | int or null | `null` | Stop after N documents |
+| `packing_strategy` | str | `"simple"` | Packing strategy: `"simple"`, `"bfd"`, `"bfd_split"` |
+| `max_packed_len` | int | `8192` | Maximum length of a packed bin |
+| `truncation_mode` | str | `"keep_start"` | How to truncate sequences: `"keep_start"` or `"keep_end"` |
+
+### `output`
+
+| Field | Type | Default | Description |
+|-------|------|---------|-------------|
+| `domain_key` | str or null | `null` | JSONL key for domain grouping |
+| `storage_format` | str | `"bin"` | `"bin"` (mmap) or `"h5"` |
+| `max_tokens_per_shard` | int | `100000000` | Flush threshold in cumulative tokens |
+| `dtype` | dict[str, str] | `{}` | Per-key tensor dtype override (e.g. `{"loss_mask": "bool"}`) |
+| `position_ids_mode` | str | `"none"` | How to compute position_ids: `"none"`, `"doc_reset"`, `"continuous"` |
+
+---
+
+## Mask Algorithm
+
+### Template mode (`template: true`)
+
+For each message in the field's array:
+
+1. Prepend BOS token (masked)
+2. Render through `chat_template` for that single message
+3. Encode rendered text
+4. Apply mask rule for the message's role
+
+### Non-template mode
+
+Encode the field value as text. Mask value is 1 (train) or 0 (mask) per the section's `action`.
+
+### Text config detection
+
+When no section uses `template` and all sections have `action: "train"`, the builder skips mask generation entirely — all tokens are trained.
+
+---
+
+## Output Layout
+
+### Single-Shard (`bin`)
+
+```
+output/
+  __default__/
+    meta.json
+    sequence.bin
+    loss_mask.bin
+  wiki/
+    meta.json
+    sequence.bin
+    loss_mask.bin
+```
+
+### Multi-Shard (`bin`)
+
+When `max_tokens_per_shard` is exceeded:
+
+```
+output/
+  __default__/
+    shard_0000/
+      meta.json
+      sequence.bin
+      loss_mask.bin
+    shard_0001/
+      meta.json
+      sequence.bin
+      loss_mask.bin
+```
+
+`MmapStore` discovers all shards under the domain directory via `rglob("meta.json")`.
+
+---
+
+## CLI
+
+```bash
+# SFT
+python scripts/tools/preprocess.py data/sft/*.jsonl -o output/sft/ -c configs/sft_chat.json
+
+# DPO
+python scripts/tools/preprocess.py data/dpo/*.jsonl -o output/dpo/ -c configs/dpo.json --tokenizer_path params
+
+# GRPO
+python scripts/tools/preprocess.py data/grpo/*.jsonl -o output/grpo/ -c configs/grpo.json
+```
+
+---
+
+## Python API
+
+```python
+from astrai.preprocessing.pipeline import Pipeline
+from astrai.config.preprocess_config import PipelineConfig
+
+config = PipelineConfig.from_json("sft.json")
+Pipeline(
+    config,
+    ["data_part1.jsonl", "data_part2.jsonl"],
+    output_dir="output/",
+    tokenizer_path="params",
+).run()
+```
+
+> Document Update Time: 2026-06-03

assets/docs/training.md

@@ -1,37 +1,17 @@
 # Training
 
-## Model Architecture
+## Contents
 
-The model uses a decoder-only Transformer with **GQA** (Grouped Query Attention) and optional **MLA** (Multi-head Latent Attention). 1.0 billion parameters, Chinese–English bilingual.
+- [Autoregression](#autoregression)
-
+- [Causal Mask](#causal-mask)
-```mermaid
+- [Rotary Position Embedding (RoPE)](#rotary-position-embedding-rope)
-flowchart TB
+- [Training Loop](#training-loop)
-    subgraph Layers["Transformer Layers"]
+- [Strategies](#strategies) — SEQ, SFT, DPO, GRPO
-        direction TB
+- [LR Schedulers](#lr-schedulers)
-        A[Input Embedding] --> B[Transformer Block\nLayer 1]
+- [Gradient Checkpointing](#gradient-checkpointing)
-        B --> C[Transformer Block\nLayer ...]
+- [Checkpoint](#checkpoint)
-        C --> D[Transformer Block\nLayer ...]
+- [TrainContextBuilder](#traincontextbuilder-builder-pattern)
-        D --> E[RMSNorm]
+- [Training CLI](#training-cli)
-        E --> F[Linear]
-        F --> G[SoftMax]
-    end
-
-    subgraph TransformerBlock["Transformer Block"]
-        direction TB
-        H[x] --> I[RMSNorm]
-        I --> J[Linear → Q/K/V]
-        J --> K[Q]; J --> L[K]; J --> M[V]
-        K --> N[RoPE]; L --> O[RoPE]
-        N --> P["Q @ K^T / sqrt(d)"]; O --> P
-        P --> Q[Masked SoftMax]; Q --> R[S @ V]; M --> R
-        R --> S[Linear]; S --> T[+]; H --> T
-        T --> U[RMSNorm]
-        U --> V["Linear (gate)"]; U --> W["Linear (up)"]
-        V --> X[SiLU]; X --> Y[×]; W --> Y
-        Y --> Z["Linear (down)"]; Z --> AA[+]; T --> AA
-        AA --> BB[x']
-    end
-```
 
 ### Autoregression
 
@@ -69,14 +49,16 @@ Two-level loop: **epoch** → **batch**. Optimizer step fires every `grad_accum_
 
 ```
 on_train_begin
+  model.train()
   on_epoch_begin
     for batch in dataloader:
       on_batch_begin
       with executor.accumulate(model):
-        loss = strategy(batch)
+        loss = strategy.compute_loss(batch)
+        context.loss = loss.item()
         stand_loss = loss / executor.grad_accum_steps
         executor.backward(stand_loss)
-        iteration += 1
+        context.iteration += 1
         on_batch_end
 
         if executor.sync_gradients:
@@ -94,9 +76,13 @@ on_train_end
 | Hook | Fires | Default callback |
 |------|-------|-----------------|
 | `on_train_begin` | Before training starts | `GradientCheckpointingCallback` |
+| `on_epoch_begin` | Start of each epoch | `ProgressBarCallback` |
+| `on_batch_begin` | Every batch | — |
 | `on_optimizer_step` | Every accumulation window | `GradientClippingCallback`, `ValidationCallback` |
 | `on_batch_end` | Every batch | `CheckpointCallback`, `MetricLoggerCallback`, `ProgressBarCallback` |
-| `on_train_end` | Training ends | `CheckpointCallback`, `MetricLoggerCallback` (final save) |
+| `on_epoch_end` | End of each epoch | `ProgressBarCallback` |
+| `on_error` | On exception during training | `CheckpointCallback`, `MetricLoggerCallback` |
+| `on_train_end` | Training ends (always via finally) | `CheckpointCallback`, `MetricLoggerCallback`, `GradientCheckpointingCallback` |
 
 Default callbacks (in order): `gradient_checkpointing` (activation checkpointing, optional), `checkpoint` (safetensors, rank-0), `metric_logger` (JSONL, rank-0), `progress_bar` (tqdm), `gradient_clipping`, `validation` (periodic validation on val_dataset).
 
@@ -110,7 +96,7 @@ $$
 L_{\text{PT}} = -\sum_{t=1}^{T} \log P(x_t \mid x_{\lt t}; \theta)
 $$
 
-Keys: `input_ids`, `target_ids`
+Keys: `input_ids`, `target_ids`. Optional: `label_smoothing`.
 
 ### SFT (Supervised Fine-Tuning)
 
@@ -120,7 +106,7 @@ $$
 L_{\text{SFT}} = -\sum_{t=P+1}^{P+L} \log P(s_t \mid s_{\lt t}; \theta)
 $$
 
-Keys: `input_ids`, `target_ids`, `loss_mask`
+Keys: `input_ids`, `target_ids`, `loss_mask`. Optional: `label_smoothing`.
 
 ### DPO (Direct Preference Optimization)
 
@@ -130,7 +116,7 @@ $$
 L_{\text{DPO}} = -\mathbb{E}\left[\log\sigma\left(\beta\log\frac{\pi_\theta(y_w\mid x)}{\pi_{\text{ref}}(y_w\mid x)} - \beta\log\frac{\pi_\theta(y_l\mid x)}{\pi_{\text{ref}}(y_l\mid x)}\right)\right]
 $$
 
-Parameters: `beta=0.1`. Keys: `chosen`, `rejected`, `chosen_mask`, `rejected_mask`.
+Parameters: `beta=0.1`, `reduction="mean"`. Keys: `chosen`, `rejected`, `chosen_mask`, `rejected_mask`.
 
 ### GRPO (Group Relative Policy Optimization)
 
@@ -144,7 +130,7 @@ $$
 L_{\text{GRPO}} = -\mathbb{E}\left[\min\left(\frac{\pi_\theta}{\pi_{\text{ref}}}A,\; \text{clip}\left(\frac{\pi_\theta}{\pi_{\text{ref}}}, 1-\epsilon, 1+\epsilon\right)A\right)\right] + \lambda \cdot \mathbb{E}\left[(\log\pi_\theta - \log\pi_{\text{ref}})^2\right]
 $$
 
-Parameters: `group_size=4`, `clip_eps=0.2`, `kl_coef=0.01`, `sync_interval=200`.
+Parameters: `group_size=4`, `clip_eps=0.2`, `kl_coef=0.01`, `sync_interval=200`, `reduction="mean"`.
 
 Keys: `prompts`, `responses`, `masks`, `rewards`.
 
@@ -154,8 +140,9 @@ Keys: `prompts`, `responses`, `masks`, `rewards`.
 |------|-------|-------------|
 | Cosine | `CosineScheduler` | Linear warmup → cosine decay to `min_rate` |
 | SGDR | `SGDRScheduler` | Cosine annealing with warm restarts (`t_mult=2`) |
+| WSD | `WSDScheduler` | Warmup-Stable-Decay with sqrt cooldown |
 
-Created by `SchedulerFactory.create(optimizer, schedule_type, **kwargs)`.
+Created by `SchedulerFactory.create(schedule_type, optimizer, **kwargs)`. Valid types: `"cosine"`, `"sgdr"`, `"wsd"`. Omit to use no scheduler.
 
 ## Gradient Checkpointing
 
@@ -172,8 +159,8 @@ Callback wraps each `DecoderBlock.forward` with `torch.utils.checkpoint.checkpoi
 
 ```
 Checkpoint(state_dict, epoch, iteration, extra, meta, config)
-  ├── save(save_dir)    rank-0 only: meta.json (epoch/iteration/timestamp) + config.json (model config) + state_dict.safetensors + optional {key}.pt (optimizer.pt, scheduler.pt)
+  ├── save(save_dir)    rank-0 only: meta.json (epoch/iteration/timestamp) + config.json (model config) + model.safetensors + optional {key}.pt (optimizer.pt, scheduler.pt)
-  └── load(save_dir)    broadcasts metadata from rank-0
+  └── load(save_dir, broadcast=False)    loads from local disk; set broadcast=True to broadcast metadata from rank-0
 ```
 
 Optimizer/scheduler state persisted by default via `Checkpoint.extra`.  
@@ -194,7 +181,7 @@ context = (
 - Creates executor via `ExecutorFactory.create(cfg.parallel_mode, grad_accum_steps=cfg.grad_accum_steps, **cfg.executor_kwargs)`
 - Calls `executor.prepare(model, optimizer, dataloader, scheduler)` for model distribution (e.g. DDP) + gradient accumulation wrappers
 - Creates `ResumableDistributedSampler` for shuffle+resume
-- Builds strategy via `StrategyFactory.create(train_type, ...)`
+- Builds strategy via `StrategyFactory.create(train_type, model, device, **kwargs)`
 
 ## Training CLI
 
@@ -203,6 +190,7 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 nohup python scripts/tools/train.py \
     --nprocs=4 \
+    --parallel_mode=ddp \
     --train_type=seq \
     --data_root_path=/path/to/dataset \
     --param_path=/path/to/model \
@@ -224,4 +212,4 @@ nohup python scripts/tools/train.py \
 
 Full parameter reference at [params.md](params.md).
 
-> Document Update Time: 2026-05-28
+> Document Update Time: 2026-05-30

astrai/__init__.py

@@ -3,32 +3,98 @@ __author__ = "ViperEkura"
 
 from astrai.config import (
     AutoRegressiveLMConfig,
+    BaseModelConfig,
+    ConfigFactory,
     EncoderConfig,
+    PipelineConfig,
     TrainConfig,
 )
-from astrai.dataset import DatasetFactory
+from astrai.dataset import (
+    BaseDataset,
+    DatasetFactory,
+    ResumableDistributedSampler,
+    Store,
+    StoreFactory,
+)
 from astrai.factory import BaseFactory
 from astrai.inference import (
     GenerationRequest,
     InferenceEngine,
+    ProtocolHandler,
+    SamplingPipeline,
+    get_app,
+    run_server,
+    sample,
+)
+from astrai.model import (
+    AutoModel,
+    AutoRegressiveLM,
+    EmbeddingEncoder,
+    LoRAConfig,
+    inject_lora,
+)
+from astrai.parallel import (
+    ExecutorFactory,
+    get_rank,
+    get_world_size,
+    only_on_rank,
+    spawn_parallel_fn,
+)
+from astrai.preprocessing import Pipeline, filter_by_length
+from astrai.serialization import Checkpoint
+from astrai.tokenize import AutoTokenizer, ChatTemplate
+from astrai.trainer import (
+    BaseScheduler,
+    BaseStrategy,
+    CallbackFactory,
+    Muon,
+    SchedulerFactory,
+    StrategyFactory,
+    TrainCallback,
+    Trainer,
 )
-from astrai.model import AutoModel, AutoRegressiveLM
-from astrai.tokenize import AutoTokenizer
-from astrai.trainer import CallbackFactory, SchedulerFactory, StrategyFactory, Trainer
 
 __all__ = [
     "AutoRegressiveLM",
     "AutoRegressiveLMConfig",
-    "EncoderConfig",
+    "AutoModel",
-    "TrainConfig",
-    "DatasetFactory",
     "AutoTokenizer",
+    "BaseDataset",
+    "BaseFactory",
+    "BaseModelConfig",
+    "BaseScheduler",
+    "BaseStrategy",
+    "CallbackFactory",
+    "ChatTemplate",
+    "Checkpoint",
+    "ConfigFactory",
+    "DatasetFactory",
+    "EmbeddingEncoder",
+    "EncoderConfig",
+    "ExecutorFactory",
     "GenerationRequest",
     "InferenceEngine",
-    "Trainer",
+    "LoRAConfig",
-    "CallbackFactory",
+    "Muon",
-    "StrategyFactory",
+    "Pipeline",
+    "PipelineConfig",
+    "ProtocolHandler",
+    "ResumableDistributedSampler",
+    "SamplingPipeline",
     "SchedulerFactory",
-    "BaseFactory",
+    "Store",
-    "AutoModel",
+    "StoreFactory",
+    "StrategyFactory",
+    "TrainCallback",
+    "TrainConfig",
+    "Trainer",
+    "filter_by_length",
+    "get_app",
+    "get_rank",
+    "get_world_size",
+    "inject_lora",
+    "only_on_rank",
+    "run_server",
+    "sample",
+    "spawn_parallel_fn",
 ]

astrai/config/__init__.py

@@ -4,13 +4,22 @@ from astrai.config.model_config import (
     ConfigFactory,
     EncoderConfig,
 )
+from astrai.config.preprocess_config import (
+    InputConfig,
+    OutputConfig,
+    PipelineConfig,
+    ProcessingConfig,
+)
 from astrai.config.train_config import TrainConfig
 
 __all__ = [
-    # Model configuration
     "BaseModelConfig",
     "AutoRegressiveLMConfig",
     "EncoderConfig",
     "ConfigFactory",
     "TrainConfig",
+    "InputConfig",
+    "OutputConfig",
+    "PipelineConfig",
+    "ProcessingConfig",
 ]

astrai/config/base.py

@@ -1,6 +1,7 @@
 import json
 from dataclasses import MISSING, dataclass, fields
-from typing import Any, Dict, Optional, Self, get_type_hints
+from pathlib import Path
+from typing import Any, Dict, Optional, Self, Union, get_type_hints
 
 
 @dataclass
@@ -83,4 +84,15 @@ class BaseConfig:
             return value
         if isinstance(value, target_type):
             return value
+        if isinstance(value, dict) and issubclass(target_type, BaseConfig):
+            return target_type.from_dict(value)
         raise TypeError
+
+    @classmethod
+    def from_file(cls, path: Union[str, Path]) -> Self:
+        with open(path, "r", encoding="utf-8") as f:
+            return cls.from_dict(json.load(f))
+
+    def to_file(self, path: Union[str, Path]):
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(self.to_dict(), f, indent=2, ensure_ascii=False)

astrai/config/model_config.py

@@ -1,6 +1,5 @@
-import json
 from dataclasses import dataclass
-from typing import Any, Dict, Optional, Self
+from typing import Any, Dict, Optional
 
 from astrai.config.base import BaseConfig
 from astrai.factory import BaseFactory
@@ -21,18 +20,7 @@ class BaseModelConfig(BaseConfig):
     """Base config with ``model_type`` dispatch and file I/O."""
 
     model_type: Optional[str] = None
-
+    neftune_alpha: float = 0.0
-    @classmethod
-    def from_file(cls, config_path: str) -> Self:
-        with open(config_path, "r") as f:
-            raw: Dict[str, Any] = json.load(f)
-        return cls.from_dict(raw)
-
-    def to_file(self, config_path: str):
-        d = self.to_dict()
-        config_dict = {k: v for k, v in d.items() if v is not None}
-        with open(config_path, "w") as f:
-            json.dump(config_dict, f, indent=4)
 
 
 @dataclass
@@ -83,10 +71,12 @@ class EncoderConfig(BaseModelConfig):
     rope_theta: Optional[float] = None
     rope_scaling: Optional[dict] = None
 
+    attn_type: str = "gqa"
     n_heads: Optional[int] = None
     n_kv_heads: Optional[int] = None
     use_qk_norm: Optional[bool] = None
     use_gated_attention: Optional[bool] = None
 
+    ffn_type: str = "mlp"
     pooling_type: Optional[str] = None
     normalize_embeddings: Optional[bool] = None

astrai/config/preprocess_config.py

@@ -0,0 +1,109 @@
+"""Pipeline configuration for JSONL preprocessing.
+
+Supports single-sequence (SFT/pretrain) and multi-output (DPO/GRPO)
+modes, both driven declaratively through ``input.sections`` or
+``input.sources``.
+"""
+
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional
+
+from astrai.config.base import BaseConfig
+
+
+@dataclass
+class InputConfig(BaseConfig):
+    """Declarative input mapping.
+
+    Single-output mode (backward-compatible)::
+
+        {"input": {"sections": [{"field": "messages", ...}]}}
+
+    Multi-output mode (DPO / GRPO)::
+
+        {"input": {"sources": {
+            "chosen": {"sections": [{"field": "chosen", ...}]},
+            "rejected": {"sections": [{"field": "rejected", ...}]},
+        }}}
+    """
+
+    sections: Optional[List[Dict]] = None
+    sources: Optional[Dict[str, Dict]] = None
+
+
+@dataclass
+class ProcessingConfig(BaseConfig):
+    """Processing configuration.
+
+    Parameters
+    ----------
+    max_seq_len : int
+        Maximum sequence length (default: 2048).
+    min_chars : int
+        Minimum number of characters to keep (default: 50).
+    max_chars : int
+        Maximum number of characters to keep (default: 2_000_000).
+    max_items : Optional[int]
+        Maximum number of items to process (default: None, unlimited).
+    packing_strategy : str
+        How to pack sequences into a contiguous stream.
+
+        - ``"simple"``: sequential concatenation (default, backward compatible).
+        - ``"bfd"``: best-fit decreasing bin packing, minimises wasted tokens.
+        - ``"bfd_split"``: BFD with over-length sequences split into chunks.
+    max_packed_len : int
+        Maximum length of a packed bin. Sequences longer than this are
+        truncated or split depending on ``packing_strategy`` (default: 8192).
+    truncation_mode : str
+        How to truncate sequences longer than ``max_packed_len``.
+
+        - ``"keep_start"``: keep the first ``max_packed_len`` tokens (default).
+        - ``"keep_end"``: keep the last ``max_packed_len`` tokens.
+    """
+
+    max_seq_len: int = 2048
+    min_chars: int = 50
+    max_chars: int = 2_000_000
+    max_items: Optional[int] = None
+    packing_strategy: str = "simple"
+    max_packed_len: int = 8192
+    truncation_mode: str = "keep_start"
+
+
+@dataclass
+class OutputConfig(BaseConfig):
+    """Output configuration.
+
+    Parameters
+    ----------
+    domain_key : Optional[str]
+        Domain key for the output store (default: None).
+    storage_format : str
+        Storage format, one of ``"bin"``, ``"jsonl"`` (default: ``"bin"``).
+    max_tokens_per_shard : int
+        Maximum tokens per shard before splitting (default: 100_000_000).
+    dtype : Dict[str, str]
+        Per-key dtype overrides, e.g. ``{"input_ids": "int32"}`` (default: {}).
+    position_ids_mode : Optional[str]
+        How to compute position_ids in packed sequences.
+
+         - ``"none"``: do not generate (default).
+        - ``"doc_reset"``: reset to 0 at each document boundary.
+        - ``"continuous"``: sequential 0, 1, 2, ... (pretrain, single doc).
+    """
+
+    domain_key: Optional[str] = None
+    storage_format: str = "bin"
+    max_tokens_per_shard: int = 100_000_000
+    dtype: Dict[str, str] = field(default_factory=dict)
+    position_ids_mode: str = "none"
+
+
+@dataclass
+class PipelineConfig(BaseConfig):
+    version: int = 1
+    input: InputConfig = field(default_factory=InputConfig)
+    mask: Dict[str, str] = field(default_factory=dict)
+    mask_default: str = "mask"
+    preprocessing: ProcessingConfig = field(default_factory=ProcessingConfig)
+    output: OutputConfig = field(default_factory=OutputConfig)

astrai/config/train_config.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass, field, fields
-from typing import Callable, List, Optional
+from typing import Any, Callable, Dict, List, Optional
 
 import torch.nn as nn
 from torch.optim import Optimizer
@@ -40,7 +40,7 @@ class TrainConfig(BaseConfig):
     max_grad_norm: float = field(
         default=1.0, metadata={"help": "Maximum gradient norm."}
     )
-    gradient_checkpointing_modules: list = field(
+    gradient_checkpointing_modules: List[str] = field(
         default_factory=list,
         metadata={"help": "Module types to enable activation checkpointing for."},
     )
@@ -118,16 +118,26 @@ class TrainConfig(BaseConfig):
     val_dataset: Optional[Dataset] = field(
         default=None, metadata={"help": "Dataset for validation."}
     )
+    val_split: Optional[float] = field(
+        default=None,
+        metadata={
+            "help": "Ratio to split from training dataset for validation (e.g. 0.05). Ignored if val_dataset is set."
+        },
+    )
     val_step: int = field(
         default=1000,
         metadata={"help": "Number of optimizer steps between validation runs."},
     )
+    neftune_alpha: float = field(
+        default=0.0,
+        metadata={"help": "NEFTune noise alpha (0=disabled, typical: 5.0)."},
+    )
 
-    executor_kwargs: dict = field(
+    executor_kwargs: Dict[str, Any] = field(
         default_factory=dict,
         metadata={"help": "Extra kwargs passed to ExecutorFactory.create()."},
     )
-    extra_kwargs: dict = field(
+    extra_kwargs: Dict[str, Any] = field(
         default_factory=dict, metadata={"help": "Other arguments."}
     )
 

astrai/dataset/dataset.py

@@ -136,26 +136,6 @@ class DatasetFactory(BaseFactory["BaseDataset"]):
         dataset = DatasetFactory.create("custom", window_size, stride)
     """
 
-    @classmethod
-    def _validate_component(cls, dataset_cls: type):
-        """Validate that the dataset class inherits from BaseDataset."""
-        if not issubclass(dataset_cls, BaseDataset):
-            raise TypeError(f"{dataset_cls.__name__} must inherit from BaseDataset")
-
-    @classmethod
-    def create(cls, train_type: str, window_size: int, stride: int) -> "BaseDataset":
-        """Create a dataset instance.
-
-        Args:
-            train_type: Type of training ("seq", "sft", "dpo", "grpo")
-            window_size: Window size for data sampling
-            stride: Stride between consecutive samples
-
-        Returns:
-            Dataset instance
-        """
-        return super().create(train_type, window_size, stride)
-
     @classmethod
     def load(
         cls,
@@ -185,19 +165,11 @@ class DatasetFactory(BaseFactory["BaseDataset"]):
 
         return dataset
 
-    @classmethod
-    def available_types(cls) -> list:
-        """Return list of registered dataset type names."""
-        return cls.list_registered()
-
 
 @DatasetFactory.register("seq")
 class SEQDataset(BaseDataset):
     """Dataset for sequential next-token prediction training."""
 
-    def __init__(self, window_size: int, stride: int):
-        super().__init__(window_size, stride)
-
     @property
     def required_keys(self) -> List[str]:
         return ["sequence"]
@@ -218,12 +190,9 @@ class SEQDataset(BaseDataset):
 class SFTDataset(BaseDataset):
     """Dataset for supervised fine-tuning with loss masking."""
 
-    def __init__(self, window_size: int, stride: int):
-        super().__init__(window_size, stride)
-
     @property
     def required_keys(self) -> List[str]:
-        return ["sequence", "loss_mask"]
+        return ["sequence", "loss_mask", "position_ids"]
 
     def _fetch_data(self, begin_idx: int, end_idx: int, key: str) -> Tensor:
         return self.storage.fetch(begin_idx, end_idx, key)
@@ -231,24 +200,23 @@ class SFTDataset(BaseDataset):
     def __getitem__(self, index):
         begin_idx, end_idx = self.get_index(index)
 
-        x = self._fetch_data(begin_idx, end_idx, "sequence").to(dtype=torch.long)
+        x = self._fetch_data(begin_idx, end_idx, "sequence")
-        y = self._fetch_data(begin_idx + 1, end_idx + 1, "sequence").to(
+        y = self._fetch_data(begin_idx + 1, end_idx + 1, "sequence")
-            dtype=torch.long
+        position_ids = self._fetch_data(begin_idx, end_idx, "position_ids")
-        )
+        loss_mask = self._fetch_data(begin_idx + 1, end_idx + 1, "loss_mask")
-        loss_mask = self._fetch_data(begin_idx + 1, end_idx + 1, "loss_mask").to(
-            dtype=torch.bool
-        )
 
-        return {"input_ids": x, "target_ids": y, "loss_mask": loss_mask}
+        return {
+            "input_ids": x.to(dtype=torch.long),
+            "target_ids": y.to(dtype=torch.long),
+            "position_ids": position_ids.to(dtype=torch.long),
+            "loss_mask": loss_mask.to(dtype=torch.bool),
+        }
 
 
 @DatasetFactory.register("dpo")
 class DPODataset(BaseDataset):
     """Dataset for Direct Preference Optimization training."""
 
-    def __init__(self, window_size: int, stride: int):
-        super().__init__(window_size, stride)
-
     @property
     def required_keys(self) -> List[str]:
         return ["chosen", "rejected", "chosen_mask", "rejected_mask"]
@@ -280,9 +248,6 @@ class DPODataset(BaseDataset):
 class GRPODataset(BaseDataset):
     """Dataset for Group Relative Policy Optimization training."""
 
-    def __init__(self, window_size: int, stride: int):
-        super().__init__(window_size, stride)
-
     @property
     def required_keys(self) -> List[str]:
         return ["prompts", "responses", "masks", "rewards"]

astrai/dataset/storage.py

@@ -18,6 +18,7 @@ Key properties:
 """
 
 import bisect
+import glob
 import json
 import os
 from abc import ABC, abstractmethod
@@ -113,12 +114,20 @@ def detect_format(load_path: str) -> str:
             return "h5"
         raise ValueError(f"Unsupported file format: {suffix}")
 
-    h5_files = list(root.rglob("*.h5")) + list(root.rglob("*.hdf5"))
+    h5_files = [
+        Path(p)
+        for pattern in ("*.h5", "*.hdf5")
+        for p in glob.glob(str(root / "**" / pattern), recursive=True)
+    ]
     if h5_files:
         return "h5"
-    bin_files = list(root.rglob("*.bin"))
+    bin_files = [Path(p) for p in glob.glob(str(root / "**" / "*.bin"), recursive=True)]
-    if bin_files and (root / "meta.json").exists():
+    if bin_files:
-        return "bin"
+        has_meta = (root / "meta.json").exists() or len(
+            [Path(p) for p in glob.glob(str(root / "**" / "meta.json"), recursive=True)]
+        ) > 0
+        if has_meta:
+            return "bin"
     raise FileNotFoundError(f"No supported data files found at {load_path}")
 
 
@@ -213,11 +222,6 @@ class StoreFactory(BaseFactory["Store"]):
             ...
     """
 
-    @classmethod
-    def _validate_component(cls, store_cls: type):
-        if not issubclass(store_cls, Store):
-            raise TypeError(f"{store_cls.__name__} must inherit from Store")
-
 
 @StoreFactory.register("h5")
 class H5Store(Store):
@@ -244,7 +248,19 @@ class MmapStore(Store):
 
     def load(self, path: str):
         self._mmap_refs = []
-        raw = load_bin(path)
+        root = Path(path)
-        self._normalize(raw)
+        all_raw: Dict[str, List[Tensor]] = {}
+        meta_paths = [
+            Path(p) for p in glob.glob(str(root / "**" / "meta.json"), recursive=True)
+        ]
+        for meta_path in meta_paths:
+            raw = load_bin(str(meta_path.parent))
+            for key, tensors in raw.items():
+                if key not in all_raw:
+                    all_raw[key] = []
+                all_raw[key].extend(tensors)
+        if not meta_paths:
+            raise FileNotFoundError(f"No meta.json found under {path}")
+        self._normalize(all_raw)
         for tensors in self._data.values():
             self._mmap_refs.extend(tensors)

astrai/factory.py

@@ -1,149 +1,103 @@
-"""Base factory class for extensible component registration."""
+"""Base factory with decorator-based registration and kwarg-filtered instantiation."""
 
 import inspect
+import sys
 from abc import ABC
-from typing import Callable, Dict, Generic, List, Optional, Tuple, Type, TypeVar
+from typing import (
+    Callable,
+    Dict,
+    ForwardRef,
+    Generic,
+    List,
+    Optional,
+    Type,
+    TypeVar,
+    Union,
+)
+from typing import get_args as _get_args
+from typing import get_origin as _get_origin
 
 T = TypeVar("T")
 
 
-class Registry:
+def _resolve_type(
-    """Flexible registry for component classes with category and priority support.
+    arg: Union[Type, str, ForwardRef], factory_cls: type
+) -> Optional[Type]:
+    """Resolve a generic type-arg (str forward-ref, ForwardRef, or class)."""
+    if not isinstance(arg, (str, ForwardRef)):
+        return arg
 
-    This registry stores component classes with optional metadata (category, priority).
+    name = arg if isinstance(arg, str) else arg.__forward_arg__
-    It provides methods for registration, retrieval, and listing with filtering.
+    if name == factory_cls.__name__:
-    """
+        return factory_cls
 
-    def __init__(self):
+    mod = sys.modules.get(factory_cls.__module__)
-        self._entries = {}  # name -> (component_cls, category, priority)
+    if mod is None:
+        return None
+    ns = vars(mod)
 
-    def register(
+    if isinstance(arg, ForwardRef):
-        self,
+        return arg._evaluate(ns, None, frozenset(), recursive_guard=frozenset())
-        name: str,
-        component_cls: Type,
-        category: Optional[str] = None,
-        priority: int = 0,
-    ):
-        """Register a component class with optional category and priority."""
-        if name in self._entries:
-            raise ValueError(f"Component '{name}' is already registered")
-        self._entries[name] = (component_cls, category, priority)
 
-    def get(self, name: str) -> Type:
+    return ns.get(name)
-        """Get component class by name."""
-        if name not in self._entries:
-            raise KeyError(f"Component '{name}' not found in registry")
-        return self._entries[name][0]
-
-    def get_with_metadata(self, name: str) -> Tuple[Type, Optional[str], int]:
-        """Get component class with its metadata."""
-        entry = self._entries.get(name)
-        if entry is None:
-            raise KeyError(f"Component '{name}' not found in registry")
-        return entry
-
-    def contains(self, name: str) -> bool:
-        """Check if a name is registered."""
-        return name in self._entries
-
-    def list_names(self) -> List[str]:
-        """Return list of registered component names."""
-        return sorted(self._entries.keys())
-
-    def list_by_category(self, category: str) -> List[str]:
-        """Return names of components belonging to a specific category."""
-        return sorted(
-            name for name, (_, cat, _) in self._entries.items() if cat == category
-        )
-
-    def list_by_priority(self, reverse: bool = False) -> List[str]:
-        """Return names sorted by priority (default ascending)."""
-        return sorted(
-            self._entries.keys(),
-            key=lambda name: self._entries[name][2],
-            reverse=reverse,
-        )
-
-    def entries(self) -> Dict[str, Tuple[Type, Optional[str], int]]:
-        """Return raw entries dictionary."""
-        return self._entries.copy()
 
 
 class BaseFactory(ABC, Generic[T]):
-    """Generic factory class for component registration and creation.
+    """Generic factory with decorator-based component registration.
 
-    This base class provides a decorator-based registration pattern
+        class MyFactory(BaseFactory[MyBase]):
-    for creating extensible component factories.
-
-    Example usage:
-        class MyFactory(BaseFactory[MyBaseClass]):
             pass
 
         @MyFactory.register("custom")
-        class CustomComponent(MyBaseClass):
+        class CustomComponent(MyBase):
             ...
 
-        component = MyFactory.create("custom", *args, **kwargs)
+        obj = MyFactory.create("custom", *args, **kwargs)
+
+    ``create()`` filters kwargs to match the component's ``__init__``
+    signature so components don't need ``**kwargs`` just to absorb
+    unrelated parameters.
     """
 
-    _registry: Registry
+    _entries: Dict[str, Type[T]]
 
     def __init_subclass__(cls, **kwargs):
         super().__init_subclass__(**kwargs)
-        cls._registry = Registry()
+        for orig_base in getattr(cls, "__orig_bases__", ()):
+            if _get_origin(orig_base) is BaseFactory:
+                (arg,) = _get_args(orig_base)
+                cls._entries = {}
+                cls._component_base = _resolve_type(arg, cls)
+                return
 
     @classmethod
-    def register(
+    def register(cls, name: str) -> Callable[[Type[T]], Type[T]]:
-        cls, name: str, category: Optional[str] = None, priority: int = 0
+        """Decorator to register a component class.
-    ) -> Callable[[Type[T]], Type[T]]:
-        """Decorator to register a component class with optional category and priority.
 
-        Args:
+        Validates that the decorated class inherits from the generic
-            name: Registration name for the component
+        type parameter ``T`` declared on the factory.
-            category: Optional category for grouping components
-            priority: Priority for ordering (default 0)
-
-        Returns:
-            Decorator function that registers the component class
-
-        Raises:
-            TypeError: If the decorated class doesn't inherit from the base type
         """
 
         def decorator(component_cls: Type[T]) -> Type[T]:
             cls._validate_component(component_cls)
-            cls._registry.register(
+            if name in cls._entries:
-                name, component_cls, category=category, priority=priority
+                raise ValueError(f"Component '{name}' is already registered")
-            )
+            cls._entries[name] = component_cls
             return component_cls
 
         return decorator
 
     @classmethod
     def create(cls, name: str, *args, **kwargs) -> T:
-        """Create a component instance by name.
+        """Create a component instance by name, filtering kwargs to match
-
+        the component's ``__init__`` signature.
-        Filters kwargs to match the component's __init__ signature,
-        so components don't need to declare **kwargs just to absorb
-        parameters meant for other components.
-
-        Args:
-            name: Registered name of the component
-            *args: Positional arguments passed to component constructor
-            **kwargs: Keyword arguments passed to component constructor
-
-        Returns:
-            Component instance
-
-        Raises:
-            ValueError: If the component name is not registered
         """
-        if not cls._registry.contains(name):
+        entry = cls._entries.get(name)
+        if entry is None:
             raise ValueError(
-                f"Unknown component: '{name}'. "
+                f"Unknown component: '{name}'. Supported types: {sorted(cls._entries)}"
-                f"Supported types: {sorted(cls._registry.list_names())}"
             )
-        component_cls = cls._registry.get(name)
+        component_cls = entry
         sig = inspect.signature(component_cls.__init__)
         has_var_kwargs = any(
             p.kind == inspect.Parameter.VAR_KEYWORD for p in sig.parameters.values()
@@ -159,68 +113,32 @@ class BaseFactory(ABC, Generic[T]):
 
     @classmethod
     def _validate_component(cls, component_cls: Type[T]):
-        """Validate that the component class is valid for this factory.
+        """Validate the decorated class inherits from the factory's base type.
 
-        Override this method in subclasses to add custom validation.
+        Override for custom validation beyond ``issubclass``.
-
-        Args:
-            component_cls: Component class to validate
-
-        Raises:
-            TypeError: If the component class is invalid
         """
-        pass
+        base = cls._component_base
+        if base is not None and not issubclass(component_cls, base):
+            raise TypeError(
+                f"{component_cls.__name__} must inherit from {base.__name__}"
+            )
 
     @classmethod
     def get_component_class(cls, name: str) -> Type[T]:
-        """Get the registered component class by name without instantiating it.
+        """Get the registered component class without instantiating it."""
-
+        entry = cls._entries.get(name)
-        Args:
+        if entry is None:
-            name: Registered name of the component
-
-        Returns:
-            The component class itself
-
-        Raises:
-            ValueError: If the component name is not registered
-        """
-        if not cls._registry.contains(name):
             raise ValueError(
-                f"Unknown component: '{name}'. "
+                f"Unknown component: '{name}'. Supported types: {sorted(cls._entries)}"
-                f"Supported types: {sorted(cls._registry.list_names())}"
             )
-        return cls._registry.get(name)
+        return entry
 
     @classmethod
-    def list_registered(cls) -> list:
+    def list_registered(cls) -> List[str]:
-        """List all registered component names.
+        """List all registered component names."""
-
+        return sorted(cls._entries)
-        Returns:
-            List of registered component names
-        """
-        return cls._registry.list_names()
 
     @classmethod
     def is_registered(cls, name: str) -> bool:
-        """Check if a component name is registered.
+        """Check if a component name is registered."""
-
+        return name in cls._entries
-        Args:
-            name: Component name to check
-
-        Returns:
-            True if registered, False otherwise
-        """
-        return cls._registry.contains(name)
-
-    @classmethod
-    def list_by_category(cls, category: str) -> List[str]:
-        """List registered component names in a category."""
-        return cls._registry.list_by_category(category)
-
-    @classmethod
-    def list_by_priority(cls, reverse: bool = False) -> List[str]:
-        """List registered component names sorted by priority."""
-        return cls._registry.list_by_priority(reverse)
-
-
-__all__ = ["Registry", "BaseFactory"]

astrai/inference/__init__.py

@@ -11,13 +11,18 @@ Layers:
 
 from astrai.inference.api import (
     AnthropicMessage,
+    BaseToolParser,
     ChatCompletionRequest,
     ChatMessage,
+    FunctionDef,
     GenContext,
     MessagesRequest,
     ProtocolHandler,
+    SimpleJsonToolParser,
     StopChecker,
-    app,
+    ToolDef,
+    ToolParserFactory,
+    get_app,
     run_server,
 )
 from astrai.inference.api.anthropic import AnthropicResponseBuilder
@@ -74,12 +79,17 @@ __all__ = [
     "ProtocolHandler",
     "StopChecker",
     "GenContext",
+    "BaseToolParser",
+    "SimpleJsonToolParser",
+    "ToolParserFactory",
     "OpenAIResponseBuilder",
     "AnthropicResponseBuilder",
     "ChatMessage",
     "ChatCompletionRequest",
+    "FunctionDef",
+    "ToolDef",
     "AnthropicMessage",
     "MessagesRequest",
-    "app",
+    "get_app",
     "run_server",
 ]

astrai/inference/api/__init__.py

@@ -1,23 +1,39 @@
-"""Inference API: protocol handler, stop checker, and FastAPI server."""
+"""Inference API: protocol handler, stop checker, tool parsers, and FastAPI server.
+
+``app`` is no longer a module-level global. Use :func:`get_app` to access the
+lazy singleton FastAPI instance.
+"""
 
 from astrai.inference.api.protocol import GenContext, ProtocolHandler, StopChecker
 from astrai.inference.api.server import (
     AnthropicMessage,
     ChatCompletionRequest,
     ChatMessage,
+    FunctionDef,
     MessagesRequest,
-    app,
+    ToolDef,
+    get_app,
     run_server,
 )
+from astrai.inference.api.tool_parser import (
+    BaseToolParser,
+    SimpleJsonToolParser,
+    ToolParserFactory,
+)
 
 __all__ = [
     "ProtocolHandler",
     "StopChecker",
     "GenContext",
+    "BaseToolParser",
+    "SimpleJsonToolParser",
+    "ToolParserFactory",
     "AnthropicMessage",
     "ChatCompletionRequest",
     "ChatMessage",
+    "FunctionDef",
+    "ToolDef",
     "MessagesRequest",
-    "app",
+    "get_app",
     "run_server",
 ]

astrai/inference/api/anthropic.py

@@ -1,5 +1,6 @@
 """Anthropic message completion response builder."""
 
+import time
 import uuid
 from typing import Any, Dict, List, Tuple, Union
 
@@ -39,9 +40,8 @@ class AnthropicResponseBuilder(ResponseBuilder):
         prompt = engine.tokenizer.apply_chat_template(messages, tokenize=False)
         ctx = GenContext(
             resp_id=f"msg_{uuid.uuid4().hex[:24]}",
-            created=0,
+            created=int(time.time()),
             model=request.model,
-            prompt_tokens=0,
         )
         stop_sequences = getattr(request, "stop_sequences", None) or []
         return prompt, ctx, stop_sequences
@@ -72,15 +72,17 @@ class AnthropicResponseBuilder(ResponseBuilder):
             ),
         ]
 
-    def format_chunk(self, token: str) -> str:
+    def format_chunk(self, token: str, **kwargs) -> List[str]:
-        return sse_event(
+        return [
-            {
+            sse_event(
-                "type": "content_block_delta",
+                {
-                "index": 0,
+                    "type": "content_block_delta",
-                "delta": {"type": "text_delta", "text": token},
+                    "index": 0,
-            },
+                    "delta": {"type": "text_delta", "text": token},
-            event="content_block_delta",
+                },
-        )
+                event="content_block_delta",
+            )
+        ]
 
     def format_stream_end(self, ctx: GenContext, stop: StopInfo) -> List[str]:
         events: List[str] = []

astrai/inference/api/openai.py

@@ -1,7 +1,9 @@
 """OpenAI chat completion response builder."""
 
+import logging
+import time
 import uuid
-from typing import Any, Dict, List, Tuple
+from typing import Any, Dict, List, Optional, Tuple, Union
 
 from pydantic import BaseModel
 
@@ -11,24 +13,79 @@ from astrai.inference.api.protocol import (
     StopInfo,
     sse_event,
 )
+from astrai.inference.api.tool_parser import BaseToolParser, ToolParserFactory
 from astrai.inference.engine import InferenceEngine
 
+logger = logging.getLogger(__name__)
+
+_UNSUPPORTED_PARAMS = (
+    "n",
+    "presence_penalty",
+    "frequency_penalty",
+    "logit_bias",
+    "user",
+)
+
+
+def _resolve_tool_choice(
+    request: BaseModel,
+) -> Union[str, Dict[str, Any]]:
+    tc = getattr(request, "tool_choice", None)
+    if tc is None:
+        return "auto"
+    if isinstance(tc, str):
+        return tc
+    if isinstance(tc, dict):
+        return tc
+    return "auto"
+
+
+def _resolve_tools(request: BaseModel) -> Optional[List[Dict[str, Any]]]:
+    raw = getattr(request, "tools", None)
+    if not raw:
+        return None
+    if isinstance(raw, list):
+        return [t.model_dump() if hasattr(t, "model_dump") else t for t in raw]
+    return None
+
 
 class OpenAIResponseBuilder(ResponseBuilder):
     def prepare(
         self, request: BaseModel, engine: InferenceEngine
     ) -> Tuple[str, GenContext, List[str]]:
         messages = [{"role": m.role, "content": m.content} for m in request.messages]
-        prompt = engine.tokenizer.apply_chat_template(messages, tokenize=False)
+        tools = _resolve_tools(request)
+        prompt = engine.tokenizer.apply_chat_template(
+            messages, tokenize=False, tools=tools or []
+        )
 
         self._resp_id = f"chatcmpl-{uuid.uuid4().hex[:12]}"
         self._model = request.model
 
+        for param in _UNSUPPORTED_PARAMS:
+            value = getattr(request, param, None)
+            fields = getattr(type(request), "model_fields", {})
+            default = fields[param].default if param in fields else None
+            if value is not None and value != default:
+                logger.warning(
+                    "ChatCompletionRequest param '%s'=%r is not supported"
+                    " and will be ignored",
+                    param,
+                    value,
+                )
+
+        self._parser: Optional[BaseToolParser] = None
+        if tools:
+            tool_choice = _resolve_tool_choice(request)
+            self._parser = ToolParserFactory.create(
+                "simple_json", tools=tools, tool_choice=tool_choice
+            )
+        self._content_started = False
+
         ctx = GenContext(
             resp_id=self._resp_id,
-            created=0,
+            created=int(time.time()),
             model=self._model,
-            prompt_tokens=0,
         )
         stop = request.stop
         stop_sequences = (
@@ -55,7 +112,82 @@ class OpenAIResponseBuilder(ResponseBuilder):
             )
         ]
 
-    def format_chunk(self, token: str) -> str:
+    def format_chunk(self, token: str, **kwargs) -> List[str]:
+        body = kwargs.get("body", "")
+        if self._parser is not None:
+            return self._format_tool_chunk(body, **kwargs)
+
+        return [
+            sse_event(
+                {
+                    "id": self._resp_id,
+                    "object": "chat.completion.chunk",
+                    "created": 0,
+                    "model": self._model,
+                    "choices": [
+                        {
+                            "index": 0,
+                            "delta": {"content": token},
+                            "finish_reason": None,
+                        }
+                    ],
+                }
+            )
+        ]
+
+    def _format_tool_chunk(self, body: str, **kwargs) -> List[str]:
+        deltas = self._parser.feed(
+            body,
+            current_token_ids=kwargs.get("current_token_ids"),
+            delta_token_ids=kwargs.get("delta_token_ids"),
+        )
+        events: List[str] = []
+        for d in deltas:
+            if "content" in d:
+                if not self._content_started:
+                    events.append(self._role_chunk())
+                    self._content_started = True
+                events.append(
+                    sse_event(
+                        {
+                            "id": self._resp_id,
+                            "object": "chat.completion.chunk",
+                            "created": 0,
+                            "model": self._model,
+                            "choices": [
+                                {
+                                    "index": 0,
+                                    "delta": {"content": d["content"]},
+                                    "finish_reason": None,
+                                }
+                            ],
+                        }
+                    )
+                )
+            elif "tool_calls" in d:
+                if not self._content_started:
+                    events.append(self._role_chunk())
+                    self._content_started = True
+                events.append(
+                    sse_event(
+                        {
+                            "id": self._resp_id,
+                            "object": "chat.completion.chunk",
+                            "created": 0,
+                            "model": self._model,
+                            "choices": [
+                                {
+                                    "index": 0,
+                                    "delta": {"tool_calls": d["tool_calls"]},
+                                    "finish_reason": None,
+                                }
+                            ],
+                        }
+                    )
+                )
+        return events
+
+    def _role_chunk(self) -> str:
         return sse_event(
             {
                 "id": self._resp_id,
@@ -63,12 +195,19 @@ class OpenAIResponseBuilder(ResponseBuilder):
                 "created": 0,
                 "model": self._model,
                 "choices": [
-                    {"index": 0, "delta": {"content": token}, "finish_reason": None}
+                    {
+                        "index": 0,
+                        "delta": {"role": "assistant"},
+                        "finish_reason": None,
+                    }
                 ],
             }
         )
 
     def format_stream_end(self, ctx: GenContext, stop: StopInfo) -> List[str]:
+        finish_reason = "stop"
+        if self._parser is not None and self._parser.has_tool_calls:
+            finish_reason = "tool_calls"
         return [
             sse_event(
                 {
@@ -76,7 +215,9 @@ class OpenAIResponseBuilder(ResponseBuilder):
                     "object": "chat.completion.chunk",
                     "created": ctx.created,
                     "model": self._model,
-                    "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}],
+                    "choices": [
+                        {"index": 0, "delta": {}, "finish_reason": finish_reason}
+                    ],
                 }
             ),
             sse_event(
@@ -91,6 +232,32 @@ class OpenAIResponseBuilder(ResponseBuilder):
     def format_response(
         self, ctx: GenContext, content: str, stop: StopInfo
     ) -> Dict[str, Any]:
+        if self._parser is not None:
+            parsed = self._parser.parse_complete(content)
+            if parsed and parsed.get("tool_calls"):
+                return {
+                    "id": self._resp_id,
+                    "object": "chat.completion",
+                    "created": ctx.created,
+                    "model": self._model,
+                    "choices": [
+                        {
+                            "index": 0,
+                            "message": {
+                                "role": "assistant",
+                                "content": parsed.get("content"),
+                                "tool_calls": parsed["tool_calls"],
+                            },
+                            "finish_reason": "tool_calls",
+                        }
+                    ],
+                    "usage": {
+                        "prompt_tokens": ctx.prompt_tokens,
+                        "completion_tokens": ctx.completion_tokens,
+                        "total_tokens": ctx.prompt_tokens + ctx.completion_tokens,
+                    },
+                }
+
         return {
             "id": self._resp_id,
             "object": "chat.completion",

astrai/inference/api/protocol.py

@@ -35,7 +35,7 @@ class GenContext:
     resp_id: str
     created: int
     model: str
-    prompt_tokens: int
+    prompt_tokens: int = 0
     completion_tokens: int = 0
 
 
@@ -64,7 +64,7 @@ class StopChecker:
 class ResponseBuilder(ABC):
     """Interface for protocol-specific response formatting.
 
-    A new protocol requires one concrete builder implementing 6 methods.
+    A new protocol requires one concrete builder implementing 5 methods.
     """
 
     @abstractmethod
@@ -78,8 +78,15 @@ class ResponseBuilder(ABC):
         """SSE events that open the stream."""
 
     @abstractmethod
-    def format_chunk(self, token: str) -> str:
+    def format_chunk(self, token: str, **kwargs) -> List[str]:
-        """SSE event for a single generated token."""
+        """SSE events for a single generated token.
+
+        ``body`` (the full accumulated text so far) is always provided
+        as a keyword argument. Additional keyword arguments such as
+        ``current_token_ids`` and ``delta_token_ids`` may be included
+        for tool parsers that need token-level information.
+        Returns a list of SSE event strings (may be empty).
+        """
 
     @abstractmethod
     def format_stream_end(self, ctx: GenContext, stop: StopInfo) -> List[str]:
@@ -137,15 +144,25 @@ class ProtocolHandler:
             body = ""
             yielded = ""
             matched = None
+            token_ids: List[int] = []
             async for token in agen:
-                ctx.completion_tokens += 1
                 body += token
 
+                new_ids = self.engine.tokenizer.encode(token)
+                token_ids.extend(new_ids)
+
                 matched = checker.check(body)
                 if matched:
                     break
 
-                yield self.builder.format_chunk(token)
+                ctx.completion_tokens += 1
+                for event in self.builder.format_chunk(
+                    token,
+                    body=body,
+                    current_token_ids=token_ids,
+                    delta_token_ids=new_ids,
+                ):
+                    yield event
                 yielded += token
 
             stop = StopInfo(matched=matched, body=body, yielded=yielded)
@@ -168,7 +185,6 @@ class ProtocolHandler:
         matched = None
 
         async for token in agen:
-            ctx.completion_tokens += 1
             chunks.append(token)
             body += token
 
@@ -176,6 +192,8 @@ class ProtocolHandler:
             if matched:
                 break
 
+            ctx.completion_tokens += 1
+
         content = "".join(chunks)
         stop = StopInfo(matched=matched, body=body)
         return self.builder.format_response(ctx, content, stop)

astrai/inference/api/server.py

@@ -3,6 +3,9 @@ OpenAI / Anthropic-compatible chat completion server backed by continuous-batchi
 
 Protocol-specific formatting is delegated to ``astrai.inference.protocol``.
 This module owns the FastAPI app, request/response schemas, and dependency wiring.
+
+``app`` is lazily constructed — importing this module does NOT create a FastAPI instance.
+Use :func:`get_app` to access the singleton.
 """
 
 import logging
@@ -12,7 +15,7 @@ from typing import Any, Dict, List, Optional, Union
 
 import torch
 import uvicorn
-from fastapi import FastAPI, HTTPException
+from fastapi import APIRouter, FastAPI, HTTPException
 from pydantic import BaseModel, Field
 
 from astrai.inference.api.anthropic import AnthropicResponseBuilder
@@ -24,12 +27,25 @@ from astrai.tokenize import AutoTokenizer
 
 logger = logging.getLogger(__name__)
 
-_project_root = Path(__file__).parent.parent.parent
+_app_instance: Optional[FastAPI] = None
 
 
 class ChatMessage(BaseModel):
     role: str
-    content: str
+    content: Optional[str] = None
+    tool_calls: Optional[List[Dict[str, Any]]] = None
+    tool_call_id: Optional[str] = None
+
+
+class FunctionDef(BaseModel):
+    name: str
+    description: Optional[str] = None
+    parameters: Optional[Dict[str, Any]] = None
+
+
+class ToolDef(BaseModel):
+    type: str = "function"
+    function: FunctionDef
 
 
 class ChatCompletionRequest(BaseModel):
@@ -48,6 +64,8 @@ class ChatCompletionRequest(BaseModel):
     frequency_penalty: Optional[float] = Field(default=0.0, ge=-2.0, le=2.0)
     logit_bias: Optional[Dict[int, float]] = None
     user: Optional[str] = None
+    tools: Optional[List[ToolDef]] = None
+    tool_choice: Optional[Union[str, Dict[str, Any]]] = "auto"
 
 
 class AnthropicMessage(BaseModel):
@@ -84,17 +102,15 @@ async def lifespan(app: FastAPI):
         logger.info("Inference engine shutdown complete")
 
 
-app = FastAPI(title="AstrAI Inference Server", version="0.2.0", lifespan=lifespan)
+router = APIRouter()
 
 
 def _create_engine(
-    param_path: Optional[Path] = None,
+    param_path: Path,
     device: str = "cuda",
     dtype: torch.dtype = torch.bfloat16,
     max_batch_size: int = 16,
 ) -> InferenceEngine:
-    if param_path is None:
-        param_path = _project_root / "params"
     if not param_path.exists():
         raise FileNotFoundError(f"Parameter directory not found: {param_path}")
 
@@ -112,34 +128,50 @@ def _create_engine(
     return engine
 
 
+def get_app() -> FastAPI:
+    """Return the singleton FastAPI instance (lazily created on first call)."""
+    global _app_instance
+    if _app_instance is None:
+        _app_instance = FastAPI(
+            title="AstrAI Inference Server",
+            version="0.2.0",
+            lifespan=lifespan,
+        )
+        _app_instance.include_router(router)
+        _app_instance.state.server_config = {}
+        _app_instance.state.engine = None
+    return _app_instance
+
+
 def _get_engine() -> InferenceEngine:
-    engine = app.state.engine
+    engine = get_app().state.engine
     if engine is None:
         raise HTTPException(status_code=503, detail="Engine not initialized")
     return engine
 
 
-@app.get("/health")
+@router.get("/health")
 async def health():
+    app = get_app()
     return {
         "status": "ok",
         "model_loaded": app.state.engine is not None,
     }
 
 
-@app.get("/stats")
+@router.get("/stats")
 async def get_stats():
     return _get_engine().get_stats()
 
 
-@app.post("/v1/chat/completions")
+@router.post("/v1/chat/completions")
 async def chat_completion(request: ChatCompletionRequest):
     engine = _get_engine()
     handler = ProtocolHandler(request, engine, OpenAIResponseBuilder())
     return await handler.handle()
 
 
-@app.post("/v1/messages")
+@router.post("/v1/messages")
 async def create_message(request: MessagesRequest):
     engine = _get_engine()
     handler = ProtocolHandler(request, engine, AnthropicResponseBuilder())
@@ -147,14 +179,15 @@ async def create_message(request: MessagesRequest):
 
 
 def run_server(
+    param_path: Path,
     host: str = "0.0.0.0",
     port: int = 8000,
     reload: bool = False,
     device: str = "cuda",
     dtype: torch.dtype = torch.bfloat16,
-    param_path: Optional[Path] = None,
     max_batch_size: int = 16,
 ):
+    app = get_app()
     app.state.server_config = {
         "device": device,
         "dtype": dtype,

astrai/inference/api/tool_parser.py

@@ -0,0 +1,325 @@
+"""Tool call parsers for extracting structured tool calls from model output.
+
+Patterned after vLLM's ToolParser abstraction. Each parser knows how to
+detect and incrementally extract tool calls from raw generated text.
+
+Subclasses may optionally consume ``token_ids`` for token-level parsing
+(e.g. Harmony / VLM-style parsers).
+"""
+
+import re
+import uuid
+from abc import ABC, abstractmethod
+from typing import Dict, List, Optional
+
+from astrai.factory import BaseFactory
+
+
+class BaseToolParser(ABC):
+    """Abstract tool call parser — one instance per request.
+
+    Maintains streaming state internally so that each call to :meth:`feed`
+    can diff against previously emitted content.
+
+    Parameters
+    ----------
+    tools : list of dict, optional
+        Tool definitions from the request.
+    tool_choice : str
+        ``"auto"`` / ``"required"`` / ``"none"`` or a named tool choice
+        dict.
+    """
+
+    def __init__(self, tools: Optional[List[Dict]] = None, tool_choice: str = "auto"):
+        self.tools = tools or []
+        self.tool_choice = tool_choice
+
+    @abstractmethod
+    def feed(
+        self,
+        body: str,
+        current_token_ids: Optional[List[int]] = None,
+        delta_token_ids: Optional[List[int]] = None,
+    ) -> List[Dict]:
+        """Feed the *full* accumulated text each step.
+
+        Returns a list of delta dicts to emit. Each delta is one of:
+
+        - ``{"content": "text"}``      — plain text delta
+        - ``{"tool_calls": [...]}``    — tool-call delta (OpenAI format)
+
+        Returns an empty list when nothing new should be emitted.
+
+        Parameters
+        ----------
+        body : str
+            The complete accumulated generated text so far.
+        current_token_ids : list of int, optional
+            All token IDs decoded into *body* (cumulative).
+        delta_token_ids : list of int, optional
+            Only the token IDs for this chunk.
+        """
+
+    @abstractmethod
+    def parse_complete(self, body: str) -> Optional[Dict]:
+        """Parse the *complete* generated text after generation ends.
+
+        Returns ``None`` when no tool calls were found, otherwise a dict
+        with ``content`` (str or None) and ``tool_calls`` (list of dicts).
+        """
+
+    @property
+    @abstractmethod
+    def has_tool_calls(self) -> bool:
+        """True if the parser detected at least one tool call in the stream."""
+
+
+class ToolParserFactory(BaseFactory["BaseToolParser"]):
+    pass
+
+
+_TOOL_CALL_HEAD_RE = re.compile(r'\{\s*"name"\s*:')
+
+
+def _scan_json(text: str, start: int = 0):
+    """Scan for a complete JSON object starting at *start*.
+
+    Returns ``(end, complete)`` where *end* is one-past the closing
+    brace (or ``len(text)`` if unclosed), and *complete* is a bool.
+    """
+    depth = 0
+    in_string = False
+    escape = False
+    for i in range(start, len(text)):
+        c = text[i]
+        if escape:
+            escape = False
+            continue
+        if c == "\\":
+            escape = True
+            continue
+        if c == '"':
+            in_string = not in_string
+            continue
+        if in_string:
+            continue
+        if c == "{":
+            depth += 1
+        elif c == "}":
+            depth -= 1
+            if depth == 0:
+                return i + 1, True
+    return len(text), False
+
+
+def _parse_tool_call_json(json_str: str, complete: bool):
+    """Extract *name* and *arguments* from a tool-call JSON string.
+
+    Returns ``(name, args, valid)``.
+    """
+    name_match = re.search(r'"name"\s*:\s*"([^"]*)"', json_str)
+    if not name_match:
+        return None, "", False
+    name = name_match.group(1)
+
+    args_match = re.search(r'"arguments"\s*:\s*(.*)', json_str, re.DOTALL)
+    if not args_match:
+        return name, "", True
+
+    raw = args_match.group(1).rstrip()
+    if complete and raw.endswith("}"):
+        raw = raw[:-1].rstrip()
+    if raw.startswith("{"):
+        inner = raw[1:].rstrip()
+        if inner.endswith("}"):
+            inner = inner[:-1].rstrip()
+        raw = inner
+    return name, raw, True
+
+
+def _find_tool_calls(text: str, start_pos: int = 0):
+    """Find all complete ``{...}`` tool-call objects in *text*.
+
+    Returns a list of dicts with keys *start*, *end*, *name*, *args*,
+    *complete*.
+    """
+    results = []
+    pos = start_pos
+
+    while True:
+        brace = text.find("{", pos)
+        if brace == -1:
+            break
+
+        end, complete = _scan_json(text, brace)
+        if not complete:
+            break
+
+        json_str = text[brace:end]
+        if not _TOOL_CALL_HEAD_RE.search(json_str):
+            pos = end
+            continue
+
+        name, args, valid = _parse_tool_call_json(json_str, complete=True)
+        if not valid or name is None:
+            pos = end
+            continue
+
+        results.append(
+            {
+                "start": brace,
+                "end": end,
+                "name": name,
+                "args": args,
+                "complete": True,
+            }
+        )
+        pos = end
+
+    return results
+
+
+def _find_partial_tool_call(text: str, start_pos: int = 0):
+    """Find one incomplete (still-generating) tool-call JSON object."""
+    brace = text.find("{", start_pos)
+    if brace == -1:
+        return None
+
+    json_str = text[brace:]
+    if not _TOOL_CALL_HEAD_RE.search(json_str):
+        return None
+
+    name, args, valid = _parse_tool_call_json(json_str, complete=False)
+    if not valid or name is None:
+        return None
+
+    return {
+        "start": brace,
+        "name": name,
+        "args": args,
+        "complete": False,
+    }
+
+
+@ToolParserFactory.register("simple_json")
+class SimpleJsonToolParser(BaseToolParser):
+    """Parser for models that output tool calls as plain JSON objects.
+
+    Detects ``{"name": "<func>", "arguments": {...}}`` anywhere in the
+    generated text.  Handles single and (non-overlapping) multiple tool
+    calls.  Text preceding the first tool call is emitted as plain
+    ``content`` deltas.
+    """
+
+    def __init__(self, tools=None, tool_choice="auto"):
+        super().__init__(tools, tool_choice)
+        self._emitted_content_len = 0
+        self._tc_state: List[Dict] = []
+        self._has_tool_calls = False
+
+    # -------------------------------------------------------------- feed
+
+    def feed(
+        self,
+        body: str,
+        current_token_ids: Optional[List[int]] = None,
+        delta_token_ids: Optional[List[int]] = None,
+    ) -> List[Dict]:
+        deltas: List[Dict] = []
+
+        completed = _find_tool_calls(body)
+
+        if not completed:
+            partial = _find_partial_tool_call(body)
+            if not partial:
+                return self._emit_plain_content(body, deltas)
+            all_tcs = [partial]
+        else:
+            all_tcs = completed
+            partial = _find_partial_tool_call(body, completed[-1]["end"])
+            if partial:
+                all_tcs = completed + [partial]
+
+        first_start = all_tcs[0]["start"]
+        if first_start > self._emitted_content_len:
+            content = body[self._emitted_content_len : first_start]
+            self._emitted_content_len = first_start
+            if content:
+                deltas.append({"content": content})
+
+        for i, tc in enumerate(all_tcs):
+            if i >= len(self._tc_state):
+                self._tc_state.append(
+                    {
+                        "id": f"call_{uuid.uuid4().hex[:12]}",
+                        "name_emitted": False,
+                        "args_emitted_len": 0,
+                    }
+                )
+                self._has_tool_calls = True
+            st = self._tc_state[i]
+
+            if not st["name_emitted"]:
+                st["name_emitted"] = True
+                deltas.append(
+                    {
+                        "tool_calls": [
+                            {
+                                "index": i,
+                                "id": st["id"],
+                                "type": "function",
+                                "function": {"name": tc["name"], "arguments": ""},
+                            }
+                        ]
+                    }
+                )
+
+            new_args = tc["args"]
+            if len(new_args) > st["args_emitted_len"]:
+                diff = new_args[st["args_emitted_len"] :]
+                st["args_emitted_len"] = len(new_args)
+                deltas.append(
+                    {
+                        "tool_calls": [
+                            {
+                                "index": i,
+                                "function": {"arguments": diff},
+                            }
+                        ]
+                    }
+                )
+
+        return deltas
+
+    def _emit_plain_content(self, body: str, deltas: List[Dict]) -> List[Dict]:
+        new_content = body[self._emitted_content_len :]
+        if new_content:
+            self._emitted_content_len = len(body)
+            deltas.append({"content": new_content})
+        return deltas
+
+    # -------------------------------------------------------- complete
+
+    def parse_complete(self, body: str) -> Optional[Dict]:
+        completed = _find_tool_calls(body)
+        if not completed:
+            return None
+
+        content = body[: completed[0]["start"]].strip() or None
+        tool_calls = []
+        for i, tc in enumerate(completed):
+            tool_calls.append(
+                {
+                    "id": f"call_{uuid.uuid4().hex[:12]}",
+                    "type": "function",
+                    "function": {
+                        "name": tc["name"],
+                        "arguments": tc["args"],
+                    },
+                }
+            )
+        return {"content": content, "tool_calls": tool_calls}
+
+    @property
+    def has_tool_calls(self) -> bool:
+        return self._has_tool_calls

astrai/inference/core/scheduler.py

@@ -70,7 +70,8 @@ class InferenceScheduler:
             dtype=self.dtype,
         )
 
-        self._running = False
+        self._stop_event = threading.Event()
+        self._loop_thread: Optional[threading.Thread] = None
 
     def add_task(self, prompt: str, **kwargs) -> str:
         return self._task_mgr.add_task(prompt, **kwargs)
@@ -85,7 +86,7 @@ class InferenceScheduler:
     def _run_generation_loop(self):
         stop_ids = self._task_mgr.tokenizer.stop_ids
         try:
-            while self._running:
+            while not self._stop_event.is_set():
                 finished = self._task_mgr.remove_finished_tasks(stop_ids)
                 for task in finished:
                     self._page_cache.task_free(task.task_id)
@@ -175,6 +176,7 @@ class InferenceScheduler:
                                     t.stream_callback(STOP)
 
         except Exception as e:
+            self._stop_event.set()
             logger.error(f"Scheduler loop crashed: {e}", exc_info=True)
             for task in self._task_mgr.get_active_tasks():
                 if task.stream_callback:
@@ -184,22 +186,28 @@ class InferenceScheduler:
                 if task.stream_callback:
                     task.stream_callback(STOP)
             self._task_mgr.clear_queues()
-            raise
 
     def start(self):
-        if not self._running:
+        if self._loop_thread is not None and self._loop_thread.is_alive():
-            self._running = True
+            return
-            t = threading.Thread(target=self._run_generation_loop, daemon=True)
+        self._stop_event.clear()
-            t.start()
+        t = threading.Thread(target=self._run_generation_loop, daemon=True)
-            self._loop_thread = t
+        t.start()
+        self._loop_thread = t
 
     def stop(self):
-        self._running = False
+        self._stop_event.set()
         self._task_mgr.wake()
-        if hasattr(self, "_loop_thread"):
+        if self._loop_thread is not None:
             self._loop_thread.join(timeout=2.0)
+        self._loop_thread = None
         for task in self._task_mgr.get_active_tasks():
+            if task.stream_callback:
+                task.stream_callback(STOP)
             self._page_cache.task_free(task.task_id)
+        for task in self._task_mgr.get_waiting_tasks():
+            if task.stream_callback:
+                task.stream_callback(STOP)
         self._task_mgr.clear_queues()
         if torch.cuda.is_available():
             torch.cuda.empty_cache()

astrai/inference/core/task.py

@@ -186,7 +186,10 @@ class TaskManager:
         return bool(self.active_tasks or self.waiting_queue)
 
     def wait_for_tasks(self, timeout: float = 1.0):
-        self._task_event.clear()
+        with self._lock:
+            if self.waiting_queue or self.active_tasks:
+                return
+            self._task_event.clear()
         self._task_event.wait(timeout=timeout)
 
     def get_active_tasks(self) -> List[Task]:

astrai/inference/engine.py

@@ -79,8 +79,8 @@ class GenerationRequest:
             raise ValueError("top_k must be a non-negative integer")
         if not (0.0 <= top_p <= 1.0):
             raise ValueError("top_p must be a float between 0.0 and 1.0")
-        if not (isinstance(temperature, (int, float)) and temperature >= 0):
+        if not (isinstance(temperature, (int, float)) and temperature > 0):
-            raise ValueError("temperature must be a non-negative number")
+            raise ValueError("temperature must be a positive number")
 
         self.messages = messages
         self.top_k = top_k

astrai/inference/sample.py

@@ -29,6 +29,7 @@ class BaseSamplingStrategy(ABC):
         Returns:
             Transformed logits tensor.
         """
+        raise NotImplementedError
 
 
 class TemperatureStrategy(BaseSamplingStrategy):
@@ -41,13 +42,15 @@ class TemperatureStrategy(BaseSamplingStrategy):
     def __init__(self, temperature: Union[float, Tensor] = 1.0):
         self.temperature = temperature
 
-    def apply(self, logits, filter_value=-float("inf")):
+    def apply(self, logits: Tensor, filter_value: float = -float("inf")) -> Tensor:
         t = self.temperature
         if isinstance(t, Tensor):
+            t = t.to(logits.device, non_blocking=True).view(-1, 1)
+            t = torch.clamp(t, min=1e-8)
             if (t != 1.0).any():
-                logits = logits / t.to(logits.device, non_blocking=True).view(-1, 1)
+                logits = logits / t
         elif t != 1.0:
-            logits = logits / t
+            logits = logits / max(t, 1e-8)
         return logits
 
 
@@ -61,7 +64,7 @@ class TopKStrategy(BaseSamplingStrategy):
     def __init__(self, top_k: Union[int, Tensor] = 0):
         self.top_k = top_k
 
-    def apply(self, logits, filter_value=-float("inf")):
+    def apply(self, logits: Tensor, filter_value: float = -float("inf")) -> Tensor:
         tk = self.top_k
         if isinstance(tk, Tensor):
             tk = tk.to(logits.device, non_blocking=True).long().clamp(min=0)
@@ -98,7 +101,9 @@ class TopPStrategy(BaseSamplingStrategy):
     def __init__(self, top_p: Union[float, Tensor] = 1.0):
         self.top_p = top_p
 
-    def _apply(self, logits, top_p, filter_value):
+    def _apply(
+        self, logits: Tensor, top_p: Union[float, Tensor], filter_value: float
+    ) -> Tensor:
         sorted_logits, sorted_indices = torch.sort(logits, descending=True, dim=-1)
         cum_probs = torch.cumsum(torch.softmax(sorted_logits, dim=-1), dim=-1)
         remove = cum_probs > top_p
@@ -109,7 +114,7 @@ class TopPStrategy(BaseSamplingStrategy):
         logits[mask] = filter_value
         return logits
 
-    def apply(self, logits, filter_value=-float("inf")):
+    def apply(self, logits: Tensor, filter_value: float = -float("inf")) -> Tensor:
         tp = self.top_p
         if isinstance(tp, Tensor):
             tp = tp.to(logits.device, non_blocking=True)
@@ -140,7 +145,7 @@ class SamplingPipeline(BaseSamplingStrategy):
     def __init__(self, strategies: List[BaseSamplingStrategy]):
         self.strategies = strategies
 
-    def apply(self, logits, filter_value=-float("inf")):
+    def apply(self, logits: Tensor, filter_value: float = -float("inf")) -> Tensor:
         for strategy in self.strategies:
             logits = strategy.apply(logits, filter_value)
         return logits

astrai/model/components/attention.py

@@ -24,9 +24,7 @@ def repeat_kv(x: Tensor, n_rep: int) -> Tensor:
 
 
 class AttnFactory(BaseFactory[nn.Module]):
-    @classmethod
+    pass
-    def create(cls, attn_type: str, **kwargs) -> nn.Module:
-        return super().create(attn_type, **kwargs)
 
 
 @AttnFactory.register("gqa")
@@ -40,6 +38,7 @@ class GQA(nn.Module):
         norm_eps: float,
         use_gated_attention: bool,
         layer_id: int,
+        n_layers: int = 1,
     ):
         super().__init__()
         assert dim % n_heads == 0
@@ -57,7 +56,7 @@ class GQA(nn.Module):
         self.q_proj = Linear(dim, n_heads * self.head_dim)
         self.k_proj = Linear(dim, n_kv_heads * self.head_dim)
         self.v_proj = Linear(dim, n_kv_heads * self.head_dim)
-        self.o_proj = Linear(dim, dim)
+        self.o_proj = Linear(dim, dim, init_std=0.02 / (2 * n_layers) ** 0.5)
 
         if self.use_qk_norm:
             self.q_norm = RMSNorm(self.head_dim, norm_eps)
@@ -123,6 +122,7 @@ class MLA(nn.Module):
         use_qk_norm: bool,
         use_gated_attention: bool,
         layer_id: int,
+        n_layers: int = 1,
     ):
         super().__init__()
         self.dim = dim
@@ -150,7 +150,9 @@ class MLA(nn.Module):
             n_kv_heads * (2 * self.head_dim),
         )
 
-        self.o_proj = Linear(dim, dim, bias=False)
+        self.o_proj = Linear(
+            dim, dim, bias=False, init_std=0.02 / (2 * n_layers) ** 0.5
+        )
 
         if use_gated_attention:
             self.gate = Linear(dim, dim, bias=False)

astrai/model/components/decoder_block.py

@@ -1,3 +1,4 @@
+from dataclasses import asdict
 from typing import Optional
 
 import torch.nn as nn
@@ -10,35 +11,14 @@ from astrai.model.components.norm import RMSNorm
 
 
 class DecoderBlock(nn.Module):
-    def __init__(
+    def __init__(self, config, layer_id: int):
-        self,
-        dim: int,
-        n_heads: int,
-        dim_ffn: int,
-        n_kv_heads: int,
-        norm_eps: float,
-        use_qk_norm: bool,
-        use_gated_attention: bool,
-        layer_id: int,
-        attn_type: str = "gqa",
-        ffn_type: str = "mlp",
-        **kwargs,
-    ):
         super().__init__()
-        self.attention = AttnFactory.create(
+        cfg = asdict(config)
-            attn_type,
+        cfg["down_init_std"] = 0.02 / (2 * config.n_layers) ** 0.5
-            dim=dim,
+        self.attention = AttnFactory.create(config.attn_type, **cfg, layer_id=layer_id)
-            n_heads=n_heads,
+        self.input_norm = RMSNorm(config.dim, config.norm_eps)
-            n_kv_heads=n_kv_heads,
+        self.post_attention_norm = RMSNorm(config.dim, config.norm_eps)
-            use_qk_norm=use_qk_norm,
+        self.mlp = FFNFactory.create(config.ffn_type, **cfg)
-            norm_eps=norm_eps,
-            use_gated_attention=use_gated_attention,
-            layer_id=layer_id,
-            **kwargs,
-        )
-        self.input_norm = RMSNorm(dim, norm_eps)
-        self.post_attention_norm = RMSNorm(dim, norm_eps)
-        self.mlp = FFNFactory.create(ffn_type, dim, dim_ffn, **kwargs)
 
     def forward(
         self,

astrai/model/components/embedding.py

@@ -1,3 +1,5 @@
+import math
+
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -5,12 +7,20 @@ from torch import Tensor
 
 
 class Embedding(nn.Module):
-    def __init__(self, vocab_size: int, embedding_dim: int):
+    def __init__(self, vocab_size: int, embedding_dim: int, neftune_alpha: float = 0.0):
         super().__init__()
         self.weight = nn.Parameter(torch.empty((vocab_size, embedding_dim)))
+        self.neftune_noise_alpha = neftune_alpha
+
+    def set_neftune_alpha(self, alpha: float):
+        self.neftune_noise_alpha = alpha
 
     def reset_parameters(self):
         nn.init.normal_(self.weight, mean=0.0, std=0.02)
 
     def forward(self, x: Tensor) -> Tensor:
-        return F.embedding(x, self.weight)
+        out = F.embedding(x, self.weight)
+        if self.training and self.neftune_noise_alpha > 0.0:
+            eps = self.neftune_noise_alpha / math.sqrt(out.size(1))
+            out = out + eps * torch.randn_like(out)
+        return out

astrai/model/components/linear.py

@@ -5,13 +5,16 @@ from torch import Tensor
 
 
 class Linear(nn.Module):
-    def __init__(self, in_dim: int, out_dim: int, bias: bool = False):
+    def __init__(
+        self, in_dim: int, out_dim: int, bias: bool = False, init_std: float = 0.02
+    ):
         super().__init__()
         self.weight = nn.Parameter(torch.empty((out_dim, in_dim)))
         self.bias = nn.Parameter(torch.zeros(out_dim)) if bias else None
+        self.init_std = init_std
 
     def reset_parameters(self):
-        nn.init.kaiming_uniform_(self.weight, a=5**0.5)
+        nn.init.normal_(self.weight, mean=0.0, std=self.init_std)
         if self.bias is not None:
             fan_in, _ = nn.init._calculate_fan_in_and_fan_out(self.weight)
             bound = 1 / (fan_in**0.5)

astrai/model/components/mlp.py

@@ -8,18 +8,16 @@ from astrai.model.components.linear import Linear
 
 
 class FFNFactory(BaseFactory[nn.Module]):
-    @classmethod
+    pass
-    def create(cls, ffn_type: str, dim: int, dim_ffn: int, **kwargs) -> nn.Module:
-        return super().create(ffn_type, dim, dim_ffn, **kwargs)
 
 
 @FFNFactory.register("mlp")
 class MLP(nn.Module):
-    def __init__(self, dim: int, dim_ffn: int):
+    def __init__(self, dim: int, dim_ffn: int, down_init_std: float = 0.02):
         super().__init__()
         self.up = Linear(dim, dim_ffn)
         self.gate = Linear(dim, dim_ffn)
-        self.down = Linear(dim_ffn, dim)
+        self.down = Linear(dim_ffn, dim, init_std=down_init_std)
 
     def forward(self, x: Tensor) -> Tensor:
         gated = self.up(x) * F.silu(self.gate(x))
@@ -37,6 +35,7 @@ class DeepSeekMoE(nn.Module):
         n_shared_experts: int = 1,
         n_activated_experts: int = 2,
         topk_method: str = "greedy",
+        n_layers: int = 1,
     ):
         super().__init__()
         self.dim = dim
@@ -46,12 +45,20 @@ class DeepSeekMoE(nn.Module):
         self.topk_method = topk_method
 
         self.router = Linear(dim, n_routed_experts, bias=False)
+        moe_scale = 1 / max(n_shared_experts, 1) + 1 / n_activated_experts
+        down_init_std = 0.02 / (2 * n_layers * moe_scale) ** 0.5
 
         self.shared_experts = nn.ModuleList(
-            [MLP(dim, dim_ffn) for _ in range(n_shared_experts)]
+            [
+                MLP(dim, dim_ffn, down_init_std=down_init_std)
+                for _ in range(n_shared_experts)
+            ]
         )
         self.routed_experts = nn.ModuleList(
-            [MLP(dim, dim_ffn) for _ in range(n_routed_experts)]
+            [
+                MLP(dim, dim_ffn, down_init_std=down_init_std)
+                for _ in range(n_routed_experts)
+            ]
         )
 
     def forward(self, x: Tensor) -> Tensor:

astrai/model/encoder.py

@@ -23,22 +23,12 @@ class EmbeddingEncoder(AutoModel):
         self.rotary_embedding = RotaryEmbedding(
             rope_dim, config.max_len, rope_base, rope_scaling=config.rope_scaling
         )
-        self.embed_tokens = Embedding(config.vocab_size, config.dim)
+        self.embed_tokens = Embedding(
+            config.vocab_size, config.dim, neftune_alpha=config.neftune_alpha
+        )
 
         self.layers = nn.ModuleList(
-            [
+            [DecoderBlock(config, layer_id) for layer_id in range(config.n_layers)]
-                DecoderBlock(
-                    config.dim,
-                    config.n_heads,
-                    config.dim_ffn,
-                    config.n_kv_heads,
-                    config.norm_eps,
-                    config.use_qk_norm,
-                    config.use_gated_attention,
-                    layer_id,
-                )
-                for layer_id in range(config.n_layers)
-            ]
         )
 
         self.norm = RMSNorm(config.dim, config.norm_eps)

astrai/model/transformer.py

@@ -26,24 +26,21 @@ def process_attention_mask(
         return input_mask
 
     device = input_tensor.device
-    dtype = input_tensor.dtype
+    B = input_tensor.size(0)
-    B, S = input_tensor.size()[:2]
     T = position_ids.max().item() + 1
 
     if input_mask is None:
         if position_ids.min().item() == 0 and is_causal:
             return None
-        pad = torch.ones(B, T, dtype=torch.bool, device=device)
+        attend = torch.ones(B, 1, T, dtype=torch.bool, device=device)
     else:
-        pad = input_mask[:, :T].to(device=device, dtype=torch.bool)
+        attend = input_mask[:, :T].to(device=device, dtype=torch.bool).unsqueeze(1)
 
-    attend = pad.view(B, 1, T).expand(B, S, T).clone()
     if is_causal:
-        attend &= position_ids.unsqueeze(-1) >= torch.arange(T, device=device)
+        causal = position_ids.unsqueeze(-1) >= torch.arange(T, device=device)
+        attend = attend & causal
 
-    return torch.full(
+    return attend.unsqueeze(1)
-        (B, 1, S, T), -torch.finfo(dtype).max / 2, dtype=dtype, device=device
-    ).masked_fill_(attend.unsqueeze(1), 0.0)
 
 
 @AutoModel.register("autoregressive_lm")
@@ -62,31 +59,12 @@ class AutoRegressiveLM(AutoModel):
         self.rotary_embedding = RotaryEmbedding(
             rope_dim, config.max_len, rope_base, rope_scaling=config.rope_scaling
         )
-        self.embed_tokens = Embedding(config.vocab_size, config.dim)
+        self.embed_tokens = Embedding(
+            config.vocab_size, config.dim, neftune_alpha=config.neftune_alpha
+        )
 
         self.layers = nn.ModuleList(
-            [
+            [DecoderBlock(config, layer_id) for layer_id in range(config.n_layers)]
-                DecoderBlock(
-                    config.dim,
-                    config.n_heads,
-                    config.dim_ffn,
-                    config.n_kv_heads,
-                    config.norm_eps,
-                    config.use_qk_norm,
-                    config.use_gated_attention,
-                    layer_id,
-                    attn_type=config.attn_type,
-                    ffn_type=config.ffn_type,
-                    n_routed_experts=config.n_routed_experts,
-                    n_shared_experts=config.n_shared_experts,
-                    n_activated_experts=config.n_activated_experts,
-                    topk_method=config.topk_method,
-                    kv_lora_rank=config.kv_lora_rank,
-                    qk_nope_head_dim=config.qk_nope_head_dim,
-                    qk_rope_head_dim=config.qk_rope_head_dim,
-                )
-                for layer_id in range(config.n_layers)
-            ]
         )
 
         self.norm = RMSNorm(config.dim, config.norm_eps)

astrai/parallel/executor.py

@@ -2,11 +2,13 @@
 
 import contextlib
 import logging
+import os
 from contextlib import contextmanager
 from typing import Optional, Tuple
 
 import torch
 import torch.nn as nn
+from torch.distributed.fsdp import FullStateDictConfig, StateDictType
 from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
 from torch.nn.parallel import DistributedDataParallel as DDP
 from torch.optim import Optimizer
@@ -115,8 +117,8 @@ class BaseExecutor:
     def backward(self, loss: torch.Tensor):
         loss.backward()
 
-    def unwrap_model(self, model: nn.Module) -> nn.Module:
+    def unwrap_model(self, model: nn.Module):
-        return model
+        return model.state_dict()
 
     @property
     def use_distributed(self) -> bool:
@@ -180,7 +182,7 @@ class DDPExecutor(BaseExecutor):
         if not self.use_distributed:
             logger.warning("DDP backend selected but world_size=1, model not wrapped")
             return model
-        local_rank = get_rank()
+        local_rank = int(os.environ.get("LOCAL_RANK", get_rank()))
         model = DDP(
             model,
             device_ids=[local_rank],
@@ -195,10 +197,10 @@ class DDPExecutor(BaseExecutor):
             return model.no_sync()
         return contextlib.nullcontext()
 
-    def unwrap_model(self, model: nn.Module) -> nn.Module:
+    def unwrap_model(self, model: nn.Module):
         if isinstance(model, DDP):
-            return model.module
+            return model.module.state_dict()
-        return model
+        return model.state_dict()
 
 
 @ExecutorFactory.register("fsdp")
@@ -217,7 +219,6 @@ class FSDPExecutor(BaseExecutor):
         sync_module_states: bool = False,
         forward_prefetch: bool = False,
         limit_all_gathers: bool = True,
-        use_orig_params: bool = False,
         ignored_states=None,
         device_mesh=None,
     ):
@@ -236,7 +237,7 @@ class FSDPExecutor(BaseExecutor):
                 sync_module_states=sync_module_states,
                 forward_prefetch=forward_prefetch,
                 limit_all_gathers=limit_all_gathers,
-                use_orig_params=use_orig_params,
+                use_orig_params=True,
                 ignored_states=ignored_states,
                 device_mesh=device_mesh,
             ).items()
@@ -259,9 +260,13 @@ class FSDPExecutor(BaseExecutor):
             return model.no_sync()
         return contextlib.nullcontext()
 
-    def unwrap_model(self, model: nn.Module) -> nn.Module:
+    def unwrap_model(self, model: nn.Module):
-        if self._original_model is not None:
+        if isinstance(model, FSDP) and self.use_distributed:
-            return self._original_model
+            with FSDP.state_dict_type(
-        if isinstance(model, FSDP):
+                model,
-            return model._fsdp_wrapped_module
+                StateDictType.FULL_STATE_DICT,
-        return model
+                FullStateDictConfig(offload_to_cpu=True, rank0_only=False),
+            ):
+                return model.state_dict()
+
+        return model.state_dict()

astrai/parallel/setup.py

@@ -1,4 +1,5 @@
 import os
+from abc import ABC, abstractmethod
 from contextlib import contextmanager
 from functools import wraps
 from typing import Callable
@@ -30,6 +31,7 @@ def get_rank() -> int:
 def setup_parallel(
     rank: int,
     world_size: int,
+    local_rank: int,
     backend: str = "nccl",
     master_addr: str = "localhost",
     master_port: str = "29500",
@@ -41,14 +43,18 @@ def setup_parallel(
         return
 
     if world_size <= 1:
+        device_id = torch.device(device_type, local_rank)
+        os.environ["LOCAL_RANK"] = str(local_rank)
+        os.environ["WORLD_SIZE"] = "1"
+        os.environ["LOCAL_DEVICE"] = str(device_id)
         yield None
         return
 
-    device_id = torch.device(device_type, rank)
+    device_id = torch.device(device_type, local_rank)
 
     os.environ["MASTER_ADDR"] = master_addr
     os.environ["MASTER_PORT"] = master_port
-    os.environ["LOCAL_RANK"] = str(rank)
+    os.environ["LOCAL_RANK"] = str(local_rank)
     os.environ["WORLD_SIZE"] = str(world_size)
     os.environ["LOCAL_DEVICE"] = str(device_id)
 
@@ -90,7 +96,7 @@ def only_on_rank(rank, sync=False):
     return decorator
 
 
-def wrapper_spawn_func(
+def _run_single_rank(
     rank: int,
     world_size: int,
     backend: str,
@@ -100,20 +106,108 @@ def wrapper_spawn_func(
     func: Callable,
     kwargs: dict,
 ):
-    try:
+    with setup_parallel(
+        rank=rank,
+        world_size=world_size,
+        local_rank=rank,
+        backend=backend,
+        master_addr=master_addr,
+        master_port=master_port,
+        device_type=device_type,
+    ):
+        func(**kwargs)
+
+
+class LaunchStrategy(ABC):
+    """Strategy for launching a function in a distributed context."""
+
+    def __init__(
+        self,
+        world_size: int,
+        backend: str,
+        master_addr: str,
+        master_port: str,
+        device_type: str,
+        start_method: str,
+    ):
+        self.world_size = world_size
+        self.backend = backend
+        self.master_addr = master_addr
+        self.master_port = master_port
+        self.device_type = device_type
+        self.start_method = start_method
+
+    @abstractmethod
+    def launch(self, func: Callable, **kwargs):
+        raise NotImplementedError
+
+
+class TorchrunStrategy(LaunchStrategy):
+    """External orchestrator (torchrun, SLURM, K8s) — env vars pre-set."""
+
+    def launch(self, func: Callable, **kwargs):
+        rank = int(os.environ["RANK"])
+        world_size = int(os.environ["WORLD_SIZE"])
+        local_rank = int(os.environ.get("LOCAL_RANK", rank))
         with setup_parallel(
             rank=rank,
             world_size=world_size,
-            backend=backend,
+            local_rank=local_rank,
-            master_addr=master_addr,
+            backend=self.backend,
-            master_port=master_port,
+            master_addr=os.environ.get("MASTER_ADDR", self.master_addr),
-            device_type=device_type,
+            master_port=os.environ.get("MASTER_PORT", self.master_port),
+            device_type=self.device_type,
         ):
             func(**kwargs)
 
-    except Exception as e:
+
-        print(f"Error in rank {rank}: {e}")
+class LocalStrategy(LaunchStrategy):
-        raise
+    """Local launcher — single-process or mp.start_processes."""
+
+    def launch(self, func: Callable, **kwargs):
+        args = (
+            self.world_size,
+            self.backend,
+            self.master_addr,
+            self.master_port,
+            self.device_type,
+            func,
+            kwargs,
+        )
+
+        if self.world_size == 1:
+            _run_single_rank(0, *args)
+            return
+
+        ctx = mp.start_processes(
+            _run_single_rank,
+            args=args,
+            nprocs=self.world_size,
+            start_method=self.start_method,
+            join=False,
+        )
+        try:
+            while not ctx.join():
+                pass
+        except BaseException:
+            for p in ctx.processes:
+                p.terminate()
+            ctx.join()
+            raise
+
+
+def _detect_launcher() -> str:
+    """Detect the distributed launcher from environment.
+
+    Returns one of: "torchelastic", "torchrun", "external", "local".
+    """
+    if dist.is_torchelastic_launched():
+        return "torchelastic"
+    if "LOCAL_WORLD_SIZE" in os.environ:
+        return "torchrun"
+    if "RANK" in os.environ and "WORLD_SIZE" in os.environ:
+        return "external"
+    return "local"
 
 
 def spawn_parallel_fn(
@@ -126,41 +220,13 @@ def spawn_parallel_fn(
     start_method: str = "spawn",
     **kwargs,
 ):
-    # clear environment variables
+    launcher = _detect_launcher()
-    for key in [
+    if launcher in ("torchelastic", "torchrun", "external"):
-        "MASTER_ADDR",
+        strategy = TorchrunStrategy(
-        "MASTER_PORT",
+            world_size, backend, master_addr, master_port, device_type, start_method
-        "RANK",
+        )
-        "WORLD_SIZE",
+    else:
-        "LOCAL_RANK",
+        strategy = LocalStrategy(
-        "LOCAL_DEVICE",
+            world_size, backend, master_addr, master_port, device_type, start_method
-    ]:
+        )
-        if key in os.environ:
+    strategy.launch(func, **kwargs)
-            del os.environ[key]
-
-    if world_size == 1:
-        device_id = torch.device(device_type, 0)
-        os.environ["LOCAL_RANK"] = "0"
-        os.environ["WORLD_SIZE"] = "1"
-        os.environ["LOCAL_DEVICE"] = str(device_id)
-
-        func(**kwargs)
-        return
-
-    wrapper_spawn_func_args = (
-        world_size,
-        backend,
-        master_addr,
-        master_port,
-        device_type,
-        func,
-        kwargs,
-    )
-
-    mp.start_processes(
-        wrapper_spawn_func,
-        args=wrapper_spawn_func_args,
-        nprocs=world_size,
-        start_method=start_method,
-        join=True,
-    )

astrai/preprocessing/__init__.py

@@ -0,0 +1,32 @@
+from astrai.preprocessing.builder import (
+    BaseMaskBuilder,
+    MaskBuilderFactory,
+    SectionedMaskBuilder,
+)
+from astrai.preprocessing.packing import (
+    PackingStrategy,
+    PackingStrategyFactory,
+)
+from astrai.preprocessing.pipeline import Pipeline, filter_by_length
+from astrai.preprocessing.position_id import (
+    PositionIdStrategy,
+    PositionIdStrategyFactory,
+)
+from astrai.preprocessing.writer import (
+    StoreWriter,
+    StoreWriterFactory,
+)
+
+__all__ = [
+    "BaseMaskBuilder",
+    "MaskBuilderFactory",
+    "PackingStrategy",
+    "PackingStrategyFactory",
+    "Pipeline",
+    "PositionIdStrategy",
+    "PositionIdStrategyFactory",
+    "SectionedMaskBuilder",
+    "StoreWriter",
+    "StoreWriterFactory",
+    "filter_by_length",
+]

astrai/preprocessing/builder.py

@@ -0,0 +1,315 @@
+"""Mask building for preprocessing pipeline.
+
+:class:`SectionRenderer` converts section specs into token ids and loss
+masks (template / text / value extraction).  :class:`SectionedMaskBuilder`
+orchestrates single-output / multi-output (DPO / GRPO) assembly.
+"""
+
+from abc import ABC, abstractmethod
+from typing import Optional
+
+from astrai.factory import BaseFactory
+
+
+def _extract_domain(item: dict, domain_key: Optional[str]) -> str:
+    if not domain_key:
+        return "__default__"
+    val = item.get(domain_key, "__default__")
+    return val if isinstance(val, str) else "__default__"
+
+
+def _resolve_action(action: str, role: str, config) -> str:
+    if action == "$role":
+        return config.mask.get(role, config.mask_default)
+    return action
+
+
+class SectionRenderer:
+    """Render section specs into ``(ids, loss_mask)`` tuples."""
+
+    def process_sections(
+        self,
+        item: dict,
+        sections: list,
+        config,
+        tokenizer,
+        *,
+        is_top_level: bool = False,
+    ):
+        all_ids: list[int] = []
+        loss_mask: list[int] = []
+
+        has_template = any(s.get("template") for s in sections)
+        is_text_config = not has_template and all(
+            s["action"] == "train" for s in sections
+        )
+
+        if is_top_level and has_template and tokenizer.bos_token_id is not None:
+            all_ids.append(tokenizer.bos_token_id)
+            loss_mask.append(0)
+
+        first_section = True
+        for sec in sections:
+            field = sec["field"]
+            action = sec["action"]
+            use_template = sec.get("template", False)
+            add_special = sec.get(
+                "add_special_tokens", not use_template and first_section
+            )
+
+            if use_template:
+                success = self._append_template(
+                    item, field, action, tokenizer, config, all_ids, loss_mask
+                )
+                if not success:
+                    continue
+            else:
+                success = self._append_text(
+                    item,
+                    field,
+                    action,
+                    tokenizer,
+                    add_special,
+                    is_text_config,
+                    config,
+                    all_ids,
+                    loss_mask,
+                )
+                if not success:
+                    continue
+
+            first_section = False
+
+        max_len = config.preprocessing.max_seq_len
+        all_ids = all_ids[:max_len]
+        loss_mask = loss_mask[: len(all_ids)]
+
+        if not all_ids:
+            return None, None
+
+        if is_top_level and has_template and len(all_ids) <= 1:
+            return None, None
+
+        return all_ids, loss_mask
+
+    def process_list_field(self, item: dict, sections: list, config, tokenizer):
+        all_ids: list[int] = []
+        loss_mask: list[int] = []
+
+        for sec in sections:
+            field = sec["field"]
+            action = sec["action"]
+            use_template = sec.get("template", False)
+
+            values = item.get(field)
+            if not isinstance(values, list):
+                continue
+
+            for val in values:
+                if use_template:
+                    if isinstance(val, list):
+                        wrapper = {field: val}
+                        self._append_template(
+                            wrapper,
+                            field,
+                            action,
+                            tokenizer,
+                            config,
+                            all_ids,
+                            loss_mask,
+                        )
+                else:
+                    wrapper = {field: str(val)}
+                    self._append_text(
+                        wrapper,
+                        field,
+                        action,
+                        tokenizer,
+                        False,
+                        False,
+                        config,
+                        all_ids,
+                        loss_mask,
+                    )
+
+        max_len = config.preprocessing.max_seq_len
+        all_ids = all_ids[:max_len]
+        loss_mask = loss_mask[: len(all_ids)]
+
+        if not all_ids:
+            return None, None
+        return all_ids, loss_mask
+
+    @staticmethod
+    def is_value_section(sections: list) -> bool:
+        return len(sections) == 1 and sections[0].get("action") == "value"
+
+    @staticmethod
+    def extract_raw_value(item: dict, sections: list):
+        sec = sections[0]
+        field = sec["field"]
+        raw = item.get(field)
+        if raw is None:
+            return None
+        if isinstance(raw, list):
+            return [float(v) for v in raw]
+        return [float(raw)]
+
+    def _append_template(
+        self, item, field, action, tokenizer, config, all_ids, loss_mask
+    ):
+        messages = item.get(field)
+        if not isinstance(messages, list) or not messages:
+            return False
+        for msg in messages:
+            role = msg.get("role", "")
+            act = _resolve_action(action, role, config)
+            rendered = tokenizer.apply_chat_template(
+                [msg], tokenize=False, add_generation_prompt=False
+            )
+            ids = tokenizer.encode(rendered, add_special_tokens=False)
+            all_ids.extend(ids)
+            val = 1 if act == "train" else 0
+            loss_mask.extend([val] * len(ids))
+        return True
+
+    def _append_text(
+        self,
+        item,
+        field,
+        action,
+        tokenizer,
+        add_special,
+        is_text_config,
+        config,
+        all_ids,
+        loss_mask,
+    ):
+        text = str(item.get(field, ""))
+        if not text.strip():
+            return False
+        if is_text_config:
+            pp = config.preprocessing
+            if pp.min_chars > 0 and len(text) < pp.min_chars:
+                return False
+            if len(text) > pp.max_chars:
+                return False
+        ids = tokenizer.encode(text, add_special_tokens=add_special)
+        all_ids.extend(ids)
+        val = 1 if action == "train" else 0
+        loss_mask.extend([val] * len(ids))
+        return True
+
+
+class BaseMaskBuilder(ABC):
+    """Convert a JSONL item into token ids and optional loss_mask."""
+
+    @abstractmethod
+    def build(self, item: dict, config, tokenizer) -> Optional[dict]: ...
+
+
+class MaskBuilderFactory(BaseFactory["BaseMaskBuilder"]):
+    pass
+
+
+@MaskBuilderFactory.register("sectioned")
+class SectionedMaskBuilder(BaseMaskBuilder):
+    """Config-driven builder supporting single and multi-output modes.
+
+    Single-output::
+
+        {"input": {"sections": [
+            {"field": "messages", "action": "$role", "template": true}
+        ]}}
+        → {"sequence": [...], "loss_mask": [...], "domain": "..."}
+
+    Multi-output (DPO / GRPO)::
+
+        {"input": {"sources": {
+            "chosen": {"sections": [{"field": "chosen", "action": "$role", "template": true}]},
+            "rejected": {"sections": [{"field": "rejected", "action": "$role", "template": true}]},
+        }}}
+        → {"chosen": [...], "chosen_mask": [...], "rejected": [...], "rejected_mask": [...], "domain": "..."}
+
+    Output spec fields::
+
+        sections      – list of section specs (same format as single-output)
+        list_field    – True when JSONL field holds a list (GRPO responses)
+        mask_key      – explicit loss-mask output key (default: ``"{output_key}_mask"``)
+    """
+
+    def __init__(self):
+        self.renderer = SectionRenderer()
+
+    def build(self, item: dict, config, tokenizer) -> Optional[dict]:
+        sources_spec = getattr(config.input, "sources", None)
+        if sources_spec:
+            return self._build_multi(item, sources_spec, config, tokenizer)
+        return self._build_single(item, config, tokenizer)
+
+    def _build_single(self, item: dict, config, tokenizer) -> Optional[dict]:
+        sections = config.input.sections
+        if not sections:
+            return None
+
+        ids, mask = self.renderer.process_sections(
+            item, sections, config, tokenizer, is_top_level=True
+        )
+        if ids is None:
+            return None
+
+        result: dict = {
+            "sequence": ids,
+            "domain": _extract_domain(item, config.output.domain_key),
+        }
+        if not all(m == 1 for m in mask):
+            result["loss_mask"] = mask
+        return result
+
+    def _build_multi(
+        self, item: dict, sources_spec: dict, config, tokenizer
+    ) -> Optional[dict]:
+        result: dict = {}
+        any_output = False
+
+        for output_key, spec in sources_spec.items():
+            sections = spec.get("sections", [])
+            if not sections:
+                continue
+
+            if self.renderer.is_value_section(sections):
+                ids = self.renderer.extract_raw_value(item, sections)
+                if ids is None:
+                    continue
+                result[output_key] = ids
+                any_output = True
+                continue
+
+            list_field = spec.get("list_field", False)
+            mask_key = spec.get("mask_key", f"{output_key}_mask")
+
+            if list_field:
+                ids, mask = self.renderer.process_list_field(
+                    item, sections, config, tokenizer
+                )
+            else:
+                ids, mask = self.renderer.process_sections(
+                    item, sections, config, tokenizer, is_top_level=True
+                )
+
+            if ids is None:
+                continue
+
+            result[output_key] = ids
+            if not all(m == 1 for m in mask):
+                result[mask_key] = mask
+            elif "mask_key" in spec:
+                result[mask_key] = mask
+
+            any_output = True
+
+        if not any_output:
+            return None
+
+        result["domain"] = _extract_domain(item, config.output.domain_key)
+        return result

astrai/preprocessing/packing.py

@@ -0,0 +1,121 @@
+"""Sequence packing strategies for shard-level reordering and truncation.
+
+Each strategy receives the accumulated ``{key: [list of token lists]}``
+dict for a shard and returns a reordered / truncated version.  The
+pipeline later flattens the result into contiguous tensors.
+"""
+
+from abc import ABC, abstractmethod
+from typing import Dict, List
+
+from astrai.factory import BaseFactory
+
+
+def _truncate(seq: List[int], max_len: int, mode: str) -> List[int]:
+    if len(seq) <= max_len:
+        return seq
+    if mode == "keep_end":
+        return seq[-max_len:]
+    return seq[:max_len]
+
+
+class PackingStrategy(ABC):
+    """Reorder and truncate sequences within a shard."""
+
+    @abstractmethod
+    def apply(
+        self,
+        keys: Dict[str, List[List[int]]],
+        max_packed_len: int,
+        truncation_mode: str,
+    ) -> Dict[str, List[List[int]]]:
+        raise NotImplementedError
+
+
+class PackingStrategyFactory(BaseFactory["PackingStrategy"]):
+    pass
+
+
+@PackingStrategyFactory.register("simple")
+class SimplePacking(PackingStrategy):
+    def apply(
+        self,
+        keys: Dict[str, List[List[int]]],
+        max_packed_len: int,
+        truncation_mode: str,
+    ) -> Dict[str, List[List[int]]]:
+        return {
+            k: [_truncate(v, max_packed_len, truncation_mode) for v in vals]
+            for k, vals in keys.items()
+        }
+
+
+@PackingStrategyFactory.register("bfd")
+class BFDPacking(PackingStrategy):
+    """Best-Fit Decreasing bin packing.
+
+    Assigns sequences to bins using a best-fit heuristic (sorted by
+    decreasing length) and concatenates sequences within each bin into
+    a single packed sequence.  Packed sequences are truncated to
+    *max_packed_len* so that each packed bin fits within one context
+    window during training.
+    """
+
+    def apply(
+        self,
+        keys: Dict[str, List[List[int]]],
+        max_packed_len: int,
+        truncation_mode: str,
+    ) -> Dict[str, List[List[int]]]:
+        sequences = keys.get("sequence", [])
+        if not sequences:
+            return keys
+        bins = self._plan(sequences, max_packed_len, truncation_mode)
+
+        packed: Dict[str, List[List[int]]] = {}
+        for k, vals in keys.items():
+            packed[k] = [
+                _truncate(
+                    self._concat_bin(vals, bin_indices),
+                    max_packed_len,
+                    truncation_mode,
+                )
+                for bin_indices in bins
+            ]
+        return packed
+
+    @staticmethod
+    def _concat_bin(vals: List[List[int]], indices: List[int]) -> List[int]:
+        result: List[int] = []
+        for i in indices:
+            result.extend(vals[i])
+        return result
+
+    @staticmethod
+    def _plan(
+        sequences: List[List[int]], max_packed_len: int, truncation_mode: str
+    ) -> List[List[int]]:
+        n = len(sequences)
+        order = sorted(range(n), key=lambda i: len(sequences[i]), reverse=True)
+        bins: List[List[int]] = []
+        bin_lengths: List[int] = []
+
+        for orig_idx in order:
+            seq_len = len(
+                _truncate(sequences[orig_idx], max_packed_len, truncation_mode)
+            )
+            best_bin = None
+            best_remain = max_packed_len + 1
+            for i, bl in enumerate(bin_lengths):
+                remain = max_packed_len - bl
+                if seq_len <= remain < best_remain:
+                    best_remain = remain
+                    best_bin = i
+            if best_bin is not None:
+                bins[best_bin].append(orig_idx)
+                bin_lengths[best_bin] += seq_len
+            else:
+                bins.append([orig_idx])
+                bin_lengths.append(seq_len)
+
+        return bins

astrai/preprocessing/pipeline.py

@@ -0,0 +1,185 @@
+"""Config-driven JSONL preprocessing pipeline.
+
+Composes a :class:`BaseMaskBuilder` (selected by ``input.type``) with
+sharding and flush to ``.h5`` / ``.bin`` storage.  Packing, position-id
+generation and storage writing are each delegated to pluggable strategies,
+dispatched by configuration keys.
+"""
+
+import json
+import logging
+import os
+from collections import defaultdict
+from itertools import chain
+from typing import Dict, List, Optional
+
+import torch
+import tqdm
+
+from astrai.config.preprocess_config import PipelineConfig
+from astrai.preprocessing.builder import MaskBuilderFactory
+from astrai.preprocessing.packing import PackingStrategyFactory
+from astrai.preprocessing.position_id import PositionIdStrategyFactory
+from astrai.preprocessing.writer import StoreWriterFactory
+from astrai.tokenize import AutoTokenizer
+
+logger = logging.getLogger(__name__)
+
+_STR_TO_DTYPE: dict[str, torch.dtype] = {
+    "bool": torch.bool,
+    "uint8": torch.uint8,
+    "int8": torch.int8,
+    "int16": torch.int16,
+    "int32": torch.int32,
+    "int64": torch.int64,
+    "float16": torch.float16,
+    "float32": torch.float32,
+    "float64": torch.float64,
+}
+
+
+def filter_by_length(text: str, min_len: int = 50, max_len: int = 2_000_000) -> bool:
+    return min_len <= len(text) <= max_len
+
+
+class Pipeline:
+    """Tokenization pipeline driven by a declarative :class:`PipelineConfig`.
+
+    Usage::
+
+        config = PipelineConfig.from_file("sft_pipeline.json")
+        Pipeline(config, ["data.jsonl"], output_dir="out", tokenizer_path="params").run()
+    """
+
+    def __init__(
+        self,
+        config: PipelineConfig,
+        input_paths: list[str],
+        output_dir: str,
+        tokenizer_path: str,
+    ):
+        os.makedirs(output_dir, exist_ok=True)
+        self.config = config
+        self.paths = input_paths
+        self.output_dir = output_dir
+        self.tokenizer_path = tokenizer_path
+
+        self.mask_builder = MaskBuilderFactory.create("sectioned")
+        self._packer = PackingStrategyFactory.create(
+            config.preprocessing.packing_strategy
+        )
+        self._position_id = PositionIdStrategyFactory.create(
+            config.output.position_ids_mode
+        )
+        self._writer = StoreWriterFactory.create(config.output.storage_format)
+
+    def transform(self, item: dict) -> Optional[dict]:
+        return self.mask_builder.build(item, self.config, self._tokenizer)
+
+    def run(self):
+        self._tokenizer = AutoTokenizer.from_pretrained(self.tokenizer_path)
+        domains: dict = defaultdict(lambda: defaultdict(list))
+        total_tokens = 0
+        shard_idx: dict[str, int] = defaultdict(int)
+        count = 0
+
+        pp = self.config.preprocessing
+
+        for item in tqdm.tqdm(
+            self._iter_items(), desc="Tokenizing", unit="docs", mininterval=0.5
+        ):
+            if pp.max_items and count >= pp.max_items:
+                break
+
+            try:
+                result = self.transform(item)
+            except Exception:
+                logger.warning(
+                    "Failed to process item #%d, skipping", count + 1, exc_info=True
+                )
+                continue
+            if result is None:
+                continue
+
+            domain = result.pop("domain", "__default__")
+
+            is_multi = bool(getattr(self.config.input, "sources", None))
+            if is_multi:
+                ids = self._primary_ids(result)
+            else:
+                ids = result.pop("sequence")
+                result["sequence"] = ids
+
+            if not ids:
+                continue
+
+            bucket = domains[domain]
+            self._align_bucket(bucket, result, ids)
+            for key, val in result.items():
+                bucket[key].append(val)
+
+            count += 1
+            total_tokens += len(ids)
+
+            if total_tokens >= self.config.output.max_tokens_per_shard:
+                self._flush(domains, shard_idx)
+                domains.clear()
+                total_tokens = 0
+
+        if total_tokens > 0:
+            self._flush(domains, shard_idx)
+
+    @staticmethod
+    def _primary_ids(result: dict) -> list:
+        """Return the first list-valued entry in *result* as the primary id
+        sequence for token counting."""
+        for val in result.values():
+            if isinstance(val, list) and val and isinstance(val[0], int):
+                return val
+        return []
+
+    @staticmethod
+    def _align_bucket(bucket: dict, result: dict, ids: list):
+        """Pad previously-accumulated keys that are missing from *result*."""
+        for key in list(bucket.keys()):
+            if key in result:
+                continue
+            bucket[key].append([1] * len(ids))
+
+    def _iter_items(self):
+        for path in self.paths:
+            with open(path, "r", encoding="utf-8") as f:
+                for line in f:
+                    line = line.strip()
+                    if not line:
+                        continue
+                    yield json.loads(line)
+
+    def _flush(self, domains, shard_idx):
+        for domain, keys in domains.items():
+            idx = shard_idx[domain]
+
+            pp = self.config.preprocessing
+            keys = self._packer.apply(dict(keys), pp.max_packed_len, pp.truncation_mode)
+
+            tensors: Dict[str, List[torch.Tensor]] = {}
+            for key, ids_list in keys.items():
+                dt = _STR_TO_DTYPE.get(
+                    self.config.output.dtype.get(key, "int32"), torch.int32
+                )
+                tensors[key] = [
+                    torch.tensor(list(chain.from_iterable(ids_list)), dtype=dt)
+                ]
+
+            pos_ids = self._position_id.generate(keys.get("sequence", []))
+            if pos_ids:
+                tensors["position_ids"] = [torch.tensor(pos_ids, dtype=torch.int32)]
+
+            self._writer.save(self.output_dir, domain, idx, tensors)
+            shard_idx[domain] = idx + 1
+
+            first_key = "sequence" if "sequence" in tensors else next(iter(tensors))
+            tqdm.tqdm.write(
+                f"  saved {domain}/shard_{idx:04d}  "
+                f"({tensors[first_key][0].numel():,} tokens)"
+            )

astrai/preprocessing/position_id.py

@@ -0,0 +1,46 @@
+"""Position-id generation strategies for packed sequences.
+
+Each strategy takes the list of per-document token sequences after packing
+and returns a flat list of position ids (same total length as all
+sequences combined).  The pipeline wraps the result into a tensor and
+attaches it as ``position_ids``.
+"""
+
+from abc import ABC, abstractmethod
+from typing import List
+
+from astrai.factory import BaseFactory
+
+
+class PositionIdStrategy(ABC):
+    """Generate ``position_ids`` for packed sequences."""
+
+    @abstractmethod
+    def generate(self, sequences: List[List[int]]) -> List[int]:
+        raise NotImplementedError
+
+
+class PositionIdStrategyFactory(BaseFactory["PositionIdStrategy"]):
+    pass
+
+
+@PositionIdStrategyFactory.register("none")
+class NoPositionId(PositionIdStrategy):
+    def generate(self, sequences: List[List[int]]) -> List[int]:
+        return []
+
+
+@PositionIdStrategyFactory.register("doc_reset")
+class DocResetPositionId(PositionIdStrategy):
+    def generate(self, sequences: List[List[int]]) -> List[int]:
+        pos_ids = []
+        for seq in sequences:
+            pos_ids.extend(range(len(seq)))
+        return pos_ids
+
+
+@PositionIdStrategyFactory.register("continuous")
+class ContinuousPositionId(PositionIdStrategy):
+    def generate(self, sequences: List[List[int]]) -> List[int]:
+        total = sum(len(seq) for seq in sequences)
+        return list(range(total))

astrai/preprocessing/writer.py

@@ -0,0 +1,75 @@
+"""Storage writer strategies for pipeline output.
+
+The :class:`StoreWriter` abstraction decouples the pipeline from the
+concrete storage format (bin / h5).  The pipeline builds a ``{key:
+List[Tensor]}`` dict and delegates the write to the writer selected
+by ``output.storage_format``.
+"""
+
+import logging
+import os
+import shutil
+from abc import ABC, abstractmethod
+from typing import Dict, List
+
+import torch
+
+from astrai.dataset.storage import save_bin, save_h5
+from astrai.factory import BaseFactory
+
+logger = logging.getLogger(__name__)
+
+
+class StoreWriter(ABC):
+    """Write pre-tokenized tensors to disk in a format-specific way."""
+
+    @abstractmethod
+    def save(
+        self,
+        output_dir: str,
+        domain: str,
+        shard_idx: int,
+        tensors: Dict[str, List[torch.Tensor]],
+    ) -> None: ...
+
+
+class StoreWriterFactory(BaseFactory["StoreWriter"]):
+    pass
+
+
+@StoreWriterFactory.register("bin")
+class BinWriter(StoreWriter):
+    def save(self, output_dir, domain, shard_idx, tensors):
+        shard_path = os.path.join(output_dir, domain, f"shard_{shard_idx:04d}")
+        try:
+            save_bin(shard_path, tensors)
+        except Exception:
+            if os.path.exists(shard_path):
+                shutil.rmtree(shard_path, ignore_errors=True)
+            logger.error(
+                "Failed to write shard %s/%s_%04d, cleaned up partial output",
+                domain,
+                "shard",
+                shard_idx,
+                exc_info=True,
+            )
+            raise
+
+
+@StoreWriterFactory.register("h5")
+class H5Writer(StoreWriter):
+    def save(self, output_dir, domain, shard_idx, tensors):
+        chunk_dir = os.path.join(output_dir, domain)
+        file_path = os.path.join(chunk_dir, f"data_{shard_idx:04d}.h5")
+        try:
+            save_h5(chunk_dir, f"data_{shard_idx:04d}", tensors)
+        except Exception:
+            if os.path.exists(file_path):
+                os.remove(file_path)
+            logger.error(
+                "Failed to write shard %s/data_%04d.h5, cleaned up partial output",
+                domain,
+                shard_idx,
+                exc_info=True,
+            )
+            raise

astrai/serialization.py

@@ -3,7 +3,7 @@ import json
 import time
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Any, Dict, Union
+from typing import Any, Dict, Optional, Union
 
 import safetensors.torch as st
 import torch
@@ -180,3 +180,22 @@ class Checkpoint:
             extra=extra,
             config=config,
         )
+
+    @classmethod
+    def load_any(cls, save_dir: str, broadcast: bool = False) -> Optional["Checkpoint"]:
+        save_path = Path(save_dir)
+        meta_path = save_path / _META_FILE
+        weights_path = save_path / _WEIGHTS_FILE
+
+        if meta_path.exists():
+            return cls.load(save_dir, broadcast=broadcast)
+
+        if weights_path.exists():
+            state_dict = load_state_dict(weights_path, broadcast=broadcast)
+            config = {}
+            config_path = save_path / _CONFIG_FILE
+            if config_path.exists():
+                config = load_json(config_path, broadcast)
+            return cls(state_dict=state_dict, config=config)
+
+        return None

astrai/tokenize/chat_template.py

@@ -1,13 +1,10 @@
-from dataclasses import dataclass
 from typing import Any, Dict, List, Optional
 
 from jinja2 import Template
 
-# Message type for chat messages
 type MessageType = Dict[str, Any]
 
 
-@dataclass
 class ChatTemplate:
     """A chat template with Jinja2 rendering support.
 
@@ -15,23 +12,24 @@ class ChatTemplate:
         name: Unique identifier for the template.
         template_str: Jinja2 template string.
         description: Optional description.
-        default_variables: Optional dictionary of default variable values
+        default_variables: Optional dictionary of default variable values.
-            that will be passed to the template if not overridden during rendering.
         special_tokens: Optional dictionary mapping token names to their string values.
-            These tokens are automatically added to the template variables.
     """
 
-    name: str
+    def __init__(
-    template_str: str
+        self,
-    description: str = ""
+        name: str = "",
-    default_variables: Dict[str, Any] = None
+        template_str: str = "",
-    special_tokens: Dict[str, str] = None
+        description: str = "",
-
+        default_variables: Optional[Dict[str, Any]] = None,
-    def __post_init__(self):
+        special_tokens: Optional[Dict[str, str]] = None,
-        if self.default_variables is None:
+    ):
-            self.default_variables = {}
+        self.name = name
-        if self.special_tokens is None:
+        self.template_str = template_str
-            self.special_tokens = {}
+        self.description = description
+        self.default_variables = default_variables or {}
+        self.special_tokens = special_tokens or {}
+        self._compiled: Template = Template(template_str)
 
     @classmethod
     def from_string(
@@ -43,7 +41,7 @@ class ChatTemplate:
     ) -> "ChatTemplate":
         """Create a ChatTemplate instance directly from a template string."""
         return cls(
-            name="",  # empty name for ad‑hoc templates
+            name="",
             template_str=template_str,
             description=description,
             default_variables=default_variables,
@@ -73,5 +71,4 @@ class ChatTemplate:
         if system_prompt is not None:
             variables["system_prompt"] = system_prompt
 
-        jinja_template = Template(self.template_str)
+        return self._compiled.render(**variables)
-        return jinja_template.render(**variables)

astrai/trainer/schedule.py

@@ -2,7 +2,7 @@
 
 import math
 from abc import ABC, abstractmethod
-from typing import Any, Dict, List, Type
+from typing import Any, Dict, List
 
 from torch.optim.lr_scheduler import LRScheduler
 
@@ -31,7 +31,6 @@ class SchedulerFactory(BaseFactory["BaseScheduler"]):
     """Factory class for creating learning rate schedulers.
 
     Supports decorator-based registration for extensible scheduler types.
-    Also supports creation from ScheduleConfig objects.
 
     Example usage:
         @SchedulerFactory.register("custom")
@@ -41,33 +40,6 @@ class SchedulerFactory(BaseFactory["BaseScheduler"]):
         scheduler = SchedulerFactory.create("custom", optimizer, **kwargs)
     """
 
-    @classmethod
-    def _validate_component(cls, scheduler_cls: Type[BaseScheduler]):
-        """Validate that the scheduler class inherits from BaseScheduler."""
-        if not issubclass(scheduler_cls, BaseScheduler):
-            raise TypeError(f"{scheduler_cls.__name__} must inherit from BaseScheduler")
-
-    @classmethod
-    def create(
-        cls, optimizer, schedule_type: str = "none", **kwargs
-    ) -> "BaseScheduler":
-        """Create a scheduler instance by type name.
-
-        Args:
-            optimizer: PyTorch optimizer
-            schedule_type: Type of scheduler ("cosine", "sgdr")
-            **kwargs: Arguments passed to the scheduler constructor
-
-        Returns:
-            Scheduler instance
-        """
-        return super().create(schedule_type, optimizer, **kwargs)
-
-    @classmethod
-    def available_types(cls) -> list:
-        """Return list of registered scheduler type names."""
-        return cls.list_registered()
-
 
 # ----------- Scheduler implementations -----------
 
@@ -192,3 +164,66 @@ class SGDRScheduler(BaseScheduler):
         self.min_rate = state_dict.pop("min_rate")
         self.t_mult = state_dict.pop("t_mult")
         super().load_state_dict(state_dict)
+
+
+@SchedulerFactory.register("wsd")
+class WSDScheduler(BaseScheduler):
+    """WSD (Warmup-Stable-Decay) scheduler with sqrt cooldown.
+
+    warmup_steps:  linear warmup from min_rate to 1.0
+    stable_steps:   constant at base_lr
+    decay_steps:    sqrt decay from base_lr to min_rate
+    min_rate:       minimum lr as fraction of base_lr (default 0.0)
+    """
+
+    def __init__(
+        self,
+        optimizer,
+        warmup_steps: int,
+        stable_steps: int,
+        decay_steps: int,
+        min_rate: float = 0.0,
+        last_epoch: int = -1,
+    ):
+        self.warmup_steps = warmup_steps
+        self.stable_steps = stable_steps
+        self.decay_steps = decay_steps
+        self.min_rate = min_rate
+        self.total_steps = warmup_steps + stable_steps + decay_steps
+        super().__init__(optimizer, last_epoch)
+
+    def get_lr(self) -> List[float]:
+        if self.last_epoch < self.warmup_steps:
+            factor = self.last_epoch / max(self.warmup_steps, 1)
+            return [base_lr * factor for base_lr in self.base_lrs]
+
+        offset = self.last_epoch - self.warmup_steps
+
+        if offset < self.stable_steps:
+            return list(self.base_lrs)
+
+        decay_ratio = (offset - self.stable_steps) / max(self.decay_steps, 1)
+        decay_ratio = min(decay_ratio, 1.0)
+        factor = (1.0 - self.min_rate) * (1.0 - decay_ratio) ** 2 + self.min_rate
+        return [base_lr * factor for base_lr in self.base_lrs]
+
+    def state_dict(self):
+        state = super().state_dict()
+        state.update(
+            {
+                "warmup_steps": self.warmup_steps,
+                "stable_steps": self.stable_steps,
+                "decay_steps": self.decay_steps,
+                "min_rate": self.min_rate,
+                "total_steps": self.total_steps,
+            }
+        )
+        return state
+
+    def load_state_dict(self, state_dict):
+        self.warmup_steps = state_dict.pop("warmup_steps")
+        self.stable_steps = state_dict.pop("stable_steps")
+        self.decay_steps = state_dict.pop("decay_steps")
+        self.min_rate = state_dict.pop("min_rate")
+        self.total_steps = state_dict.pop("total_steps")
+        super().load_state_dict(state_dict)

astrai/trainer/strategy.py

@@ -1,41 +1,28 @@
 """Training strategy implementations with factory pattern."""
 
-import copy
 from abc import ABC, abstractmethod
-from typing import Any, Callable, Dict, Union
+from typing import Callable, Dict, Union
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from torch import Tensor
-from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
-from torch.nn.parallel import DistributedDataParallel as DDP
 
 from astrai.factory import BaseFactory
 
 
-def unwrap_model(model: nn.Module) -> nn.Module:
+def create_ref_model(
-    if isinstance(model, DDP):
+    model_fn: Callable[[], nn.Module], state_dict: Dict[str, Tensor]
-        return model.module
+) -> nn.Module:
-    if isinstance(model, FSDP):
+    """Create a frozen reference model from model_fn + full state dict."""
-        return model._fsdp_wrapped_module
+    ref_model = model_fn()
-    return model
+    ref_model.load_state_dict(state_dict)
-
-
-def create_ref_model(model: nn.Module) -> nn.Module:
-    """Create a reference model for DPO/GRPO training.
-
-    Handles DDP-wrapped models safely by unwrapping first,
-    then creating a deep copy with frozen gradients.
-    """
-    original_model = unwrap_model(model)
-    ref_model = copy.deepcopy(original_model)
     ref_model.requires_grad_(False)
     ref_model.eval()
     return ref_model
 
 
-def move_to_device(batch: Dict[str, Tensor], device: str) -> Any:
+def move_to_device(batch: Dict[str, Tensor], device: str) -> Dict[str, Tensor]:
     """Move batch tensors to specified device with non-blocking transfer."""
     return {key: value.to(device, non_blocking=True) for key, value in batch.items()}
 
@@ -45,7 +32,7 @@ def get_logprobs(
     input_ids: Tensor,
     mask: Tensor,
     reduction: str,
-):
+) -> Tensor:
     """Compute token-wise log probabilities from model outputs.
 
     Args:
@@ -83,14 +70,35 @@ def get_logprobs(
         return token_logprobs * shifted_mask
 
 
+def make_doc_boundary_mask(position_ids: Tensor) -> Tensor:
+    S = position_ids.size(1)
+    device = position_ids.device
+    boundaries = position_ids[:, 1:] <= position_ids[:, :-1]
+    doc_ids = torch.cat(
+        [
+            torch.zeros(position_ids.size(0), 1, dtype=torch.long, device=device),
+            boundaries.long().cumsum(dim=1),
+        ],
+        dim=1,
+    )
+    same_doc = doc_ids.unsqueeze(-1) == doc_ids.unsqueeze(-2)
+    causal = torch.tril(torch.ones(S, S, dtype=torch.bool, device=device))
+    return (same_doc & causal).unsqueeze(1)
+
+
 class BaseStrategy(ABC):
     """Abstract base class for training strategies."""
 
     def __init__(
-        self, model: Union[Callable[..., Dict[str, Tensor]]], device: str, **kwargs
+        self,
+        model: Union[nn.Module, Callable[..., Dict[str, Tensor]]],
+        device: str,
+        **kwargs,
     ):
         self.model = model
         self.device = device
+        self.executor = kwargs.pop("executor", None)
+        self.model_fn = kwargs.pop("model_fn", None)
         self.extra_kwargs = kwargs
 
     @abstractmethod
@@ -124,32 +132,6 @@ class StrategyFactory(BaseFactory["BaseStrategy"]):
         strategy = StrategyFactory.create("custom", model, device)
     """
 
-    @classmethod
-    def _validate_component(cls, strategy_cls: type):
-        """Validate that the strategy class inherits from BaseStrategy."""
-        if not issubclass(strategy_cls, BaseStrategy):
-            raise TypeError(f"{strategy_cls.__name__} must inherit from BaseStrategy")
-
-    @classmethod
-    def create(cls, train_type: str, model, device: str, **kwargs) -> "BaseStrategy":
-        """Create a strategy instance based on training type.
-
-        Args:
-            train_type: Type of training ("seq", "sft", "dpo", "grpo")
-            model: Model instance for the strategy
-            device: Device to run the strategy on
-            **kwargs: Additional arguments passed to strategy constructor
-
-        Returns:
-            Strategy instance
-        """
-        return super().create(train_type, model, device, **kwargs)
-
-    @classmethod
-    def available_strategies(cls) -> list:
-        """Return list of registered strategy names."""
-        return cls.list_registered()
-
 
 # ============== Strategy Classes ==============
 # All strategies are registered at class definition time using the decorator
@@ -162,7 +144,13 @@ class SEQStrategy(BaseStrategy):
     Computes cross-entropy loss for next token prediction.
     """
 
-    def __init__(self, model, device, label_smoothing: float = 0.0, **kwargs):
+    def __init__(
+        self,
+        model: Union[nn.Module, Callable[..., Dict[str, Tensor]]],
+        device: str,
+        label_smoothing: float = 0.0,
+        **kwargs,
+    ):
         super().__init__(model, device, **kwargs)
         self.label_smoothing = label_smoothing
 
@@ -187,21 +175,31 @@ class SFTStrategy(BaseStrategy):
     Applies cross-entropy loss only to tokens where loss_mask is True.
     """
 
-    def __init__(self, model, device, label_smoothing: float = 0.0, **kwargs):
+    def __init__(
+        self,
+        model: Union[nn.Module, Callable[..., Dict[str, Tensor]]],
+        device: str,
+        label_smoothing: float = 0.0,
+        **kwargs,
+    ):
         super().__init__(model, device, **kwargs)
         self.label_smoothing = label_smoothing
 
     def compute_loss(self, batch: Dict[str, Tensor]) -> Tensor:
         batch = move_to_device(batch, self.device)
-        input_ids, target_ids, loss_mask = (
+        input_ids, target_ids, position_ids, loss_mask = (
             batch["input_ids"],
             batch["target_ids"],
+            batch["position_ids"],
             batch["loss_mask"],
         )
 
         ignore_index = -100
-        logits = self.model(input_ids=input_ids)["logits"]
+        input_mask = make_doc_boundary_mask(position_ids)
         target_ids = target_ids.masked_fill(loss_mask == 0, ignore_index)
+        logits = self.model(
+            input_ids=input_ids, position_ids=position_ids, input_mask=input_mask
+        )["logits"]
 
         loss = F.cross_entropy(
             input=logits.flatten(0, 1).float(),
@@ -230,7 +228,9 @@ class DPOStrategy(BaseStrategy):
         **kwargs,
     ):
         super().__init__(model, device, **kwargs)
-        self.ref_model = create_ref_model(model)
+        self.ref_model = create_ref_model(
+            self.model_fn, self.executor.unwrap_model(model)
+        ).to(device=self.device)
         self.beta = beta
         self.reduction = reduction
 
@@ -284,7 +284,9 @@ class GRPOStrategy(BaseStrategy):
         **kwargs,
     ):
         super().__init__(model, device, **kwargs)
-        self.ref_model = create_ref_model(model)
+        self.ref_model = create_ref_model(
+            self.model_fn, self.executor.unwrap_model(model)
+        ).to(device=self.device)
         self.clip_eps = clip_eps
         self.kl_coef = kl_coef
         self.group_size = group_size
@@ -294,8 +296,7 @@ class GRPOStrategy(BaseStrategy):
 
     def sync_ref_model(self):
         """Copy current model weights to ref model."""
-        ref_state = self.model.state_dict()
+        self.ref_model.load_state_dict(self.executor.unwrap_model(self.model))
-        self.ref_model.load_state_dict(ref_state)
 
     def compute_loss(self, batch: Dict[str, Tensor]) -> Tensor:
         self._step += 1

astrai/trainer/train_callback.py

@@ -146,8 +146,7 @@ class CheckpointCallback(TrainCallback):
         self.last_ckpt_iter = 0
 
     def _save_checkpoint(self, context: TrainContext):
-        unwrapped = context.executor.unwrap_model(context.model)
+        state_dict = context.executor.unwrap_model(context.model)
-        state_dict = unwrapped.state_dict()
         self.last_ckpt_iter = context.iteration
 
         if get_rank() == 0:
@@ -155,11 +154,13 @@ class CheckpointCallback(TrainCallback):
                 self.save_dir, f"epoch_{context.epoch}_iter_{context.iteration}"
             )
             extra = self.save_extra_fn(context)
+            meta = context.config.to_dict()
             context.checkpoint = Checkpoint(
                 state_dict=state_dict,
                 epoch=context.epoch,
                 iteration=context.iteration,
                 extra=extra,
+                meta=meta,
                 config=context.model_config,
             )
             context.checkpoint.save(save_path)
@@ -214,7 +215,7 @@ class ProgressBarCallback(TrainCallback):
             "loss": f"{context.loss:.4f}",
             "lr": f"{context.optimizer.param_groups[-1]['lr']:.2e}",
         }
-        if context.val_loss > 0:
+        if context.val_loss is not None:
             postfix["val_loss"] = f"{context.val_loss:.4f}"
         self.progress_bar.set_postfix(postfix)
         self.progress_bar.update(1)
@@ -236,6 +237,7 @@ class MetricLoggerCallback(TrainCallback):
         metrics: List[str] = None,
     ):
         self.last_log_iter = 0
+        self._last_val_loss = None
         self.save_interval = save_interval
         self.log_interval = log_interval
         self.metrics = metrics or ["loss", "lr"]
@@ -257,41 +259,54 @@ class MetricLoggerCallback(TrainCallback):
             "grad_nan_num": ctx_get_grad_nan_num,
         }
 
-    def _get_log_data(self, context: TrainContext):
+    def _metrics(self, context: TrainContext, names):
         return {
-            "timestamp": time.strftime("%Y-%m-%dT%H:%M:%S"),
+            m: self._metric_funcs[m](context)
-            "epoch": context.epoch,
+            for m in names
-            "iter": context.iteration,
+            if self._metric_funcs[m](context) is not None
-            **{m: self._metric_funcs[m](context) for m in self.metrics},
         }
 
     @only_on_rank(0)
-    def _add_log(self, log_data):
+    def _append(self, event_type: str, context: TrainContext, **extra):
-        self.log_cache.append(log_data)
+        entry = {
+            "type": event_type,
+            "timestamp": time.strftime("%Y-%m-%dT%H:%M:%S"),
+            "epoch": context.epoch,
+            "iter": context.iteration,
+            **extra,
+        }
+        self.log_cache.append(entry)
 
     @only_on_rank(0)
-    def _save_log(self, epoch, iter):
+    def _flush(self, epoch, iter):
         log_file = self.log_dir / f"epoch_{epoch}_iter_{iter}_metric.jsonl"
-
+        log_file.parent.mkdir(parents=True, exist_ok=True)
         with open(log_file, "w") as f:
             for log in self.log_cache:
                 f.write(json.dumps(log) + "\n")
 
     def on_batch_end(self, context):
         if context.iteration % self.log_interval == 0:
-            log_data = self._get_log_data(context)
+            step_metrics = [m for m in self.metrics if m != "val_loss"]
-            self._add_log(log_data)
+            self._append("step", context, **self._metrics(context, step_metrics))
-
         if context.iteration - self.last_log_iter >= self.save_interval:
-            self._save_log(context.epoch, context.iteration)
+            self._flush(context.epoch, context.iteration)
             self.last_log_iter = context.iteration
 
+    def on_optimizer_step(self, context):
+        if context.val_loss is not None and context.val_loss != self._last_val_loss:
+            self._append("validation", context, val_loss=context.val_loss)
+            self._last_val_loss = context.val_loss
+
+    def on_epoch_end(self, context):
+        self._append("epoch", context)
+
     def on_train_end(self, context):
         if context.iteration != self.last_log_iter:
-            self._save_log(context.epoch, context.iteration)
+            self._flush(context.epoch, context.iteration)
 
     def on_error(self, context):
-        self._save_log(context.epoch, context.iteration)
+        self._flush(context.epoch, context.iteration)
 
 
 @CallbackFactory.register("validation")

astrai/trainer/train_context.py

@@ -1,9 +1,10 @@
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Optional, Self
+from typing import Any, Dict, Optional, Self
 
+import torch
 import torch.nn as nn
-from torch.utils.data import DataLoader
+from torch.utils.data import DataLoader, random_split
 
 from astrai.config.train_config import TrainConfig
 from astrai.dataset import ResumableDistributedSampler
@@ -11,7 +12,7 @@ from astrai.model.components.lora import inject_lora
 from astrai.parallel.executor import BaseExecutor, ExecutorFactory
 from astrai.parallel.setup import get_current_device, get_rank, get_world_size
 from astrai.protocols import OptimizerProtocol, SchedulerProtocol
-from astrai.serialization import Checkpoint, load_json, load_model_weights
+from astrai.serialization import Checkpoint, load_json
 from astrai.trainer.strategy import BaseStrategy, StrategyFactory
 
 
@@ -30,12 +31,12 @@ class TrainContext:
     epoch: int = field(default=0)
     iteration: int = field(default=0)
     loss: float = field(default=0.0)
-    val_dataloader: DataLoader = field(default=None)
+    val_dataloader: Optional[DataLoader] = field(default=None)
-    val_loss: float = field(default=0.0)
+    val_loss: Optional[float] = field(default=None)
 
     world_size: int = field(default=1)
     rank: int = field(default=0)
-    kwargs: dict = field(default_factory=dict)
+    kwargs: Dict[str, Any] = field(default_factory=dict)
 
 
 class TrainContextBuilder:
@@ -81,21 +82,15 @@ class TrainContextBuilder:
             executor=executor,
         )
 
-        if self._resume_dir is not None:
+        if self._resume_dir:
-            resume_path = Path(self._resume_dir)
+            checkpoint = Checkpoint.load_any(self._resume_dir)
-            if (resume_path / "meta.json").exists():
+            if checkpoint is not None:
-                checkpoint = Checkpoint.load(self._resume_dir)
+                model.load_state_dict(checkpoint.state_dict, strict=False)
-                state_dict = checkpoint.state_dict
                 if checkpoint.config:
                     context.model_config = checkpoint.config
-            else:
+                context.epoch = checkpoint.epoch or cfg.start_epoch
-                checkpoint = None
+                context.iteration = checkpoint.iteration or cfg.start_batch
-                state_dict = load_model_weights(self._resume_dir)
+                context.checkpoint = checkpoint
-            model.load_state_dict(state_dict, strict=False)
-            if checkpoint is not None:
-                context.epoch = cfg.start_epoch
-                context.iteration = cfg.start_batch
-            context.checkpoint = checkpoint
 
         if cfg.lora is not None:
             inject_lora(
@@ -108,15 +103,27 @@ class TrainContextBuilder:
         context.optimizer = cfg.optimizer_fn(model)
         context.scheduler = cfg.scheduler_fn(context.optimizer)
 
+        train_dataset = cfg.dataset
+        val_dataset = cfg.val_dataset
+
+        if val_dataset is None and cfg.val_split is not None:
+            n_total = len(cfg.dataset)
+            n_val = max(1, int(n_total * cfg.val_split))
+            n_train = n_total - n_val
+            generator = torch.Generator().manual_seed(cfg.random_seed)
+            train_dataset, val_dataset = random_split(
+                cfg.dataset, [n_train, n_val], generator=generator
+            )
+
         sampler_offset = context.iteration * cfg.batch_per_device
         sampler = ResumableDistributedSampler(
-            data_source=cfg.dataset,
+            data_source=train_dataset,
             start_epoch=context.epoch,
             start_iter=sampler_offset,
             seed=cfg.random_seed,
         )
         context.dataloader = DataLoader(
-            cfg.dataset,
+            train_dataset,
             batch_size=cfg.batch_per_device,
             sampler=sampler,
             num_workers=cfg.num_workers,
@@ -124,16 +131,16 @@ class TrainContextBuilder:
             prefetch_factor=cfg.prefetch_factor,
         )
 
-        if cfg.val_dataset is not None:
+        if val_dataset is not None:
             val_sampler = ResumableDistributedSampler(
-                data_source=cfg.val_dataset,
+                data_source=val_dataset,
                 start_epoch=0,
                 start_iter=0,
                 seed=cfg.random_seed,
                 shuffle=False,
             )
             context.val_dataloader = DataLoader(
-                cfg.val_dataset,
+                val_dataset,
                 batch_size=cfg.batch_per_device,
                 sampler=val_sampler,
                 num_workers=cfg.num_workers,
@@ -159,9 +166,11 @@ class TrainContextBuilder:
                         obj.load_state_dict(extra[name])
 
         context.strategy = StrategyFactory.create(
+            cfg.strategy,
             model=context.model,
-            train_type=cfg.strategy,
             device=device,
+            executor=executor,
+            model_fn=cfg.model_fn,
             **cfg.extra_kwargs,
         )
 

astrai/trainer/trainer.py

@@ -34,6 +34,7 @@ class Trainer:
                 cfg.ckpt_dir,
                 cfg.ckpt_interval,
             ),
+            CallbackFactory.create("validation"),
             CallbackFactory.create(
                 "metric_logger",
                 log_dir=cfg.log_dir,
@@ -43,7 +44,6 @@ class Trainer:
             ),
             CallbackFactory.create("progress_bar", cfg.n_epoch),
             CallbackFactory.create("gradient_clipping", cfg.max_grad_norm),
-            CallbackFactory.create("validation"),
         ]
         return callbacks
 
@@ -68,9 +68,8 @@ class Trainer:
                 self._call_callbacks("on_epoch_begin", context)
 
                 for batch in context.dataloader:
-                    self._call_callbacks("on_batch_begin", context)
-
                     with executor.accumulate(context.model):
+                        self._call_callbacks("on_batch_begin", context)
                         loss = context.strategy(batch)
                         context.loss = loss.item()
                         stand_loss = loss / executor.grad_accum_steps

scripts/eval/evaluate_humaneval.py

@@ -0,0 +1,334 @@
+"""HumanEval code generation benchmark.
+
+Generates n completions per problem, extracts function bodies, executes
+against hidden tests, and computes pass@k.
+
+Usage::
+
+    python scripts/tools/evaluate_humaneval.py --param_path ./params \
+        --data_path HumanEval.jsonl.gz --output results.json \
+        --num_samples 200 --temperature 0.8 --max_tokens 512
+"""
+
+import argparse
+import json
+import os
+import re
+from math import prod
+from multiprocessing import Process, Queue
+from typing import Dict, List, Optional, Tuple
+
+import numpy as np
+import torch
+import tqdm
+
+from astrai.inference import InferenceEngine
+from astrai.model import AutoModel
+from astrai.tokenize import AutoTokenizer
+
+HUMANEVAL_URL = (
+    "https://github.com/openai/human-eval/raw/master/data/HumanEval.jsonl.gz"
+)
+
+_STOP_SEQUENCES = [
+    "\nclass ",
+    "\ndef ",
+    "\n# ",
+    "\nif __name__",
+    "\nprint(",
+    "\n\n\n",
+]
+
+
+def _download_humaneval(data_path: str):
+    if os.path.exists(data_path):
+        return
+    import gzip
+    import urllib.request
+
+    os.makedirs(os.path.dirname(data_path) or ".", exist_ok=True)
+    print(f"Downloading HumanEval from {HUMANEVAL_URL} ...")
+    tmp = data_path + ".tmp"
+    urllib.request.urlretrieve(HUMANEVAL_URL, tmp)
+    with gzip.open(tmp, "rb") as f_in:
+        with open(data_path, "wb") as f_out:
+            f_out.write(f_in.read())
+    os.remove(tmp)
+    print(f"  saved to {data_path}")
+
+
+def _load_problems(data_path: str) -> List[dict]:
+    problems = []
+    with open(data_path, "r", encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                problems.append(json.loads(line))
+    return problems
+
+
+def _extract_function_body(code: str, entry_point: str) -> Optional[str]:
+    """Extract the function body from a completion."""
+    pattern = rf"def\s+{re.escape(entry_point)}\b[^:]*:"
+    match = re.search(pattern, code)
+    if not match:
+        # Use the full code as-is if we can't find the function
+        return code
+
+    body_start = match.end()
+    lines = code[body_start:].split("\n")
+    body_lines = []
+    started = False
+
+    for line in lines:
+        stripped = line.rstrip()
+        if not stripped and not started:
+            continue
+        if not stripped and started:
+            body_lines.append("")
+            continue
+        if not started:
+            started = True
+        if stripped.lstrip() == stripped and started:
+            break
+        body_lines.append(stripped)
+
+    body = "\n".join(body_lines)
+    if not body.strip():
+        return None
+    return body
+
+
+def _trim_stop_sequences(text: str) -> str:
+    for stop in _STOP_SEQUENCES:
+        idx = text.find(stop)
+        if idx != -1:
+            text = text[:idx]
+    return text
+
+
+def _execute_code(problem: dict, completion: str, timeout: float = 3.0) -> bool:
+    """Run the completion against hidden tests in a subprocess."""
+
+    def _worker(queue, full_code):
+        try:
+            namespace = {}
+            exec(full_code, namespace)
+            check = namespace.get("check")
+            if check is None:
+                queue.put(False)
+                return
+            check(namespace.get(problem["entry_point"]))
+            queue.put(True)
+        except Exception:
+            queue.put(False)
+
+    full_code = problem["prompt"] + completion + "\n" + problem["test"]
+
+    queue: Queue = Queue()
+    proc = Process(target=_worker, args=(queue, full_code))
+    proc.start()
+    proc.join(timeout)
+
+    if proc.is_alive():
+        proc.terminate()
+        proc.join()
+        return False
+
+    try:
+        return queue.get_nowait()
+    except Exception:
+        return False
+
+
+def _pass_at_k(n: int, c: int, k: int) -> float:
+    """Unbiased estimator of pass@k."""
+    if n - c < k:
+        return 1.0
+    return 1.0 - float(prod(1.0 - k / np.arange(n - c + 1, n + 1)))
+
+
+def _deduplicate(completions: List[str]) -> List[str]:
+    seen = set()
+    unique = []
+    for c in completions:
+        if c not in seen:
+            seen.add(c)
+            unique.append(c)
+    return unique
+
+
+def _generate(
+    engine: InferenceEngine,
+    prompt: str,
+    num_samples: int,
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int,
+    batch_size: int,
+) -> List[str]:
+    batches = [prompt] * min(batch_size, num_samples)
+    completions = []
+    remaining = num_samples
+
+    while remaining > 0:
+        current = min(batch_size, remaining)
+        batch_prompts = batches[:current]
+        outputs = engine.generate(
+            prompt=batch_prompts,
+            stream=False,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+        )
+        if isinstance(outputs, str):
+            outputs = [outputs]
+        completions.extend(outputs)
+        remaining -= current
+
+    return _deduplicate(completions)
+
+
+def evaluate(
+    engine: InferenceEngine,
+    problems: List[dict],
+    num_samples: int,
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int,
+    batch_size: int,
+    k_values: Tuple[int, ...] = (1, 10, 100),
+) -> Dict:
+    results = {}
+    all_pass_at_k = {k: [] for k in k_values}
+
+    for problem in tqdm.tqdm(problems, desc="HumanEval", unit="problem"):
+        task_id = problem["task_id"]
+        prompt = problem["prompt"]
+        entry_point = problem["entry_point"]
+
+        raw_completions = _generate(
+            engine,
+            prompt,
+            num_samples,
+            max_tokens,
+            temperature,
+            top_p,
+            top_k,
+            batch_size,
+        )
+
+        completions = []
+        for raw in raw_completions:
+            trimmed = _trim_stop_sequences(raw)
+            body = _extract_function_body(trimmed, entry_point)
+            if body:
+                completions.append(body)
+
+        passed = 0
+        for comp in completions:
+            if _execute_code(problem, comp):
+                passed += 1
+
+        n = len(completions)
+        c = passed
+        result = {"task_id": task_id, "n": n, "passed": c}
+        for k in k_values:
+            result[f"pass@{k}"] = round(_pass_at_k(n, c, k), 4)
+            all_pass_at_k[k].append(_pass_at_k(n, c, k))
+        results[task_id] = result
+
+    summary = {}
+    for k in k_values:
+        vals = all_pass_at_k[k]
+        summary[f"pass@{k}"] = round(float(np.mean(vals)), 4)
+    results["_summary"] = summary
+
+    return results
+
+
+def main():
+    parser = argparse.ArgumentParser(description="HumanEval benchmark")
+    parser.add_argument(
+        "--param_path", type=str, default="./params", help="Model directory"
+    )
+    parser.add_argument(
+        "--data_path",
+        type=str,
+        default="./humaneval/HumanEval.jsonl",
+        help="HumanEval JSONL file (auto-download if missing)",
+    )
+    parser.add_argument("--output", type=str, default=None, help="Output JSON path")
+    parser.add_argument(
+        "--num_samples",
+        type=int,
+        default=200,
+        help="Completions per problem",
+    )
+    parser.add_argument(
+        "--max_tokens", type=int, default=512, help="Max generation tokens"
+    )
+    parser.add_argument(
+        "--temperature", type=float, default=0.8, help="Sampling temperature"
+    )
+    parser.add_argument("--top_p", type=float, default=0.95, help="Top-p sampling")
+    parser.add_argument("--top_k", type=int, default=50, help="Top-k sampling")
+    parser.add_argument(
+        "--batch_size", type=int, default=1, help="Inference batch size"
+    )
+    parser.add_argument(
+        "--problems",
+        type=int,
+        nargs="+",
+        default=None,
+        help="Specific problem indices (0-based)",
+    )
+    args = parser.parse_args()
+
+    _download_humaneval(args.data_path)
+    problems = _load_problems(args.data_path)
+    if args.problems:
+        problems = [problems[i] for i in args.problems if i < len(problems)]
+
+    model = AutoModel.from_pretrained(args.param_path)
+    tokenizer = AutoTokenizer.from_pretrained(args.param_path)
+    model.to(device="cuda", dtype=torch.bfloat16)
+
+    engine = InferenceEngine(
+        model=model,
+        tokenizer=tokenizer,
+        max_batch_size=args.batch_size,
+    )
+
+    results = evaluate(
+        engine=engine,
+        problems=problems,
+        num_samples=args.num_samples,
+        max_tokens=args.max_tokens,
+        temperature=args.temperature,
+        top_p=args.top_p,
+        top_k=args.top_k,
+        batch_size=args.batch_size,
+        k_values=(1, 10, 100),
+    )
+
+    summary = results.pop("_summary")
+    print(f"\n{'=' * 60}")
+    for k, v in summary.items():
+        print(f"  {k}: {v:.2%}")
+    print(f"{'=' * 60}")
+
+    if args.output:
+        results["_summary"] = summary
+        with open(args.output, "w", encoding="utf-8") as f:
+            json.dump(results, f, indent=2, ensure_ascii=False)
+        print(f"Results saved to {args.output}")
+
+    engine.shutdown()
+
+
+if __name__ == "__main__":
+    main()

scripts/eval/evaluate_ifd.py

@@ -0,0 +1,293 @@
+"""IFD (Instruction Following Difficulty) data quality scoring.
+
+Computes IFD scores for instruction-response pairs to guide data selection.
+IFD = conditional_NLL / unconditional_NLL, where:
+
+- conditional_NLL: average CE loss on response tokens given instruction context
+- unconditional_NLL: average CE loss on response tokens alone
+
+Higher IFD (close to 1) = instruction provides less help = harder sample.
+Lower IFD (close to 0) = instruction provides strong guidance = easy sample.
+IFD > 1 = instruction misleads the model = likely low-quality data.
+
+Usage::
+
+    python scripts/eval/ifd.py --param_path ./params \
+        --input data.jsonl --output data_with_ifd.jsonl \
+        --instr_key instruction --resp_key response
+
+Disable chat template::
+
+    python scripts/eval/ifd.py --param_path ./params \
+        --input data.jsonl --output data_with_ifd.jsonl \
+        --instr_key instruction --resp_key response \
+        --no_chat_template
+"""
+
+import argparse
+import json
+
+import torch
+import torch.nn.functional as F
+import tqdm
+
+from astrai.model import AutoModel
+from astrai.tokenize import AutoTokenizer
+
+
+def compute_ifd(
+    model,
+    tokenizer,
+    instruction: str,
+    response: str,
+    device: str,
+    max_len: int = 2048,
+    use_chat_template: bool = False,
+) -> dict:
+    if use_chat_template:
+        return _compute_ifd_with_template(
+            model, tokenizer, instruction, response, device, max_len
+        )
+    return _compute_ifd_raw(model, tokenizer, instruction, response, device, max_len)
+
+
+def _compute_ifd_raw(model, tokenizer, instruction, response, device, max_len) -> dict:
+    instr_ids = tokenizer.encode(instruction)
+    resp_ids = tokenizer.encode(response)
+
+    if not resp_ids:
+        return {
+            "L_cond": None,
+            "L_uncond": None,
+            "ifd": None,
+            "error": "empty response",
+        }
+
+    qa_len = len(instr_ids) + len(resp_ids)
+    if qa_len > max_len:
+        overflow = qa_len - max_len
+        instr_ids = instr_ids[overflow:]
+
+    instr_len = len(instr_ids)
+    resp_len = len(resp_ids)
+
+    qa_ids = instr_ids + resp_ids
+    qa_tensor = torch.tensor([qa_ids], device=device, dtype=torch.long)
+
+    with torch.inference_mode():
+        logits_qa = model(qa_tensor)["logits"][0]
+
+    resp_logits = logits_qa[instr_len - 1 : -1]
+    resp_targets = torch.tensor(resp_ids, device=device, dtype=torch.long)
+    L_cond = F.cross_entropy(resp_logits, resp_targets, reduction="mean").item()
+
+    resp_tensor = torch.tensor([resp_ids], device=device, dtype=torch.long)
+
+    with torch.inference_mode():
+        logits_resp = model(resp_tensor)["logits"][0]
+
+    unp_logits = logits_resp[:-1]
+    unp_targets = resp_tensor[0, 1:]
+    L_uncond = F.cross_entropy(unp_logits, unp_targets, reduction="mean").item()
+
+    ifd = L_cond / L_uncond if L_uncond > 0 else None
+
+    return {
+        "L_cond": round(L_cond, 6),
+        "L_uncond": round(L_uncond, 6),
+        "ifd": round(ifd, 6) if ifd is not None else None,
+        "instr_len": instr_len,
+        "resp_len": resp_len,
+        "error": None,
+    }
+
+
+def _compute_ifd_with_template(
+    model, tokenizer, instruction, response, device, max_len
+) -> dict:
+    instr_prefix = tokenizer.apply_chat_template(
+        [{"role": "user", "content": instruction}],
+        tokenize=False,
+        add_generation_prompt=True,
+    )
+    full_text = tokenizer.apply_chat_template(
+        [
+            {"role": "user", "content": instruction},
+            {"role": "assistant", "content": response},
+        ],
+        tokenize=False,
+        add_generation_prompt=False,
+    )
+
+    full_ids = tokenizer.encode(full_text)
+    prefix_ids = tokenizer.encode(instr_prefix)
+    resp_ids = tokenizer.encode(response)
+
+    if not resp_ids:
+        return {
+            "L_cond": None,
+            "L_uncond": None,
+            "ifd": None,
+            "error": "empty response",
+        }
+
+    if len(full_ids) > max_len:
+        overflow = len(full_ids) - max_len
+        full_ids = full_ids[overflow:]
+        prefix_len = len(prefix_ids) - overflow
+        prefix_len = max(0, prefix_len)
+    else:
+        prefix_len = len(prefix_ids)
+
+    cond_tensor = torch.tensor([full_ids], device=device, dtype=torch.long)
+
+    with torch.inference_mode():
+        logits_qa = model(cond_tensor)["logits"][0]
+
+    resp_start = prefix_len - 1
+    resp_end = len(full_ids) - 1
+    if resp_end <= resp_start:
+        return {
+            "L_cond": None,
+            "L_uncond": None,
+            "ifd": None,
+            "error": "response truncated entirely",
+        }
+
+    resp_logits = logits_qa[resp_start:resp_end]
+    resp_targets = torch.tensor(full_ids[prefix_len:], device=device, dtype=torch.long)
+    L_cond = F.cross_entropy(resp_logits, resp_targets, reduction="mean").item()
+
+    resp_tensor = torch.tensor([resp_ids], device=device, dtype=torch.long)
+
+    with torch.inference_mode():
+        logits_resp = model(resp_tensor)["logits"][0]
+
+    unp_logits = logits_resp[:-1]
+    unp_targets = resp_tensor[0, 1:]
+    L_uncond = F.cross_entropy(unp_logits, unp_targets, reduction="mean").item()
+
+    ifd = L_cond / L_uncond if L_uncond > 0 else None
+
+    return {
+        "L_cond": round(L_cond, 6),
+        "L_uncond": round(L_uncond, 6),
+        "ifd": round(ifd, 6) if ifd is not None else None,
+        "instr_len": prefix_len,
+        "resp_len": len(resp_ids),
+        "error": None,
+    }
+
+
+def process_file(
+    param_path: str,
+    input_file: str,
+    output_file: str,
+    instr_key: str,
+    resp_key: str,
+    max_len: int,
+    use_chat_template: bool = False,
+):
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    dtype = torch.bfloat16 if device == "cuda" else torch.float32
+
+    model = AutoModel.from_pretrained(param_path)
+    tokenizer = AutoTokenizer.from_pretrained(param_path)
+    model.to(device=device, dtype=dtype)
+    model.eval()
+
+    if use_chat_template and tokenizer._chat_template is None:
+        raise RuntimeError(
+            "--use_chat_template specified but tokenizer has no chat template. "
+            "Add a chat_template to tokenizer_config.json or omit the flag."
+        )
+
+    with open(input_file, "r", encoding="utf-8") as f:
+        data = [json.loads(line) for line in f if line.strip()]
+
+    results = []
+    ifd_values = []
+
+    with torch.inference_mode():
+        for item in tqdm.tqdm(data, desc="Computing IFD", unit="sample"):
+            instruction = item[instr_key]
+            response = item[resp_key]
+            scores = compute_ifd(
+                model,
+                tokenizer,
+                instruction,
+                response,
+                device,
+                max_len,
+                use_chat_template=use_chat_template,
+            )
+            ifd_values.append(scores["ifd"])
+            results.append({**item, "ifd": scores["ifd"], "ifd_detail": scores})
+
+    with open(output_file, "w", encoding="utf-8") as f:
+        for item in results:
+            f.write(json.dumps(item, ensure_ascii=False) + "\n")
+
+    valid_ifd = [v for v in ifd_values if v is not None]
+    if valid_ifd:
+        import statistics
+
+        print(f"\n{'=' * 50}")
+        print(f"  Samples:    {len(data)}")
+        print(f"  Valid IFD:  {len(valid_ifd)}")
+        print(f"  Mean IFD:   {statistics.mean(valid_ifd):.4f}")
+        print(f"  Median IFD: {statistics.median(valid_ifd):.4f}")
+        print(f"  Stdev IFD:  {statistics.stdev(valid_ifd):.4f}")
+        print(f"  Min IFD:    {min(valid_ifd):.4f}")
+        print(f"  Max IFD:    {max(valid_ifd):.4f}")
+        print(f"{'=' * 50}")
+
+    print(f"Results saved to {output_file}")
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Compute IFD scores for instruction-response data"
+    )
+    parser.add_argument("--param_path", type=str, required=True, help="Model directory")
+    parser.add_argument("--input", type=str, required=True, help="Input JSONL file")
+    parser.add_argument("--output", type=str, required=True, help="Output JSONL file")
+    parser.add_argument(
+        "--instr_key",
+        type=str,
+        default="instruction",
+        help="Key for instruction field",
+    )
+    parser.add_argument(
+        "--resp_key",
+        type=str,
+        default="response",
+        help="Key for response field",
+    )
+    parser.add_argument(
+        "--max_len",
+        type=int,
+        default=2048,
+        help="Max token length (instruction truncated to fit)",
+    )
+    parser.add_argument(
+        "--no_chat_template",
+        action="store_true",
+        default=False,
+        help="Disable chat template, use raw text concatenation",
+    )
+    args = parser.parse_args()
+
+    process_file(
+        args.param_path,
+        args.input,
+        args.output,
+        args.instr_key,
+        args.resp_key,
+        args.max_len,
+        use_chat_template=not args.no_chat_template,
+    )
+
+
+if __name__ == "__main__":
+    main()

scripts/eval/evaluate_ifeval.py

@@ -0,0 +1,609 @@
+"""IFEval instruction-following evaluation benchmark.
+
+Evaluates model responses against regex-based constraint verifiers.
+Supports all IFEval constraint types except language detection.
+
+Usage::
+
+    python scripts/tools/evaluate_ifeval.py --param_path ./params \
+        --data_path ifeval.jsonl --output results.json \
+        --temperature 0.1 --max_tokens 512
+"""
+
+import argparse
+import json
+import os
+import re
+import urllib.request
+from typing import Callable, Dict, List, Optional
+
+import torch
+import tqdm
+
+from astrai.inference import InferenceEngine
+from astrai.model import AutoModel
+from astrai.tokenize import AutoTokenizer
+
+IFEVAL_URL = (
+    "https://raw.githubusercontent.com/google-research/"
+    "google-research/master/instruction_following_eval/data/input_data.jsonl"
+)
+
+CONSTRAINT_VERIFIERS: Dict[str, Callable[[str, dict], bool]] = {}
+
+
+def register(instruction_id: str):
+    def decorator(fn):
+        CONSTRAINT_VERIFIERS[instruction_id] = fn
+        return fn
+
+    return decorator
+
+
+@register("keywords:existence")
+def check_keyword_existence(response: str, kwargs: dict) -> bool:
+    for kw in kwargs["keywords"]:
+        if not re.search(re.escape(kw), response, re.IGNORECASE):
+            return False
+    return True
+
+
+@register("keywords:frequency")
+def check_keyword_frequency(response: str, kwargs: dict) -> bool:
+    keyword = kwargs["keyword"]
+    frequency = kwargs.get("frequency", 1)
+    relation = kwargs.get("relation", "at least")
+    count = len(re.findall(re.escape(keyword), response, re.IGNORECASE))
+    if relation == "less than":
+        return count < frequency
+    return count >= frequency
+
+
+@register("keywords:forbidden_words")
+def check_forbidden_words(response: str, kwargs: dict) -> bool:
+    for word in kwargs["forbidden_words"]:
+        if re.search(r"\b" + re.escape(word) + r"\b", response, re.IGNORECASE):
+            return False
+    return True
+
+
+@register("keywords:letter_frequency")
+def check_letter_frequency(response: str, kwargs: dict) -> bool:
+    letter = kwargs["letter"].lower()
+    frequency = kwargs.get("let_frequency", 1)
+    relation = kwargs.get("let_relation", "at least")
+    count = response.lower().count(letter)
+    if relation == "less than":
+        return count < frequency
+    return count >= frequency
+
+
+@register("detectable_content:number_placeholders")
+def check_placeholders(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_placeholders", 1)
+    placeholders = re.findall(r"\[.*?\]", response)
+    return len(placeholders) >= num
+
+
+@register("detectable_content:postscript")
+def check_postscript(response: str, kwargs: dict) -> bool:
+    marker = kwargs.get("postscript_marker", "P.S.")
+    response_lower = response.lower()
+    if marker == "P.P.S":
+        return bool(re.search(r"p\.\s?p\.\s?s", response_lower))
+    elif marker == "P.S.":
+        return bool(re.search(r"p\.\s?s\.", response_lower))
+    else:
+        return bool(re.search(re.escape(marker.lower()), response_lower))
+
+
+@register("detectable_format:number_bullet_lists")
+def check_bullet_lists(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_bullets", 1)
+    bullets = re.findall(r"^\s*\*[^\*].*$", response, re.MULTILINE)
+    dashes = re.findall(r"^\s*-.*$", response, re.MULTILINE)
+    return len(bullets) + len(dashes) == num
+
+
+@register("detectable_format:number_highlighted_sections")
+def check_highlighted_sections(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_highlights", 1)
+    highlights = re.findall(r"\*[^\n\*]+\*", response)
+    count = 0
+    for h in highlights:
+        if h.strip("*").strip():
+            count += 1
+    return count >= num
+
+
+@register("detectable_format:multiple_sections")
+def check_multiple_sections(response: str, kwargs: dict) -> bool:
+    splitter = kwargs.get("section_spliter", "Section")
+    num = kwargs.get("num_sections", 1)
+    pattern = r"\s?" + re.escape(splitter) + r"\s?\d+\s?"
+    sections = re.split(pattern, response)
+    return len(sections) - 1 >= num
+
+
+@register("detectable_format:title")
+def check_title(response: str, kwargs: dict) -> bool:
+    titles = re.findall(r"<<[^>\n]+>>", response)
+    for title in titles:
+        if title.strip("<>").strip():
+            return True
+    return False
+
+
+@register("detectable_format:json_format")
+def check_json_format(response: str, kwargs: dict) -> bool:
+    value = response.strip()
+    for prefix in ("```json", "```Json", "```JSON", "```"):
+        if value.lower().startswith(prefix.lower()):
+            value = value[len(prefix) :].strip()
+    if value.endswith("```"):
+        value = value[:-3].strip()
+    try:
+        json.loads(value)
+        return True
+    except (ValueError, json.JSONDecodeError):
+        return False
+
+
+@register("detectable_format:general_punctuation")
+def check_general_punctuation(response: str, kwargs: dict) -> bool:
+    punctuation_blacklist = kwargs.get("punctuation_blacklist", [])
+    for punct in punctuation_blacklist:
+        if punct in response:
+            return False
+    return True
+
+
+@register("detectable_format:number_highlighted_words")
+def check_highlighted_words(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_highlights", 1)
+    highlights = re.findall(r"\*[^\s\*][^\*]*[^\s\*]\*", response)
+    return len(highlights) >= num
+
+
+@register("startend:end_checker")
+def check_end_checker(response: str, kwargs: dict) -> bool:
+    end_phrase = kwargs["end_phrase"]
+    return (
+        response.strip()
+        .rstrip('"')
+        .rstrip()
+        .lower()
+        .endswith(end_phrase.strip().lower())
+    )
+
+
+@register("startend:quotation")
+def check_quotation(response: str, kwargs: dict) -> bool:
+    value = response.strip()
+    return value.startswith('"') and value.endswith('"')
+
+
+@register("startend:start_checker")
+def check_start_checker(response: str, kwargs: dict) -> bool:
+    starter = kwargs["starter"]
+    return bool(re.search(r"^\s*" + re.escape(starter), response, re.MULTILINE))
+
+
+@register("change_case:english_capital")
+def check_english_capital(response: str, kwargs: dict) -> bool:
+    return response.isupper()
+
+
+@register("change_case:english_lowercase")
+def check_english_lowercase(response: str, kwargs: dict) -> bool:
+    return response.islower()
+
+
+@register("change_case:capital_word_frequency")
+def check_capital_word_frequency(response: str, kwargs: dict) -> bool:
+    frequency = kwargs.get("capital_frequency", 1)
+    relation = kwargs.get("capital_relation", "at least")
+    capital_words = re.findall(r"\b[A-Z]{2,}\b", response)
+    count = len(capital_words)
+    if relation == "less than":
+        return count < frequency
+    return count >= frequency
+
+
+@register("punctuation:no_comma")
+def check_no_comma(response: str, kwargs: dict) -> bool:
+    return "," not in response
+
+
+def count_words(text: str) -> int:
+    return len(re.findall(r"\b\w+\b", text))
+
+
+def count_sentences(text: str) -> int:
+    text = text.strip()
+    if not text:
+        return 0
+    sentences = re.split(r"(?<=[.!?])\s+", text)
+    return len([s for s in sentences if s.strip()])
+
+
+@register("length_constraints:number_words")
+def check_number_words(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_words", 100)
+    relation = kwargs.get("relation", "at least")
+    cnt = count_words(response)
+    if relation == "less than":
+        return cnt < num
+    return cnt >= num
+
+
+@register("length_constraints:number_sentences")
+def check_number_sentences(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_sentences", 5)
+    relation = kwargs.get("relation", "at least")
+    cnt = count_sentences(response)
+    if relation == "less than":
+        return cnt < num
+    return cnt >= num
+
+
+@register("length_constraints:number_paragraphs")
+def check_number_paragraphs(response: str, kwargs: dict) -> bool:
+    num = kwargs.get("num_paragraphs", 1)
+    if "***" in response:
+        paragraphs = re.split(r"\s?\*\*\*\s?", response)
+    else:
+        paragraphs = re.split(r"\n\n+", response)
+    actual = len([p for p in paragraphs if p.strip()])
+    return actual == num
+
+
+@register("length_constraints:nth_paragraph_first_word")
+def check_nth_paragraph_first_word(response: str, kwargs: dict) -> bool:
+    num_paragraphs = kwargs.get("num_paragraphs", 1)
+    nth = kwargs.get("nth_paragraph", 1)
+    first_word = kwargs.get("first_word", "").lower()
+
+    paragraphs = re.split(r"\n\n+", response)
+    paragraphs = [p.strip() for p in paragraphs if p.strip()]
+
+    if len(paragraphs) != num_paragraphs:
+        return False
+    if nth > len(paragraphs):
+        return False
+
+    target = paragraphs[nth - 1]
+    words = target.split()
+    if not words:
+        return False
+
+    word = words[0].strip().lstrip("'\"").rstrip(".,!?:;\"'")
+    return word.lower() == first_word
+
+
+@register("length_constraints:nth_word_checker")
+def check_nth_word(response: str, kwargs: dict) -> bool:
+    nth = kwargs.get("nth_word", 1)
+    target = kwargs.get("target_word", "").lower()
+    words = re.findall(r"\b\w+\b", response)
+    if nth > len(words):
+        return False
+    return words[nth - 1].lower() == target
+
+
+@register("combination:repeat_prompt")
+def check_repeat_prompt(response: str, kwargs: dict) -> bool:
+    prompt = kwargs["prompt_to_repeat"]
+    return response.strip().lower().startswith(prompt.strip().lower())
+
+
+@register("combination:two_responses")
+def check_two_responses(response: str, kwargs: dict) -> bool:
+    parts = response.split("******")
+    valid = [p for p in parts if p.strip()]
+    if len(valid) != 2:
+        return False
+    return valid[0].strip() != valid[1].strip()
+
+
+def download_ifeval(data_path: str):
+    if os.path.exists(data_path):
+        return
+    os.makedirs(os.path.dirname(data_path) or ".", exist_ok=True)
+    print(f"Downloading IFEval from {IFEVAL_URL} ...")
+    tmp = data_path + ".tmp"
+    urllib.request.urlretrieve(IFEVAL_URL, tmp)
+    with open(tmp, "rb") as f_in:
+        content = f_in.read()
+    with open(data_path, "wb") as f_out:
+        f_out.write(content)
+    os.remove(tmp)
+    print(f"  saved to {data_path}")
+
+
+def load_problems(data_path: str) -> List[dict]:
+    problems = []
+    with open(data_path, "r", encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                problems.append(json.loads(line))
+    return problems
+
+
+def verify_response(response: str, instruction_id: str, kwargs: dict) -> Optional[bool]:
+    verifier = CONSTRAINT_VERIFIERS.get(instruction_id)
+    if verifier is None:
+        return None
+    try:
+        return verifier(response, kwargs)
+    except Exception:
+        return False
+
+
+def generate_one(
+    engine: InferenceEngine,
+    tokenizer: AutoTokenizer,
+    prompt: str,
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int,
+) -> str:
+    formatted = tokenizer.apply_chat_template(
+        [{"role": "user", "content": prompt}],
+        tokenize=False,
+        add_generation_prompt=True,
+    )
+    output = engine.generate(
+        prompt=formatted,
+        stream=False,
+        max_tokens=max_tokens,
+        temperature=temperature,
+        top_p=top_p,
+        top_k=top_k,
+    )
+    if isinstance(output, list):
+        return output[0]
+    return output
+
+
+def evaluate(
+    engine: InferenceEngine,
+    tokenizer: AutoTokenizer,
+    problems: List[dict],
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int,
+    num_samples: int = 1,
+) -> Dict:
+    results = {}
+    constraint_stats: Dict[str, Dict[str, int]] = {}
+    total_constraints = 0
+    total_passed = 0
+
+    for problem in tqdm.tqdm(problems, desc="IFEval", unit="problem"):
+        key = problem["key"]
+        prompt = problem["prompt"]
+        instruction_ids = problem["instruction_id_list"]
+        kwargs_list = problem["kwargs"]
+
+        samples = []
+        for _ in range(num_samples):
+            response = generate_one(
+                engine, tokenizer, prompt, max_tokens, temperature, top_p, top_k
+            )
+            samples.append(response)
+
+        constraint_results = []
+        passed = 0
+        verified = 0
+
+        for idx, instruction_id in enumerate(instruction_ids):
+            kwargs = kwargs_list[idx] if idx < len(kwargs_list) else {}
+            best_pass = False
+            for response in samples:
+                result = verify_response(response, instruction_id, kwargs)
+                if result is None:
+                    continue
+                if result:
+                    best_pass = True
+                    break
+
+            verifier_exists = instruction_id in CONSTRAINT_VERIFIERS
+            if verifier_exists:
+                verified += 1
+                if best_pass:
+                    passed += 1
+
+            constraint_results.append(
+                {
+                    "instruction_id": instruction_id,
+                    "passed": best_pass,
+                    "supported": verifier_exists,
+                    "kwargs": kwargs,
+                }
+            )
+
+            if verifier_exists:
+                if instruction_id not in constraint_stats:
+                    constraint_stats[instruction_id] = {
+                        "total": 0,
+                        "passed": 0,
+                    }
+                constraint_stats[instruction_id]["total"] += 1
+                if best_pass:
+                    constraint_stats[instruction_id]["passed"] += 1
+
+        total_constraints += verified
+        total_passed += passed
+
+        accuracy = passed / verified if verified > 0 else None
+        results[str(key)] = {
+            "key": key,
+            "prompt": prompt,
+            "response": samples[0],
+            "num_samples": num_samples,
+            "num_constraints": len(instruction_ids),
+            "num_verified": verified,
+            "num_passed": passed,
+            "accuracy": round(accuracy, 4) if accuracy is not None else None,
+            "constraints": constraint_results,
+        }
+
+    overall_accuracy = (
+        round(total_passed / total_constraints, 4) if total_constraints > 0 else 0.0
+    )
+
+    type_summary = {}
+    for inst_id, stats in sorted(constraint_stats.items()):
+        type_summary[inst_id] = {
+            "total": stats["total"],
+            "passed": stats["passed"],
+            "accuracy": round(stats["passed"] / stats["total"], 4)
+            if stats["total"] > 0
+            else 0.0,
+        }
+
+    unsupported_count = sum(
+        1
+        for p in problems
+        for iid in p["instruction_id_list"]
+        if iid not in CONSTRAINT_VERIFIERS
+    )
+
+    results["_summary"] = {
+        "total_problems": len(problems),
+        "total_constraints": total_constraints,
+        "total_passed": total_passed,
+        "overall_accuracy": overall_accuracy,
+        "unsupported_constraints": unsupported_count,
+        "supported_types": sorted(CONSTRAINT_VERIFIERS.keys()),
+        "per_type_accuracy": type_summary,
+    }
+
+    return results
+
+
+def main():
+    parser = argparse.ArgumentParser(description="IFEval benchmark")
+    parser.add_argument(
+        "--param_path", type=str, default="./params", help="Model directory"
+    )
+    parser.add_argument(
+        "--data_path",
+        type=str,
+        default="./ifeval/input_data.jsonl",
+        help="IFEval JSONL file (auto-download if missing)",
+    )
+    parser.add_argument("--output", type=str, default=None, help="Output JSON path")
+    parser.add_argument(
+        "--max_tokens", type=int, default=512, help="Max generation tokens"
+    )
+    parser.add_argument(
+        "--temperature",
+        type=float,
+        default=0.1,
+        help="Sampling temperature",
+    )
+    parser.add_argument("--top_p", type=float, default=0.95, help="Top-p sampling")
+    parser.add_argument("--top_k", type=int, default=50, help="Top-k sampling")
+    parser.add_argument(
+        "--num_samples",
+        type=int,
+        default=1,
+        help="Number of samples per problem (best-of-n scoring)",
+    )
+    parser.add_argument(
+        "--batch_size", type=int, default=1, help="Inference batch size"
+    )
+    parser.add_argument(
+        "--limit",
+        type=int,
+        default=None,
+        help="Limit to first N problems (for quick testing)",
+    )
+    parser.add_argument(
+        "--dump_responses",
+        type=str,
+        default=None,
+        help="Path to dump raw model responses (JSONL)",
+    )
+    args = parser.parse_args()
+
+    download_ifeval(args.data_path)
+    problems = load_problems(args.data_path)
+    if args.limit:
+        problems = problems[: args.limit]
+
+    print(f"Loaded {len(problems)} problems")
+    print(f"Supported constraint types: {len(CONSTRAINT_VERIFIERS)}")
+
+    model = AutoModel.from_pretrained(args.param_path)
+    tokenizer = AutoTokenizer.from_pretrained(args.param_path)
+    model.to(device="cuda", dtype=torch.bfloat16)
+    model.eval()
+
+    engine = InferenceEngine(
+        model=model,
+        tokenizer=tokenizer,
+        max_batch_size=args.batch_size,
+    )
+
+    results = evaluate(
+        engine=engine,
+        tokenizer=tokenizer,
+        problems=problems,
+        max_tokens=args.max_tokens,
+        temperature=args.temperature,
+        top_p=args.top_p,
+        top_k=args.top_k,
+        num_samples=args.num_samples,
+    )
+
+    summary = results.pop("_summary")
+    print(f"\n{'=' * 60}")
+    print(f"  Problems:       {summary['total_problems']}")
+    print(f"  Constraints:    {summary['total_constraints']}")
+    print(f"  Passed:         {summary['total_passed']}")
+    print(f"  Accuracy:       {summary['overall_accuracy']:.2%}")
+    print(f"  Unsupported:    {summary['unsupported_constraints']}")
+    print(f"{'=' * 60}")
+
+    print("\nPer-type accuracy:")
+    for inst_id, stats in sorted(summary["per_type_accuracy"].items()):
+        print(
+            f"  {inst_id:50s}  {stats['accuracy']:.2%}  "
+            f"({stats['passed']}/{stats['total']})"
+        )
+
+    if args.output:
+        results["_summary"] = summary
+        with open(args.output, "w", encoding="utf-8") as f:
+            json.dump(results, f, indent=2, ensure_ascii=False)
+        print(f"\nResults saved to {args.output}")
+
+    if args.dump_responses:
+        with open(args.dump_responses, "w", encoding="utf-8") as f:
+            for k, v in results.items():
+                if k.startswith("_"):
+                    continue
+                f.write(
+                    json.dumps(
+                        {
+                            "key": v["key"],
+                            "prompt": v["prompt"],
+                            "response": v["response"],
+                        },
+                        ensure_ascii=False,
+                    )
+                    + "\n"
+                )
+        print(f"Responses dumped to {args.dump_responses}")
+
+    engine.shutdown()
+
+
+if __name__ == "__main__":
+    main()

scripts/eval/evaluate_mmlu.py

@@ -5,9 +5,9 @@ import csv
 import json
 import os
 import shutil
-import urllib.request
+import tarfile
-import zipfile
 
+import requests
 import torch
 import torch.nn.functional as F
 import tqdm
@@ -15,7 +15,7 @@ import tqdm
 from astrai.model import AutoModel
 from astrai.tokenize import AutoTokenizer
 
-MMLU_URL = "https://github.com/hendrycks/test/archive/refs/heads/master.zip"
+MMLU_URL = "https://people.eecs.berkeley.edu/~hendrycks/data.tar"
 MMLU_SUBJECTS = [
     "abstract_algebra",
     "anatomy",
@@ -78,23 +78,37 @@ MMLU_SUBJECTS = [
 
 
 def _download_and_extract(url: str, data_dir: str):
-    zip_path = os.path.join(data_dir, "mmlu.zip")
+    tar_path = os.path.join(data_dir, "data.tar")
     os.makedirs(data_dir, exist_ok=True)
     print(f"Downloading MMLU data from {url}...")
-    urllib.request.urlretrieve(url, zip_path)
+    resp = requests.get(url, stream=True, timeout=300)
+    resp.raise_for_status()
+    total = int(resp.headers.get("content-length", 0))
+    with tqdm.tqdm(total=total, unit="B", unit_scale=True, desc="  Download") as bar:
+        with open(tar_path, "wb") as f:
+            for chunk in resp.iter_content(chunk_size=8192):
+                f.write(chunk)
+                bar.update(len(chunk))
     print("Extracting...")
-    with zipfile.ZipFile(zip_path, "r") as zf:
+    with tarfile.open(tar_path, "r") as tf:
-        zf.extractall(data_dir)
+        tf.extractall(data_dir)
-    os.remove(zip_path)
+    os.remove(tar_path)
 
 
 def download_mmlu(data_dir: str):
     _download_and_extract(MMLU_URL, data_dir)
-    src = os.path.join(data_dir, "test-master", "data")
+    src = os.path.join(data_dir, "data")
     if os.path.exists(src):
         for item in os.listdir(src):
-            os.rename(os.path.join(src, item), os.path.join(data_dir, item))
+            src_item = os.path.join(src, item)
-        shutil.rmtree(os.path.join(data_dir, "test-master"))
+            dst_item = os.path.join(data_dir, item)
+            if os.path.exists(dst_item):
+                if os.path.isdir(dst_item):
+                    shutil.rmtree(dst_item)
+                else:
+                    os.remove(dst_item)
+            os.rename(src_item, dst_item)
+        os.rmdir(src)
     print(f"MMLU data saved to {data_dir}")
 
 
@@ -143,10 +157,32 @@ def build_prompt(
     return prompt
 
 
+def apply_chat(
+    tokenizer, raw_prompt: str, n_shot: int, dev_data: list[dict] | None
+) -> str:
+    """Wrap raw MMLU prompt in the model's chat template format.
+
+    For few-shot, prepend example Q&A pairs as a second user/assistant exchange.
+    """
+    messages = []
+    if n_shot > 0 and dev_data:
+        for item in dev_data[:n_shot]:
+            q = f"Question: {item['question']}\n"
+            for k in ("A", "B", "C", "D"):
+                q += f"{k}. {item[k]}\n"
+            q += "Answer:"
+            messages.append({"role": "user", "content": q})
+            messages.append({"role": "assistant", "content": item["answer"]})
+    messages.append({"role": "user", "content": raw_prompt})
+    return tokenizer.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+
+
 def choice_logprob(
     model, tokenizer, context_ids: list[int], choice_letter: str, device: str
 ) -> float:
-    choice_text = f" {choice_letter}"
+    choice_text = choice_letter
     choice_ids = tokenizer.encode(choice_text, add_special_tokens=False)
     input_ids = context_ids + choice_ids
     max_len = model.config.max_len
@@ -182,8 +218,11 @@ def evaluate_subject(
     correct = 0
     total = 0
     for item in tqdm.tqdm(test_data, desc=f"{subject:40s}", leave=False):
-        prompt = build_prompt(item["question"], item, subject, n_shot, dev_data or [])
+        raw_prompt = build_prompt(
-        context_ids = tokenizer.encode(prompt)
+            item["question"], item, subject, n_shot, dev_data or []
+        )
+        context = apply_chat(tokenizer, raw_prompt, n_shot, dev_data or [])
+        context_ids = tokenizer.encode(context)
         scores = {
             c: choice_logprob(model, tokenizer, context_ids, c, device)
             for c in ("A", "B", "C", "D")
@@ -233,6 +272,7 @@ def main():
     device = args.device
     dtype = getattr(torch, args.dtype)
     model.to(device=device, dtype=dtype)
+    model.eval()
 
     subjects = args.subjects or MMLU_SUBJECTS
     results = {}

scripts/eval/evaluate_ppl.py

@@ -86,7 +86,7 @@ def process_file(
 
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Run perplexity with a Khaosz model.")
+    parser = argparse.ArgumentParser(description="Perplexity evaluation on JSONL text.")
     parser.add_argument(
         "--param_path", type=str, required=True, help="Path to the model directory."
     )

scripts/tools/generate.py

@@ -1,5 +1,6 @@
 import argparse
 import json
+from typing import Optional
 
 import torch
 
@@ -17,7 +18,7 @@ def processor(
     top_p: float,
     question_key: str,
     response_key: str,
-    max_tokens: int,
+    max_tokens: Optional[int],
     batch_size: int,
 ):
     # Load model and tokenizer
@@ -72,7 +73,7 @@ def processor(
 
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Run generate with a Khaosz model.")
+    parser = argparse.ArgumentParser(description="Batch generation from JSONL file.")
 
     parser.add_argument(
         "--param_path", type=str, required=True, help="Path to the model directory."
@@ -93,36 +94,42 @@ if __name__ == "__main__":
         "--question_key",
         type=str,
         default="question",
-        help="Key for the question in the input JSON.",
+        help="Key for the question in the input JSON (default: question).",
     )
     parser.add_argument(
         "--response_key",
         type=str,
         default="response",
-        help="Key for the response in the output JSON.",
+        help="Key for the response in the output JSON (default: response).",
     )
     parser.add_argument(
         "--temperature",
         type=float,
         default=0.60,
-        help="Temperature for generating responses.",
+        help="Temperature for generating responses (default: 0.60).",
     )
     parser.add_argument(
-        "--top_k", type=int, default=30, help="Top-k value for generating responses."
+        "--top_k",
+        type=int,
+        default=30,
+        help="Top-k value for generating responses (default: 30).",
     )
     parser.add_argument(
         "--top_p",
         type=float,
         default=0.95,
-        help="Top-p value for generating responses.",
+        help="Top-p value for generating responses (default: 0.95).",
     )
     parser.add_argument(
-        "--batch_size", type=int, default=1, help="Batch size for generating responses."
+        "--batch_size",
+        type=int,
+        default=1,
+        help="Batch size for generating responses (default: 1).",
     )
     parser.add_argument(
         "--max_tokens",
         type=int,
-        default=2048,
+        default=None,
         help="Maximum tokens to generate (default: model config max_len).",
     )
 

scripts/tools/preprocess.py

@@ -0,0 +1,38 @@
+"""CLI: JSONL → tokenized .h5/.bin via config-driven Pipeline."""
+
+import argparse
+
+from astrai.config.preprocess_config import PipelineConfig
+from astrai.preprocessing.pipeline import Pipeline
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Raw JSONL → tokenized .h5/.bin via config-driven Pipeline"
+    )
+    parser.add_argument(
+        "inputs", nargs="+", metavar="JSONL", help="One or more JSONL files"
+    )
+    parser.add_argument("--output_dir", "-o", required=True, help="Output directory")
+    parser.add_argument(
+        "--config", "-c", required=True, help="Path to pipeline config JSON"
+    )
+    parser.add_argument(
+        "--tokenizer_path",
+        default="params",
+        help="Path to tokenizer directory (default: params)",
+    )
+    args = parser.parse_args()
+
+    config = PipelineConfig.from_file(args.config)
+
+    Pipeline(
+        config=config,
+        input_paths=args.inputs,
+        output_dir=args.output_dir,
+        tokenizer_path=args.tokenizer_path,
+    ).run()
+
+
+if __name__ == "__main__":
+    main()

scripts/tools/train.py

@@ -8,6 +8,7 @@ import torch.optim as optim
 from astrai.config import AutoRegressiveLMConfig, TrainConfig
 from astrai.dataset import DatasetFactory
 from astrai.model import AutoRegressiveLM
+from astrai.model.components.decoder_block import DecoderBlock
 from astrai.trainer import SchedulerFactory, Trainer
 
 
@@ -112,9 +113,15 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "--label_smoothing",
         type=float,
-        default=0.05,
+        default=0.0,
         help="cross_entropy function label smoothing parameter",
     )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action=argparse.BooleanOptionalAction,
+        default=False,
+        help="Enable activation checkpointing for DecoderBlock modules.",
+    )
 
     parser.add_argument(
         "--ckpt_interval",
@@ -128,6 +135,36 @@ def parse_args() -> argparse.Namespace:
         default="checkpoint",
         help="Directory to save checkpoints.",
     )
+    parser.add_argument(
+        "--val_split",
+        type=float,
+        default=None,
+        help="Ratio to split from training dataset for validation (e.g. 0.05).",
+    )
+    parser.add_argument(
+        "--val_step",
+        type=int,
+        default=1000,
+        help="Number of optimizer steps between validation runs.",
+    )
+    parser.add_argument(
+        "--metrics",
+        nargs="*",
+        default=["loss", "lr"],
+        help="Metrics to log (e.g. --metrics loss lr val_loss). Default: loss lr.",
+    )
+    parser.add_argument(
+        "--log_dir",
+        type=str,
+        default="checkpoint/logs",
+        help="Directory for metric logs.",
+    )
+    parser.add_argument(
+        "--log_interval",
+        type=int,
+        default=100,
+        help="Number of batch iterations between metric logs.",
+    )
     parser.add_argument(
         "--grpo_sync_interval",
         type=int,
@@ -141,6 +178,24 @@ def parse_args() -> argparse.Namespace:
         "--start_batch", type=int, default=0, help="Start batch for training."
     )
 
+    parser.add_argument(
+        "--master_addr",
+        type=str,
+        default="localhost",
+        help="Master node address for distributed training.",
+    )
+    parser.add_argument(
+        "--master_port",
+        type=str,
+        default="29500",
+        help="Master node port for distributed training.",
+    )
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="nccl",
+        help="Distributed training backend.",
+    )
     parser.add_argument("--nprocs", type=int, default=1, help="Number of GPUs to use.")
     parser.add_argument(
         "--parallel_mode",
@@ -159,6 +214,50 @@ def parse_args() -> argparse.Namespace:
         choices=["spawn", "fork", "forkserver"],
         help="Multiprocessing start method.",
     )
+    parser.add_argument(
+        "--neftune_alpha",
+        type=float,
+        default=0.0,
+        help="NEFTune noise alpha (0=disabled, typical: 5.0).",
+    )
+
+    parser.add_argument(
+        "--schedule_type",
+        type=str,
+        default="cosine",
+        choices=["cosine", "sgdr", "wsd"],
+        help="Learning rate scheduler type.",
+    )
+    parser.add_argument(
+        "--min_rate",
+        type=float,
+        default=None,
+        help="Minimum LR as fraction of base LR. Uses scheduler default if not set (cosine/sgdr: 0.05, wsd: 0.0).",
+    )
+    parser.add_argument(
+        "--cycle_length",
+        type=int,
+        default=None,
+        help="SGDR first cycle length in steps. Defaults to total_steps - warmup_steps.",
+    )
+    parser.add_argument(
+        "--t_mult",
+        type=int,
+        default=2,
+        help="SGDR cycle length multiplier per restart.",
+    )
+    parser.add_argument(
+        "--stable_steps",
+        type=int,
+        default=None,
+        help="WSD stable plateau steps. Required when --schedule_type wsd.",
+    )
+    parser.add_argument(
+        "--decay_steps",
+        type=int,
+        default=None,
+        help="WSD decay steps. Defaults to total_steps - warmup_steps - stable_steps.",
+    )
 
     args = parser.parse_args()
 
@@ -176,7 +275,8 @@ def create_optimizer(model, **kwargs) -> optim.Optimizer:
 def create_scheduler(
     optimizer: optim.Optimizer, **kwargs
 ) -> optim.lr_scheduler.LRScheduler:
-    return SchedulerFactory.create(optimizer, **kwargs)
+    schedule_type = kwargs.pop("schedule_type")
+    return SchedulerFactory.create(schedule_type, optimizer, **kwargs)
 
 
 def compute_total_steps(
@@ -209,6 +309,11 @@ def train(
     warmup_ratio: float,
     ckpt_interval: int,
     ckpt_dir: str,
+    val_split: float,
+    val_step: int,
+    metrics: list[str],
+    log_dir: str,
+    log_interval: int,
     dpo_beta: float,
     grpo_clip_eps: float,
     grpo_kl_coef: float,
@@ -222,12 +327,23 @@ def train(
     random_seed: int,
     num_workers: int,
     pin_memory: bool,
+    gradient_checkpointing: bool,
     window_size: int,
     stride: int,
     nprocs: int,
     parallel_mode: str,
     device_type: str,
+    backend: str,
+    master_addr: str,
+    master_port: str,
     start_method: str,
+    neftune_alpha: float,
+    schedule_type: str,
+    min_rate: float,
+    cycle_length: int,
+    t_mult: int,
+    stable_steps: int,
+    decay_steps: int,
 ):
     assert train_type in ["seq", "sft", "dpo", "grpo"]
     assert os.path.exists(param_path)
@@ -237,6 +353,7 @@ def train(
     # Load config
     config_path = os.path.join(param_path, "config.json")
     config = AutoRegressiveLMConfig.from_file(config_path)
+    config.neftune_alpha = neftune_alpha
 
     if window_size is None:
         window_size = config.max_len
@@ -276,16 +393,34 @@ def train(
         len(dataset), n_epoch, batch_per_device, nprocs, grad_accum_steps
     )
     warmup_steps = int(warmup_ratio * total_steps)
+    warmup_steps = min(warmup_steps, total_steps)
+
+    scheduler_kwargs = {"warmup_steps": warmup_steps}
+
+    if schedule_type == "cosine":
+        scheduler_kwargs["lr_decay_steps"] = total_steps - warmup_steps
+    elif schedule_type == "sgdr":
+        scheduler_kwargs["cycle_length"] = cycle_length or (total_steps - warmup_steps)
+        scheduler_kwargs["t_mult"] = t_mult
+    elif schedule_type == "wsd":
+        remaining = total_steps - warmup_steps
+        stable_steps_ = stable_steps or max(1, int(remaining * 0.8))
+        scheduler_kwargs["stable_steps"] = stable_steps_
+        scheduler_kwargs["decay_steps"] = max(
+            1, decay_steps or (remaining - stable_steps_)
+        )
+
+    if min_rate is not None:
+        scheduler_kwargs["min_rate"] = min_rate
 
     scheduler_fn = partial(
         create_scheduler,
-        **{
+        schedule_type=schedule_type,
-            "schedule_type": "cosine",
+        **scheduler_kwargs,
-            "warmup_steps": min(warmup_steps, total_steps),
-            "lr_decay_steps": total_steps - min(warmup_steps, total_steps),
-        },
     )
 
+    grad_ckpt_modules = [DecoderBlock] if gradient_checkpointing else []
+
     train_config = TrainConfig(
         model_fn=model_fn,
         strategy=train_type,
@@ -304,11 +439,21 @@ def train(
         num_workers=num_workers,
         pin_memory=pin_memory,
         nprocs=nprocs,
+        backend=backend,
+        master_addr=master_addr,
+        master_port=master_port,
         parallel_mode=parallel_mode,
         device_type=device_type,
         start_method=start_method,
+        val_split=val_split,
+        val_step=val_step,
+        metrics=metrics,
+        log_dir=log_dir,
+        log_interval=log_interval,
+        gradient_checkpointing_modules=grad_ckpt_modules,
         executor_kwargs=executor_kwargs,
         extra_kwargs=strategy_kwargs,
+        neftune_alpha=neftune_alpha,
     )
 
     trainer = Trainer(train_config)

tests/data/conftest.py

@@ -0,0 +1,235 @@
+import json
+import os
+import tempfile
+
+import pytest
+from tokenizers import Tokenizer, models, pre_tokenizers, trainers
+
+from astrai.config.preprocess_config import (
+    InputConfig,
+    PipelineConfig,
+    ProcessingConfig,
+)
+from astrai.preprocessing.builder import SectionedMaskBuilder
+from astrai.tokenize import AutoTokenizer
+
+_SPECIAL_TOKENS_CONFIG = {
+    "bos_token": "<|begin_of_sentence|>",
+    "eos_token": "<|end_of_sentence|>",
+    "pad_token": "<|_pad_|>",
+    "unk_token": "<|_unk_|>",
+    "im_start": "<|im_start|>",
+    "im_end": "<|im_end|>",
+}
+
+_SPECIAL_TOKENS = list(_SPECIAL_TOKENS_CONFIG.values())
+
+_CHAT_TEMPLATE = (
+    "{% for message in messages %}"
+    "{% if message['role'] == 'system' %}"
+    "<|im_start|>system\n{{ message['content'] }}<|im_end|>\n"
+    "{% elif message['role'] == 'user' %}"
+    "<|im_start|>user\n{{ message['content'] }}<|im_end|>\n"
+    "{% elif message['role'] == 'assistant' %}"
+    "<|im_start|>assistant\n{{ message['content'] }}<|im_end|>\n"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
+)
+
+_CHAT_SECTIONS = [{"field": "messages", "action": "$role", "template": True}]
+
+_INSTRUCTION_SECTIONS = [
+    {"field": "prompt", "action": "mask", "add_special_tokens": True},
+    {"field": "response", "action": "train"},
+]
+
+_TEXT_SECTIONS = [{"field": "text", "action": "train"}]
+
+_GRPO_RESPONSE_SECTIONS = [{"field": "responses", "action": "train"}]
+
+
+def _build_chat_tokenizer():
+    tok = Tokenizer(models.BPE())
+    tok.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=False)
+    tr = trainers.BpeTrainer(
+        vocab_size=512,
+        min_frequency=1,
+        special_tokens=_SPECIAL_TOKENS,
+    )
+    train_data = [
+        "hello world",
+        "Hi there!",
+        "You are helpful.",
+        "What is 2+2?",
+        "Tell me a story about dragons and knights.",
+        "Sure, here is a tale.",
+        "Translate to French: Hello",
+        "Bonjour",
+        "Artificial Intelligence is a field of computer science.",
+        "system",
+        "user",
+        "assistant",
+        "<|im_start|>",
+        "<|im_end|>",
+        *[chr(i) for i in range(32, 127)],
+    ]
+    tok.train_from_iterator(train_data, tr)
+
+    auto_tok = AutoTokenizer()
+    auto_tok._tokenizer = tok
+    auto_tok._special_token_map = {
+        "bos_token": "<|begin_of_sentence|>",
+        "eos_token": "<|end_of_sentence|>",
+        "pad_token": "<|_pad_|>",
+        "unk_token": "<|_unk_|>",
+    }
+    auto_tok.set_chat_template(_CHAT_TEMPLATE)
+    return auto_tok
+
+
+@pytest.fixture(scope="session")
+def chat_tokenizer():
+    return _build_chat_tokenizer()
+
+
+@pytest.fixture
+def temp_dir():
+    d = tempfile.mkdtemp()
+    yield d
+    import shutil
+
+    shutil.rmtree(d, ignore_errors=True)
+
+
+def make_chat_config():
+    return PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask={"system": "mask", "user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+
+
+def make_instruction_config():
+    return PipelineConfig(
+        input=InputConfig(sections=_INSTRUCTION_SECTIONS),
+        mask={"prompt": "mask", "response": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+
+
+def make_text_config():
+    return PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(
+            max_seq_len=2048, min_chars=1, max_chars=2_000_000
+        ),
+    )
+
+
+def make_dpo_chat_config():
+    return PipelineConfig(
+        input=InputConfig(
+            sources={
+                "chosen": {
+                    "sections": [
+                        {"field": "chosen", "action": "$role", "template": True}
+                    ]
+                },
+                "rejected": {
+                    "sections": [
+                        {"field": "rejected", "action": "$role", "template": True}
+                    ]
+                },
+            }
+        ),
+        mask={"user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+
+
+def make_grpo_config():
+    return PipelineConfig(
+        input=InputConfig(
+            sources={
+                "prompts": {
+                    "sections": [
+                        {"field": "prompt", "action": "mask", "template": True}
+                    ]
+                },
+                "responses": {
+                    "sections": _GRPO_RESPONSE_SECTIONS,
+                    "list_field": True,
+                    "mask_key": "masks",
+                },
+                "rewards": {
+                    "sections": [{"field": "rewards", "action": "value"}],
+                },
+            }
+        ),
+        mask={"user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+
+
+def make_grpo_no_template_config():
+    return PipelineConfig(
+        input=InputConfig(
+            sources={
+                "prompts": {
+                    "sections": [
+                        {
+                            "field": "prompt",
+                            "action": "mask",
+                            "add_special_tokens": True,
+                        }
+                    ]
+                },
+                "responses": {
+                    "sections": _GRPO_RESPONSE_SECTIONS,
+                    "list_field": True,
+                    "mask_key": "masks",
+                },
+                "rewards": {
+                    "sections": [{"field": "rewards", "action": "value"}],
+                },
+            }
+        ),
+        mask={"user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+
+
+@pytest.fixture
+def builder():
+    return SectionedMaskBuilder()
+
+
+@pytest.fixture
+def tokenizer_dir(temp_dir, test_tokenizer):
+    d = os.path.join(temp_dir, "tok")
+    os.makedirs(d, exist_ok=True)
+    test_tokenizer._tokenizer.save(os.path.join(d, "tokenizer.json"))
+    with open(os.path.join(d, "tokenizer_config.json"), "w") as f:
+        json.dump(
+            {"special_tokens": {"pad_token": "<|_pad_|>", "unk_token": "<|_unk_|>"}}, f
+        )
+    return d
+
+
+@pytest.fixture
+def chat_tokenizer_dir(temp_dir, chat_tokenizer):
+    d = os.path.join(temp_dir, "tok")
+    os.makedirs(d, exist_ok=True)
+    chat_tokenizer._tokenizer.save(os.path.join(d, "tokenizer.json"))
+    with open(os.path.join(d, "tokenizer_config.json"), "w") as f:
+        json.dump(
+            {"special_tokens": _SPECIAL_TOKENS_CONFIG, "chat_template": _CHAT_TEMPLATE},
+            f,
+        )
+    return d

tests/data/test_dataset.py

@@ -1,4 +1,3 @@
-import json
 import os
 
 import numpy as np
@@ -8,7 +7,6 @@ import torch
 from astrai.dataset.dataset import DatasetFactory, SEQDataset
 from astrai.dataset.storage import (
     H5Store,
-    MmapStore,
     StoreFactory,
     detect_format,
     load_bin,
@@ -17,28 +15,34 @@ from astrai.dataset.storage import (
 )
 
 
+def _rand_seq(length, vocab=1000):
+    return torch.randint(0, vocab, (length,), dtype=torch.int64)
+
+
+def _make_seq_dataset(
+    test_dir, name="data", seq_length=200, train_type="seq", data=None, **load_kwargs
+):
+    if data is None:
+        data = {"sequence": [_rand_seq(seq_length)]}
+    save_h5(test_dir, name, data)
+    return DatasetFactory.load(
+        train_type,
+        test_dir,
+        window_size=load_kwargs.pop("window_size", 64),
+        **load_kwargs,
+    )
+
+
 def test_dataset_loader_random_paths(base_test_env):
     """Test dataset loader with multiple random paths"""
     test_dir = base_test_env["test_dir"]
 
-    # Create multiple mmap dataset directories with random data
     num_files = np.random.randint(2, 5)
-
     for i in range(num_files):
         seq_length = np.random.randint(200, 400)
-        dummy_data = {
+        dummy_data = {"sequence": [_rand_seq(seq_length) for _ in range(10)]}
-            "sequence": [
+        loaded_dataset = _make_seq_dataset(
-                torch.randint(0, 1000, (seq_length,), dtype=torch.int64)
+            test_dir, f"data_{i}", seq_length, data=dummy_data
-                for _ in range(10)
-            ],
-        }
-        save_h5(test_dir, f"data_{i}", dummy_data)
-
-        # Test loading with multiple paths
-        loaded_dataset = DatasetFactory.load(
-            train_type="seq",
-            load_path=test_dir,
-            window_size=64,
         )
         assert loaded_dataset is not None
         assert len(loaded_dataset) > 0
@@ -56,23 +60,15 @@ def test_dpo_strategy_with_random_data(base_test_env):
     """Test DPO strategy with randomized preference data"""
     test_dir = base_test_env["test_dir"]
 
-    # Create DPO-style data with memory mapping format
     seq_length = np.random.randint(100, 200)
-
     dummy_data = {
-        "chosen": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)],
+        "chosen": [_rand_seq(seq_length)],
-        "rejected": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)],
+        "rejected": [_rand_seq(seq_length)],
         "chosen_mask": [torch.ones(seq_length, dtype=torch.bool)],
         "rejected_mask": [torch.ones(seq_length, dtype=torch.bool)],
     }
-
+    dpo_dataset = _make_seq_dataset(
-    save_h5(test_dir, "dpo_data", dummy_data)
+        test_dir, "dpo_data", seq_length, train_type="dpo", data=dummy_data
-
-    # Load DPO dataset
-    dpo_dataset = DatasetFactory.load(
-        train_type="dpo",
-        load_path=test_dir,
-        window_size=64,
     )
 
     assert dpo_dataset is not None
@@ -94,21 +90,14 @@ def test_sft_dataset_with_random_data(base_test_env):
     """Test SFT dataset with random data"""
     test_dir = base_test_env["test_dir"]
 
-    # Create SFT-style data with memory mapping format
     seq_length = np.random.randint(100, 200)
-
     dummy_data = {
-        "sequence": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)],
+        "sequence": [_rand_seq(seq_length)],
         "loss_mask": [torch.ones(seq_length, dtype=torch.bool)],
+        "position_ids": [torch.arange(seq_length, dtype=torch.int32)],
     }
-
+    sft_dataset = _make_seq_dataset(
-    save_h5(test_dir, "sft_data", dummy_data)
+        test_dir, "sft_data", seq_length, train_type="sft", data=dummy_data
-
-    # Load SFT dataset
-    sft_dataset = DatasetFactory.load(
-        train_type="sft",
-        load_path=test_dir,
-        window_size=64,
     )
 
     assert sft_dataset is not None
@@ -129,25 +118,11 @@ def test_dataset_with_custom_stride(base_test_env):
     """Test dataset with custom stride parameter"""
     test_dir = base_test_env["test_dir"]
 
-    # Create test data
-    seq_length = 200
-    dummy_data = {
-        "sequence": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)],
-    }
-
-    save_h5(test_dir, "stride_test_data", dummy_data)
-
-    # Test with custom stride
     custom_stride = 32
-    dataset = DatasetFactory.load(
+    dataset = _make_seq_dataset(test_dir, "stride_test_data", stride=custom_stride)
-        train_type="seq", load_path=test_dir, window_size=64, stride=custom_stride
-    )
-
     assert dataset is not None
     assert len(dataset) > 0
 
-    # With stride 32 and window 64 on 200 length data, we should get more samples
-    # than with default stride (which equals window size)
     default_stride_dataset = DatasetFactory.load(
         train_type="seq",
         load_path=test_dir,
@@ -158,25 +133,11 @@ def test_dataset_with_custom_stride(base_test_env):
 
 
 def test_dataset_count_property(base_test_env):
-    """Test the count property returns correct raw token count"""
     test_dir = base_test_env["test_dir"]
-
+    dataset = _make_seq_dataset(test_dir, "count_test_data")
-    seq_length = 200
+    assert dataset.count == 200
-    dummy_data = {
+    assert dataset.count > len(dataset)
-        "sequence": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)],
+    assert len(dataset) == (200 - 1 - 64) // 64 + 1
-    }
-
-    save_h5(test_dir, "count_test_data", dummy_data)
-
-    dataset = DatasetFactory.load(
-        train_type="seq",
-        load_path=test_dir,
-        window_size=64,
-    )
-
-    assert dataset.count == seq_length
-    assert dataset.count > len(dataset)  # raw tokens > windows
-    assert len(dataset) == (seq_length - 1 - 64) // 64 + 1
 
 
 def test_empty_dataset_count():
@@ -187,17 +148,10 @@ def test_empty_dataset_count():
 
 
 def test_dataset_too_short_for_window(base_test_env):
-    """Dataset shorter than window_size returns __len__ == 0"""
     test_dir = base_test_env["test_dir"]
-    seq_length = 30
+    dataset = _make_seq_dataset(test_dir, "short", seq_length=30)
-    save_h5(
-        test_dir,
-        "short",
-        {"sequence": [torch.randint(0, 1000, (seq_length,), dtype=torch.int64)]},
-    )
-    dataset = DatasetFactory.load("seq", test_dir, window_size=64)
     assert len(dataset) == 0
-    assert dataset.count == seq_length
+    assert dataset.count == 30
 
 
 def test_unloaded_dataset_getitem_raises():
@@ -221,12 +175,8 @@ def test_store_unloaded_len():
 
 
 def test_store_fetch_begin_equals_end(base_test_env):
-    """Store.fetch with begin == end returns empty tensor"""
     test_dir = base_test_env["test_dir"]
-    dummy = {"sequence": [torch.randint(0, 1000, (100,), dtype=torch.int64)]}
+    dataset = _make_seq_dataset(test_dir, "empty_fetch", seq_length=100, window_size=32)
-    save_h5(test_dir, "empty_fetch", dummy)
-
-    dataset = DatasetFactory.load("seq", test_dir, window_size=32)
     result = dataset.storage.fetch(10, 10, "sequence")
     assert result.numel() == 0
 
@@ -300,12 +250,8 @@ def test_save_load_bin_roundtrip(base_test_env):
 
 
 def test_mmap_store_load_and_fetch(base_test_env):
-    """MmapStore loads bin data and fetches correctly"""
     test_dir = base_test_env["test_dir"]
-
+    data = {"sequence": [_rand_seq(200)]}
-    data = {
-        "sequence": [torch.randint(0, 1000, (200,), dtype=torch.int64)],
-    }
     save_bin(test_dir, data)
 
     store = StoreFactory.create("bin")
@@ -318,14 +264,9 @@ def test_mmap_store_load_and_fetch(base_test_env):
 
 
 def test_mmap_dataset_load(base_test_env):
-    """DatasetFactory.load auto-detects bin format"""
     test_dir = base_test_env["test_dir"]
-
+    data = {"sequence": [_rand_seq(200)]}
-    data = {
-        "sequence": [torch.randint(0, 1000, (200,), dtype=torch.int64)],
-    }
     save_bin(test_dir, data)
-
     dataset = DatasetFactory.load("seq", test_dir, window_size=64)
     assert len(dataset) > 0
     assert dataset.count == 200
@@ -349,19 +290,16 @@ def test_normalize_mixed_empty_key():
 
 
 def test_grpo_dataset_dtype(base_test_env):
-    """GRPODataset returns correct dtypes"""
     test_dir = base_test_env["test_dir"]
-
+    dummy_data = {
-    seq_len = 100
+        "prompts": [torch.randint(0, 100, (100,), dtype=torch.int32)],
-    data = {
+        "responses": [torch.randint(0, 100, (100,), dtype=torch.int32)],
-        "prompts": [torch.randint(0, 100, (seq_len,), dtype=torch.int32)],
+        "masks": [torch.ones(100, dtype=torch.int32)],
-        "responses": [torch.randint(0, 100, (seq_len,), dtype=torch.int32)],
+        "rewards": [torch.ones(100, dtype=torch.float32)],
-        "masks": [torch.ones(seq_len, dtype=torch.int32)],
-        "rewards": [torch.ones(seq_len, dtype=torch.float32)],
     }
-    save_h5(test_dir, "grpo_dtype", data)
+    dataset = _make_seq_dataset(
-
+        test_dir, "grpo_dtype", train_type="grpo", data=dummy_data, window_size=32
-    dataset = DatasetFactory.load("grpo", test_dir, window_size=32)
+    )
     item = dataset[0]
 
     assert item["prompts"].dtype == torch.long
@@ -371,18 +309,16 @@ def test_grpo_dataset_dtype(base_test_env):
 
 
 def test_grpo_dataset_load(base_test_env):
-    """GRPODataset loads and returns correct keys"""
     test_dir = base_test_env["test_dir"]
-    seq_len = 200
+    dummy_data = {
-    data = {
+        "prompts": [_rand_seq(200)],
-        "prompts": [torch.randint(0, 1000, (seq_len,), dtype=torch.int64)],
+        "responses": [_rand_seq(200)],
-        "responses": [torch.randint(0, 1000, (seq_len,), dtype=torch.int64)],
+        "masks": [torch.ones(200, dtype=torch.int64)],
-        "masks": [torch.ones(seq_len, dtype=torch.int64)],
+        "rewards": [torch.rand(200, dtype=torch.float32)],
-        "rewards": [torch.rand(seq_len, dtype=torch.float32)],
     }
-    save_h5(test_dir, "grpo_test", data)
+    dataset = _make_seq_dataset(
-
+        test_dir, "grpo_test", train_type="grpo", data=dummy_data
-    dataset = DatasetFactory.load("grpo", test_dir, window_size=64)
+    )
     assert len(dataset) > 0
     item = dataset[0]
     assert "prompts" in item
@@ -401,7 +337,6 @@ def test_detect_format_bin_dir(base_test_env):
 
 
 def test_store_fetch_multi_key(base_test_env):
-    """Store.fetch with List[str] returns Dict[str, Tensor]"""
     test_dir = base_test_env["test_dir"]
     save_h5(
         test_dir,
@@ -411,7 +346,6 @@ def test_store_fetch_multi_key(base_test_env):
             "loss_mask": [torch.ones(100, dtype=torch.int64)],
         },
     )
-
     store = StoreFactory.create("h5")
     store.load(test_dir)
     result = store.fetch(10, 20, ["sequence", "loss_mask"])
@@ -421,10 +355,8 @@ def test_store_fetch_multi_key(base_test_env):
 
 
 def test_store_fetch_out_of_bounds(base_test_env):
-    """Store.fetch raises ValueError for out-of-bounds indices"""
     test_dir = base_test_env["test_dir"]
     save_h5(test_dir, "bounds", {"sequence": [torch.randint(0, 100, (50,))]})
-
     store = StoreFactory.create("h5")
     store.load(test_dir)
     with pytest.raises(ValueError, match="out of bounds"):
@@ -436,10 +368,7 @@ def test_store_fetch_out_of_bounds(base_test_env):
 
 
 def test_dataset_load_explicit_storage_type(base_test_env):
-    """DatasetFactory.load with explicit storage_type bypasses auto-detect"""
     test_dir = base_test_env["test_dir"]
-    save_h5(test_dir, "explicit", {"sequence": [torch.randint(0, 100, (200,))]})
+    dataset = _make_seq_dataset(test_dir, "explicit", storage_type="h5")
-
-    dataset = DatasetFactory.load("seq", test_dir, window_size=64, storage_type="h5")
     assert len(dataset) > 0
     assert dataset.count == 200

tests/data/test_preprocess_builder.py

@@ -0,0 +1,369 @@
+import pytest
+
+from astrai.config.preprocess_config import (
+    InputConfig,
+    OutputConfig,
+    PipelineConfig,
+    ProcessingConfig,
+)
+from astrai.preprocessing.builder import (
+    MaskBuilderFactory,
+    SectionedMaskBuilder,
+)
+from tests.data.conftest import (
+    _CHAT_SECTIONS,
+    _INSTRUCTION_SECTIONS,
+    _TEXT_SECTIONS,
+    make_chat_config,
+    make_dpo_chat_config,
+    make_grpo_config,
+    make_instruction_config,
+    make_text_config,
+)
+
+
+def test_chat_simple(chat_tokenizer, builder):
+    config = make_chat_config()
+    item = {
+        "messages": [
+            {"role": "system", "content": "You are helpful."},
+            {"role": "user", "content": "Hello."},
+            {"role": "assistant", "content": "Hi there!"},
+        ]
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result is not None
+    assert "sequence" in result
+    assert "loss_mask" in result
+    assert len(result["sequence"]) == len(result["loss_mask"])
+
+    ids = chat_tokenizer.decode(result["sequence"], skip_special_tokens=False)
+    assert "system" in ids.lower() or "<|im_start|>system" in ids
+    assert "assistant" in ids.lower() or "<|im_start|>assistant" in ids
+
+    total = len(result["sequence"])
+    trained = sum(result["loss_mask"])
+    assert trained > 0
+    assert trained < total
+
+
+def test_chat_mask_only_assistant(chat_tokenizer, builder):
+    config = make_chat_config()
+    item = {
+        "messages": [
+            {"role": "user", "content": "What is 2+2?"},
+            {"role": "assistant", "content": "4"},
+        ]
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    mask = result["loss_mask"]
+    ids = result["sequence"]
+    assert len(ids) == len(mask)
+
+    trained = [i for i, m in enumerate(mask) if m == 1]
+    masked = [i for i, m in enumerate(mask) if m == 0]
+    assert len(trained) > 0
+    assert len(masked) > 0
+
+
+@pytest.mark.parametrize(
+    "mask_rules,mask_default,expect_nonzero",
+    [
+        ({"system": "mask", "user": "mask", "assistant": "mask"}, "mask", False),
+        ({}, "train", True),
+    ],
+)
+def test_chat_uniform_masking(
+    mask_rules, mask_default, expect_nonzero, chat_tokenizer, builder
+):
+    config = PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask=mask_rules,
+        mask_default=mask_default,
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+    item = {
+        "messages": [
+            {"role": "system", "content": "You are helpful."},
+            {"role": "assistant", "content": "Hi there!"},
+        ]
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    masked_count = sum(result["loss_mask"])
+    if expect_nonzero:
+        assert masked_count > 0
+    else:
+        assert masked_count == 0
+
+
+def test_chat_empty_messages(chat_tokenizer, builder):
+    config = make_chat_config()
+    assert builder.build({"messages": []}, config, chat_tokenizer) is None
+    assert builder.build({}, config, chat_tokenizer) is None
+
+
+def test_chat_domain_extraction(chat_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask={"assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+        output=OutputConfig(domain_key="source"),
+    )
+    item = {
+        "messages": [
+            {"role": "user", "content": "Hi"},
+            {"role": "assistant", "content": "Hello"},
+        ],
+        "source": "wiki",
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result["domain"] == "wiki"
+
+
+def test_chat_truncation(chat_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask={"assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=10),
+    )
+    item = {
+        "messages": [
+            {
+                "role": "user",
+                "content": "Tell me a very long story about dragons and knights and magic.",
+            },
+            {"role": "assistant", "content": "Sure! Here is a tale..."},
+        ]
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert len(result["sequence"]) <= 10
+    assert len(result["loss_mask"]) == len(result["sequence"])
+
+
+def test_instruction_basic(test_tokenizer, builder):
+    config = make_instruction_config()
+    item = {"prompt": "Translate to French: Hello", "response": "Bonjour"}
+    result = builder.build(item, config, test_tokenizer)
+    assert result is not None
+    assert len(result["sequence"]) == len(result["loss_mask"])
+
+
+def test_instruction_prompt_masked(test_tokenizer, builder):
+    config = make_instruction_config()
+    item = {"prompt": "hello", "response": "world"}
+    result = builder.build(item, config, test_tokenizer)
+    mask = result["loss_mask"]
+    ids = result["sequence"]
+
+    prompt_ids = test_tokenizer.encode("hello", add_special_tokens=True)
+    p_len = min(len(prompt_ids), len(ids))
+    assert all(m == 0 for m in mask[:p_len])
+    if p_len < len(ids):
+        assert all(m == 1 for m in mask[p_len:])
+
+
+def test_instruction_train_on_prompt(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(
+            sections=[
+                {"field": "prompt", "action": "train", "add_special_tokens": True},
+                {"field": "response", "action": "mask"},
+            ]
+        ),
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+    item = {"prompt": "hello", "response": "world"}
+    result = builder.build(item, config, test_tokenizer)
+    mask = result["loss_mask"]
+    ids = result["sequence"]
+
+    prompt_ids = test_tokenizer.encode("hello", add_special_tokens=True)
+    p_len = min(len(prompt_ids), len(ids))
+    assert all(m == 1 for m in mask[:p_len])
+
+
+def test_text_basic(test_tokenizer, builder):
+    config = make_text_config()
+    item = {"text": "Hello world. This is a test document."}
+    result = builder.build(item, config, test_tokenizer)
+    assert result is not None
+    assert "sequence" in result
+    assert len(result["sequence"]) > 0
+    assert "loss_mask" not in result
+
+
+def test_text_empty(test_tokenizer, builder):
+    config = make_text_config()
+    assert builder.build({"text": ""}, config, test_tokenizer) is None
+    assert builder.build({"text": "   "}, config, test_tokenizer) is None
+
+
+def test_text_too_short(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(min_chars=100),
+    )
+    assert builder.build({"text": "short"}, config, test_tokenizer) is None
+
+
+def test_text_truncation(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(max_seq_len=3, min_chars=1),
+    )
+    item = {"text": "This is a very long text that should be truncated"}
+    result = builder.build(item, config, test_tokenizer)
+    assert len(result["sequence"]) <= 3
+
+
+def test_sectioned_chat(chat_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask={"system": "mask", "user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+    )
+    item = {
+        "messages": [
+            {"role": "user", "content": "What is 2+2?"},
+            {"role": "assistant", "content": "4"},
+        ]
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result is not None
+    assert len(result["sequence"]) == len(result["loss_mask"])
+    assert sum(result["loss_mask"]) > 0
+    assert 0 in result["loss_mask"]
+
+
+def test_sectioned_instruction(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_INSTRUCTION_SECTIONS),
+        preprocessing=ProcessingConfig(max_seq_len=2048, min_chars=0),
+    )
+    item = {"prompt": "Q: Why?", "response": "A: Because."}
+    result = builder.build(item, config, test_tokenizer)
+    assert result is not None
+    mask = result["loss_mask"]
+    assert mask[0] == 0
+    assert mask[-1] == 1
+
+
+def test_sectioned_text(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(max_seq_len=2048, min_chars=1),
+    )
+    item = {"text": "Hello world, this is a test."}
+    result = builder.build(item, config, test_tokenizer)
+    assert result is not None
+    assert "loss_mask" not in result
+
+
+def test_sectioned_text_too_short(test_tokenizer, builder):
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(max_seq_len=2048, min_chars=100),
+    )
+    assert builder.build({"text": "short"}, config, test_tokenizer) is None
+
+
+def test_factory_registered():
+    names = MaskBuilderFactory.list_registered()
+    assert "sectioned" in names
+
+
+def test_factory_create():
+    builder_obj = MaskBuilderFactory.create("sectioned")
+    assert isinstance(builder_obj, SectionedMaskBuilder)
+
+
+def test_dpo_chat_basic(chat_tokenizer, builder):
+    config = make_dpo_chat_config()
+    item = {
+        "chosen": [
+            {"role": "user", "content": "What is 2+2?"},
+            {"role": "assistant", "content": "4"},
+        ],
+        "rejected": [
+            {"role": "user", "content": "What is 2+2?"},
+            {"role": "assistant", "content": "5"},
+        ],
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result is not None
+    assert "chosen" in result
+    assert "rejected" in result
+    assert "chosen_mask" in result
+    assert "rejected_mask" in result
+    assert len(result["chosen"]) == len(result["chosen_mask"])
+    assert len(result["rejected"]) == len(result["rejected_mask"])
+    assert sum(result["chosen_mask"]) > 0
+    assert sum(result["rejected_mask"]) > 0
+
+
+def test_dpo_chosen_only_trained(chat_tokenizer, builder):
+    config = make_dpo_chat_config()
+    item = {
+        "chosen": [
+            {"role": "user", "content": "Hi"},
+            {"role": "assistant", "content": "Hello"},
+        ],
+        "rejected": [
+            {"role": "user", "content": "Hi"},
+            {"role": "assistant", "content": "Go away"},
+        ],
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert 0 in result["chosen_mask"]
+    assert 1 in result["chosen_mask"]
+    assert 0 in result["rejected_mask"]
+    assert 1 in result["rejected_mask"]
+
+
+def test_dpo_missing_field_is_none(chat_tokenizer, builder):
+    config = make_dpo_chat_config()
+    assert builder.build({"chosen": [], "rejected": []}, config, chat_tokenizer) is None
+
+
+def test_grpo_basic(chat_tokenizer, builder):
+    config = make_grpo_config()
+    item = {
+        "prompt": [{"role": "user", "content": "What is 2+2?"}],
+        "responses": ["4", "The answer is four", "Four", "2+2=4"],
+        "rewards": [1.0, 0.5, 0.8, 0.2],
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result is not None
+    assert "prompts" in result
+    assert "responses" in result
+    assert "masks" in result
+    assert "rewards" in result
+    assert len(result["responses"]) == len(result["masks"])
+    assert result["rewards"] == [1.0, 0.5, 0.8, 0.2]
+
+
+def test_grpo_response_tokens_all_trained(chat_tokenizer, builder):
+    config = make_grpo_config()
+    item = {
+        "prompt": [{"role": "user", "content": "Q"}],
+        "responses": ["A", "B"],
+        "rewards": [0.8, 0.2],
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    masks = result["masks"]
+    assert all(m == 1 for m in masks)
+    assert len(masks) == len(result["responses"])
+
+
+def test_grpo_single_reward(chat_tokenizer, builder):
+    config = make_grpo_config()
+    item = {
+        "prompt": [{"role": "user", "content": "Q"}],
+        "responses": ["A"],
+        "rewards": 0.9,
+    }
+    result = builder.build(item, config, chat_tokenizer)
+    assert result["rewards"] == [0.9]

tests/data/test_preprocess_config.py

@@ -0,0 +1,77 @@
+import os
+
+from astrai.config.preprocess_config import (
+    InputConfig,
+    PipelineConfig,
+)
+from tests.data.conftest import (
+    _INSTRUCTION_SECTIONS,
+    _TEXT_SECTIONS,
+    make_dpo_chat_config,
+)
+
+
+def test_default_values():
+    config = PipelineConfig()
+    assert config.version == 1
+    assert config.mask == {}
+    assert config.mask_default == "mask"
+    assert config.preprocessing.max_seq_len == 2048
+    assert config.output.storage_format == "bin"
+    assert config.input.sections is None
+
+
+def test_from_dict_flat():
+    data = {
+        "version": 1,
+        "input": {
+            "sections": [{"field": "messages", "action": "$role", "template": True}]
+        },
+        "mask": {"system": "mask", "assistant": "train"},
+        "mask_default": "mask",
+        "preprocessing": {"max_seq_len": 1024},
+        "output": {"storage_format": "h5"},
+    }
+    config = PipelineConfig.from_dict(data)
+    assert config.input.sections == [
+        {"field": "messages", "action": "$role", "template": True}
+    ]
+    assert config.mask == {"system": "mask", "assistant": "train"}
+    assert config.preprocessing.max_seq_len == 1024
+    assert config.output.storage_format == "h5"
+
+
+def test_to_dict_roundtrip():
+    config = PipelineConfig(
+        input=InputConfig(sections=_INSTRUCTION_SECTIONS),
+        mask={"prompt": "mask", "response": "train"},
+        mask_default="mask",
+    )
+    d = config.to_dict()
+    config2 = PipelineConfig.from_dict(d)
+    assert config2.input.sections == _INSTRUCTION_SECTIONS
+    assert config2.mask == {"prompt": "mask", "response": "train"}
+
+
+def test_to_file_from_file(temp_dir):
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        mask={"text": "train"},
+        mask_default="mask",
+    )
+    path = os.path.join(temp_dir, "config.json")
+    config.to_file(path)
+    loaded = PipelineConfig.from_file(path)
+    assert loaded.input.sections == _TEXT_SECTIONS
+    assert loaded.mask == {"text": "train"}
+
+
+def test_dpo_config_roundtrip(temp_dir):
+    config = make_dpo_chat_config()
+    path = os.path.join(temp_dir, "config.json")
+    config.to_file(path)
+    loaded = PipelineConfig.from_file(path)
+    assert loaded.input.sources is not None
+    assert "chosen" in loaded.input.sources
+    assert "rejected" in loaded.input.sources
+    assert loaded.input.sections is None

tests/data/test_preprocess_pipeline.py

@@ -0,0 +1,264 @@
+import json
+import os
+
+from astrai.config.preprocess_config import (
+    InputConfig,
+    OutputConfig,
+    PipelineConfig,
+    ProcessingConfig,
+)
+from astrai.preprocessing.pipeline import Pipeline, filter_by_length
+from tests.data.conftest import (
+    _CHAT_SECTIONS,
+    _INSTRUCTION_SECTIONS,
+    _TEXT_SECTIONS,
+    make_dpo_chat_config,
+    make_grpo_no_template_config,
+)
+
+
+def test_filter_by_length():
+    assert filter_by_length("hello world", min_len=5)
+    assert not filter_by_length("hi", min_len=5)
+    assert not filter_by_length("x" * 100, max_len=50)
+    assert filter_by_length("just right", min_len=5, max_len=20)
+
+
+def test_full_chat_pipeline(temp_dir, chat_tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "chat.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(
+            json.dumps(
+                {
+                    "messages": [
+                        {"role": "system", "content": "You are helpful."},
+                        {"role": "user", "content": "Hi."},
+                        {"role": "assistant", "content": "Hello!"},
+                    ]
+                }
+            )
+            + "\n"
+        )
+        f.write(
+            json.dumps(
+                {
+                    "messages": [
+                        {"role": "user", "content": "What is 2+2?"},
+                        {"role": "assistant", "content": "4"},
+                    ]
+                }
+            )
+            + "\n"
+        )
+
+    config = PipelineConfig(
+        input=InputConfig(sections=_CHAT_SECTIONS),
+        mask={"system": "mask", "user": "mask", "assistant": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+        output=OutputConfig(storage_format="bin", domain_key=None),
+    )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=config,
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=chat_tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    assert os.path.exists(meta_path)
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert "sequence" in meta
+    assert "loss_mask" in meta
+    assert meta["sequence"]["dtype"] == "int32"
+    assert meta["loss_mask"]["dtype"] == "int32"
+
+
+def test_full_text_pipeline(temp_dir, tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "text.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(
+            json.dumps(
+                {
+                    "text": "Hello world this is a test document with enough characters to pass the minimum length filter."
+                }
+            )
+            + "\n"
+        )
+        f.write(
+            json.dumps(
+                {
+                    "text": "Another document for testing purposes with sufficient length to be processed."
+                }
+            )
+            + "\n"
+        )
+
+    config = PipelineConfig(
+        input=InputConfig(sections=_TEXT_SECTIONS),
+        preprocessing=ProcessingConfig(max_seq_len=2048, min_chars=10),
+        output=OutputConfig(storage_format="bin"),
+    )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=config,
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    assert os.path.exists(meta_path)
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert "sequence" in meta
+    assert "loss_mask" not in meta
+
+
+def test_full_instruction_pipeline(temp_dir, tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "instruct.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(
+            json.dumps(
+                {
+                    "prompt": "Tell me a joke",
+                    "response": "Why did the chicken cross the road?",
+                }
+            )
+            + "\n"
+        )
+        f.write(
+            json.dumps(
+                {
+                    "prompt": "What is AI?",
+                    "response": "Artificial Intelligence is a field of computer science.",
+                }
+            )
+            + "\n"
+        )
+
+    config = PipelineConfig(
+        input=InputConfig(sections=_INSTRUCTION_SECTIONS),
+        mask={"prompt": "mask", "response": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+        output=OutputConfig(storage_format="bin"),
+    )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=config,
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    assert os.path.exists(meta_path)
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert "sequence" in meta
+    assert "loss_mask" in meta
+
+
+def test_dtype_override(temp_dir, tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "data.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(json.dumps({"prompt": "Q", "response": "A"}) + "\n")
+
+    config = PipelineConfig(
+        input=InputConfig(sections=_INSTRUCTION_SECTIONS),
+        mask={"prompt": "mask", "response": "train"},
+        mask_default="mask",
+        preprocessing=ProcessingConfig(max_seq_len=2048),
+        output=OutputConfig(storage_format="bin", dtype={"loss_mask": "bool"}),
+    )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=config,
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert meta["sequence"]["dtype"] == "int32"
+    assert meta["loss_mask"]["dtype"] == "bool"
+
+
+def test_dpo_pipeline(temp_dir, chat_tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "dpo.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(
+            json.dumps(
+                {
+                    "chosen": [
+                        {"role": "user", "content": "Hi."},
+                        {"role": "assistant", "content": "Hello!"},
+                    ],
+                    "rejected": [
+                        {"role": "user", "content": "Hi."},
+                        {"role": "assistant", "content": "Go away."},
+                    ],
+                }
+            )
+            + "\n"
+        )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=make_dpo_chat_config(),
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=chat_tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    assert os.path.exists(meta_path)
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert "chosen" in meta
+    assert "rejected" in meta
+    assert "chosen_mask" in meta
+    assert "rejected_mask" in meta
+    assert "sequence" not in meta
+
+
+def test_grpo_pipeline(temp_dir, tokenizer_dir):
+    jsonl_path = os.path.join(temp_dir, "grpo.jsonl")
+    with open(jsonl_path, "w", encoding="utf-8") as f:
+        f.write(
+            json.dumps(
+                {
+                    "prompt": "Question?",
+                    "responses": ["Answer A", "Answer B"],
+                    "rewards": [0.8, 0.3],
+                }
+            )
+            + "\n"
+        )
+
+    out_dir = os.path.join(temp_dir, "output")
+    Pipeline(
+        config=make_grpo_no_template_config(),
+        input_paths=[jsonl_path],
+        output_dir=out_dir,
+        tokenizer_path=tokenizer_dir,
+    ).run()
+
+    meta_path = os.path.join(out_dir, "__default__", "shard_0000", "meta.json")
+    assert os.path.exists(meta_path)
+    with open(meta_path, "r") as f:
+        meta = json.load(f)
+    assert "prompts" in meta
+    assert "responses" in meta
+    assert "masks" in meta
+    assert "rewards" in meta
+    assert "sequence" not in meta

tests/inference/conftest.py

@@ -5,21 +5,22 @@ from unittest.mock import MagicMock
 import pytest
 from fastapi.testclient import TestClient
 
-from astrai.inference import app
+from astrai.inference import get_app
 
 
 @pytest.fixture
 def client():
     """Provide a test client for the FastAPI app."""
-    app.state.server_config = {
+    _app = get_app()
+    _app.state.server_config = {
         "device": "cpu",
         "dtype": "bfloat16",
         "param_path": None,
         "max_batch_size": 1,
         "_test": True,
     }
-    app.state.engine = None
+    _app.state.engine = None
-    return TestClient(app)
+    return TestClient(_app)
 
 
 @pytest.fixture
@@ -49,5 +50,5 @@ def mock_engine():
 @pytest.fixture
 def loaded_model(client, mock_engine):
     """Simulate that the engine is loaded."""
-    app.state.engine = mock_engine
+    get_app().state.engine = mock_engine
     return mock_engine

tests/inference/test_protocol.py

@@ -121,8 +121,8 @@ class TestOpenAIResponseBuilder:
         assert p["choices"][0]["finish_reason"] is None
 
     def test_format_chunk(self, builder):
-        event = builder.format_chunk("hello")
+        events = builder.format_chunk("hello", body="hello")
-        payload = json.loads(event.split("data: ", 1)[1])
+        payload = json.loads(events[0].split("data: ", 1)[1])
         assert payload["choices"][0]["delta"]["content"] == "hello"
         assert payload["choices"][0]["finish_reason"] is None
 
@@ -192,8 +192,8 @@ class TestAnthropicResponseBuilder:
         assert payloads[1]["type"] == "content_block_start"
 
     def test_format_chunk(self, builder):
-        event = builder.format_chunk("tok")
+        events = builder.format_chunk("tok", body="tok")
-        payload = json.loads(event.split("data: ", 1)[1])
+        payload = json.loads(events[0].split("data: ", 1)[1])
         assert payload["type"] == "content_block_delta"
         assert payload["delta"]["text"] == "tok"
 

tests/inference/test_server.py

@@ -2,12 +2,12 @@
 
 import pytest
 
-from astrai.inference import app
+from astrai.inference import get_app
 
 
 def test_health_no_model(client):
     """GET /health should return 200 even when engine not loaded."""
-    app.state.engine = None
+    get_app().state.engine = None
     response = client.get("/health")
     assert response.status_code == 200
     data = response.json()
@@ -30,7 +30,7 @@ def test_chat_completions_non_stream(client, loaded_model):
     async def async_gen():
         yield "Assistant reply"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/chat/completions",
@@ -56,7 +56,7 @@ def test_chat_completions_stream(client, loaded_model):
         yield "cumulative1"
         yield "cumulative2"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/chat/completions",
@@ -83,7 +83,7 @@ def test_messages_non_stream(client, loaded_model):
     async def async_gen():
         yield "Assistant reply"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/messages",
@@ -111,7 +111,7 @@ def test_messages_stream(client, loaded_model):
         yield "cumulative1"
         yield "cumulative2"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/messages",
@@ -141,7 +141,7 @@ def test_messages_with_system(client, loaded_model):
     async def async_gen():
         yield "Reply"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/messages",
@@ -165,7 +165,7 @@ def test_chat_completions_stop_sequence(client, loaded_model):
         yield "X"
         yield "world"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/chat/completions",
@@ -191,7 +191,7 @@ def test_chat_completions_stop_sequence_stream(client, loaded_model):
         yield "X"
         yield "world"
 
-    app.state.engine = loaded_model
+    get_app().state.engine = loaded_model
     loaded_model.generate_async.return_value = async_gen()
     response = client.post(
         "/v1/chat/completions",

tests/inference/test_tool_parser.py

@@ -0,0 +1,608 @@
+"""Unit tests for tool call parsers."""
+
+import pytest
+
+from astrai.inference.api.tool_parser import (
+    _TOOL_CALL_HEAD_RE,
+    BaseToolParser,
+    SimpleJsonToolParser,
+    ToolParserFactory,
+    _find_partial_tool_call,
+    _find_tool_calls,
+    _scan_json,
+)
+
+
+@pytest.mark.parametrize(
+    "text,expected_complete,check_end_eq_len",
+    [
+        ('{"key": "value"}', True, True),
+        ('{"outer": {"inner": 1}}', True, True),
+        ('{"key": "value"', False, False),
+        ('{"outer": {"inner": 1}', False, False),
+        ('{"key": "a{b}c"} extra', True, False),
+        (r'{"key": "a\"b"}', True, False),
+        ('{"a": {"b": {"c": {"d": {"e": 5}}}}}', True, True),
+        ('{"items": [{"x": 1}, {"x": 2}]}', True, True),
+        ('{"fn": "function() { return 1; }"}', True, False),
+        ('{"key": "\u5317\u4eac"}', True, False),
+    ],
+)
+def test_scan_json(text, expected_complete, check_end_eq_len):
+    end, complete = _scan_json(text, 0)
+    assert complete is expected_complete
+    if check_end_eq_len:
+        assert end == len(text)
+
+
+def test_find_single_tool_call():
+    text = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "get_weather"
+    assert '"city"' in results[0]["args"]
+    assert results[0]["complete"] is True
+
+
+def test_find_text_before_tool_call():
+    text = 'Some text {"name": "func", "arguments": {}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["start"] > 0
+
+
+def test_find_multiple_tool_calls():
+    text = '{"name": "f1", "arguments": {"a": 1}}{"name": "f2", "arguments": {"b": 2}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 2
+    assert results[0]["name"] == "f1"
+    assert results[1]["name"] == "f2"
+
+
+def test_find_no_tool_call():
+    results = _find_tool_calls("Hello, how are you?")
+    assert len(results) == 0
+
+
+def test_find_non_tool_json_skipped():
+    results = _find_tool_calls('{"not_a_tool": true}')
+    assert len(results) == 0
+
+
+def test_find_no_arguments_field():
+    results = _find_tool_calls('{"name": "simple_func"}')
+    assert len(results) == 1
+    assert results[0]["name"] == "simple_func"
+    assert results[0]["args"] == ""
+
+
+def test_find_deeply_nested_arguments():
+    text = '{"name": "deep", "arguments": {"a": {"b": {"c": {"d": 4}}}}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "deep"
+    assert '"d": 4' in results[0]["args"]
+
+
+def test_find_arguments_with_boolean_and_null():
+    text = '{"name": "flags", "arguments": {"active": true, "count": 0, "nick": null}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "flags"
+    assert "true" in results[0]["args"]
+    assert "null" in results[0]["args"]
+
+
+def test_find_arguments_with_array():
+    text = '{"name": "add_items", "arguments": {"items": [1, 2, 3], "name": "list"}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "add_items"
+    assert "[1, 2, 3]" in results[0]["args"]
+
+
+def test_find_arguments_with_nested_array_of_objects():
+    text = '{"name": "batch", "arguments": {"rows": [{"id": 1, "val": "a"}, {"id": 2, "val": "b"}]}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert '"rows"' in results[0]["args"]
+    assert '"id": 1' in results[0]["args"]
+
+
+def test_find_arguments_as_string_not_object():
+    text = '{"name": "echo", "arguments": "just a string"}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "echo"
+    assert "just a string" in results[0]["args"]
+
+
+def test_find_arguments_with_unicode():
+    text = (
+        '{"name": "translate", "arguments": {"text": "\u4f60\u597d\uff0c\u4e16\u754c"}}'
+    )
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "translate"
+
+
+def test_find_arguments_with_escaped_quotes():
+    text = '{"name": "format", "arguments": {"template": "he said \\"hello\\""}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert 'he said \\"hello\\"' in results[0]["args"]
+
+
+def test_find_arguments_with_braces_in_string():
+    text = '{"name": "eval", "arguments": {"code": "function(x) { return x + 1; }"}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "eval"
+    assert "function(x) { return x + 1; }" in results[0]["args"]
+
+
+def test_find_many_properties():
+    args = ",".join(f'"{chr(97 + i % 26)}" : {i}' for i in range(20))
+    text = '{"name": "many", "arguments": {' + args + "}}"
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    assert results[0]["name"] == "many"
+
+
+def test_find_empty_arguments():
+    results = _find_tool_calls('{"name": "ping", "arguments": {}}')
+    assert len(results) == 1
+    assert results[0]["name"] == "ping"
+    assert results[0]["args"] == ""
+
+
+def test_find_extracts_correct_arg_start_position():
+    text = '{"name": "f", "arguments": {"x": 1}}'
+    results = _find_tool_calls(text)
+    assert len(results) == 1
+    json_str = text[results[0]["start"] : results[0]["end"]]
+    assert json_str == text
+
+
+@pytest.mark.parametrize(
+    "text,expected_name,expected_complete",
+    [
+        ('{"name": "func", "arguments": {"city"', "func", False),
+        ('{"name": "func", "arguments": {"city": "BJ"}}', "func", None),
+        ("plain text", None, None),
+        ('{"nam', None, None),
+        ('{"name": "deep", "arguments": {"a": {"b": {"c": ', "deep", None),
+        ('{"name": "batch", "arguments": {"items": [1, 2, ', "batch", None),
+    ],
+)
+def test_find_partial_tool_call(text, expected_name, expected_complete):
+    result = _find_partial_tool_call(text)
+    if expected_name is None:
+        assert result is None
+    else:
+        assert result is not None
+        assert result["name"] == expected_name
+        if expected_complete is not None:
+            assert result["complete"] is expected_complete
+
+
+def test_feed_plain_text():
+    parser = SimpleJsonToolParser()
+    deltas = parser.feed("Hello")
+    assert len(deltas) == 1
+    assert deltas[0]["content"] == "Hello"
+
+
+def test_feed_incremental_text():
+    parser = SimpleJsonToolParser()
+    assert parser.feed("He") == [{"content": "He"}]
+    assert parser.feed("Hello") == [{"content": "llo"}]
+
+
+def test_feed_tool_call_name_delta():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    deltas = parser.feed(text)
+    tc_deltas = [d for d in deltas if "tool_calls" in d]
+    assert len(tc_deltas) >= 1
+    name_delta = tc_deltas[0]["tool_calls"][0]
+    assert name_delta["function"]["name"] == "get_weather"
+    assert name_delta["type"] == "function"
+    assert "id" in name_delta
+
+
+def test_feed_tool_call_args_streaming():
+    parser = SimpleJsonToolParser()
+    d1 = parser.feed('{"name": "f", "arguments": {"x":')
+    d2 = parser.feed('{"name": "f", "arguments": {"x": "1"}}')
+    args_deltas = [
+        d
+        for batch in (d1, d2)
+        for d in batch
+        if "tool_calls" in d
+        and "function" in d["tool_calls"][0]
+        and "arguments" in d["tool_calls"][0]["function"]
+    ]
+    assert len(args_deltas) >= 1
+
+
+def test_feed_text_before_tool_call():
+    parser = SimpleJsonToolParser()
+    text = 'Let me check. {"name": "func", "arguments": {"a": 1}}'
+    deltas = parser.feed(text)
+    content_deltas = [d for d in deltas if "content" in d]
+    assert any("Let me check" in d.get("content", "") for d in content_deltas)
+
+
+def test_has_tool_calls_false_by_default():
+    assert SimpleJsonToolParser().has_tool_calls is False
+
+
+def test_has_tool_calls_true_after_detection():
+    parser = SimpleJsonToolParser()
+    parser.feed('{"name": "f", "arguments": {}}')
+    assert parser.has_tool_calls is True
+
+
+def test_feed_no_content_when_no_new_text():
+    parser = SimpleJsonToolParser()
+    parser.feed("Hello")
+    assert parser.feed("Hello") == []
+
+
+def test_feed_multiple_tool_calls():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "f1", "arguments": {"a": 1}}{"name": "f2", "arguments": {"b": 2}}'
+    deltas = parser.feed(text)
+    tc_deltas = [d for d in deltas if "tool_calls" in d]
+    names = set()
+    for batch in tc_deltas:
+        for tc in batch["tool_calls"]:
+            if "function" in tc and "name" in tc["function"]:
+                names.add(tc["function"]["name"])
+    assert "f1" in names
+    assert "f2" in names
+
+
+def test_feed_with_tools_constructor():
+    tools = [{"type": "function", "function": {"name": "get_weather"}}]
+    parser = SimpleJsonToolParser(tools=tools, tool_choice="auto")
+    deltas = parser.feed('{"name": "get_weather", "arguments": {"city": "BJ"}}')
+    assert len(deltas) > 0
+
+
+def test_feed_content_after_tool_call_is_not_emitted():
+    parser = SimpleJsonToolParser()
+    parser.feed('{"name": "f", "arguments": {}} trailing text')
+    assert parser.has_tool_calls
+
+
+def _simulate_streaming(parser, text):
+    all_delta_names = []
+    all_args_chunks = []
+    for i in range(1, len(text) + 1):
+        deltas = parser.feed(text[:i])
+        for d in deltas:
+            if "tool_calls" in d:
+                for tc in d["tool_calls"]:
+                    fn = tc.get("function", {})
+                    if "name" in fn:
+                        all_delta_names.append(fn["name"])
+                    if "arguments" in fn and fn["arguments"]:
+                        all_args_chunks.append(fn["arguments"])
+    return all_delta_names, all_args_chunks
+
+
+def test_streaming_token_by_token_full_build():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    names, args_chunks = _simulate_streaming(parser, text)
+    assert "get_weather" in names
+    joined_args = "".join(args_chunks)
+    assert '"city"' in joined_args
+    assert "Beijing" in joined_args
+
+
+def test_streaming_token_by_token_text_then_tool():
+    parser = SimpleJsonToolParser()
+    parts = [
+        "I'll ",
+        "check ",
+        "that. ",
+        '{"',
+        'name": "search", ',
+        '"arguments": {"q": "hello"}}',
+    ]
+    body = ""
+    content_chunks = []
+    tool_names = []
+    for part in parts:
+        body += part
+        deltas = parser.feed(body)
+        for d in deltas:
+            if "content" in d:
+                content_chunks.append(d["content"])
+            if "tool_calls" in d:
+                for tc in d["tool_calls"]:
+                    fn = tc.get("function", {})
+                    if "name" in fn:
+                        tool_names.append(fn["name"])
+    full_content = "".join(content_chunks)
+    assert "I'll check that." in full_content
+    assert "search" in tool_names
+
+
+def test_streaming_multiple_tool_calls_incremental():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "f1", "arguments": {"a": 1}}{"name": "f2", "arguments": {"b": 2}}'
+    names, _ = _simulate_streaming(parser, text)
+    assert names[0] == "f1"
+    assert "f2" in names
+
+
+def test_streaming_deeply_nested_args():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "deep", "arguments": {"a": {"b": {"c": 42}}}}'
+    _, args_chunks = _simulate_streaming(parser, text)
+    joined = "".join(args_chunks)
+    assert '"c": 42' in joined
+
+
+def test_streaming_args_with_unicode():
+    parser = SimpleJsonToolParser()
+    text = (
+        '{"name": "translate", "arguments": {"text": "\u4f60\u597d\uff0c\u4e16\u754c"}}'
+    )
+    _, args_chunks = _simulate_streaming(parser, text)
+    joined = "".join(args_chunks)
+    assert "\u4f60\u597d" in joined
+
+
+def test_streaming_args_with_array():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "add", "arguments": {"items": [1, 2, 3]}}'
+    _, args_chunks = _simulate_streaming(parser, text)
+    joined = "".join(args_chunks)
+    assert "[1, 2, 3]" in joined
+
+
+def test_streaming_empty_arguments():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "ping", "arguments": {}}'
+    deltas = parser.feed(text)
+    tc_deltas = [d for d in deltas if "tool_calls" in d]
+    assert len(tc_deltas) >= 1
+    name_delta = tc_deltas[0]["tool_calls"][0]
+    assert name_delta["function"]["name"] == "ping"
+    assert "arguments" in name_delta["function"]
+
+
+def test_streaming_args_diff_only_emits_new_bytes():
+    parser = SimpleJsonToolParser()
+    step1 = parser.feed('{"name": "f", "arguments": {"city": "Bei')
+    step2 = parser.feed('{"name": "f", "arguments": {"city": "Beijing"}}')
+    all_args = []
+    for step in (step1, step2):
+        for d in step:
+            if "tool_calls" in d:
+                for tc in d["tool_calls"]:
+                    fn = tc.get("function", {})
+                    if "arguments" in fn and fn["arguments"]:
+                        all_args.append(fn["arguments"])
+    joined = "".join(all_args)
+    assert "city" in joined
+    assert "Beijing" in joined
+    assert joined.startswith('"city":')
+    assert all_args[0] != all_args[1]
+
+
+def test_streaming_distinct_tool_call_ids():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "f1", "arguments": {"a": 1}}{"name": "f2", "arguments": {"b": 2}}'
+    all_ids = []
+    for i in range(1, len(text) + 1):
+        deltas = parser.feed(text[:i])
+        for d in deltas:
+            if "tool_calls" in d:
+                for tc in d["tool_calls"]:
+                    if "id" in tc:
+                        all_ids.append(tc["id"])
+    unique = list(dict.fromkeys(all_ids))
+    assert len(unique) == 2
+
+
+def test_parse_complete_basic():
+    parser = SimpleJsonToolParser()
+    body = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    result = parser.parse_complete(body)
+    assert result is not None
+    assert result["tool_calls"][0]["function"]["name"] == "get_weather"
+    assert "Beijing" in result["tool_calls"][0]["function"]["arguments"]
+
+
+def test_parse_complete_no_tool_call():
+    assert SimpleJsonToolParser().parse_complete("Hello world") is None
+
+
+def test_parse_complete_with_content():
+    parser = SimpleJsonToolParser()
+    result = parser.parse_complete('Prefix text. {"name": "f", "arguments": {}}')
+    assert result is not None
+    assert result["content"] == "Prefix text."
+
+
+def test_parse_complete_multiple_tool_calls():
+    parser = SimpleJsonToolParser()
+    body = '{"name": "get_weather", "arguments": {"city": "Beijing"}}{"name": "get_time", "arguments": {"tz": "Asia/Shanghai"}}'
+    result = parser.parse_complete(body)
+    assert result is not None
+    assert len(result["tool_calls"]) == 2
+    assert result["tool_calls"][0]["function"]["name"] == "get_weather"
+    assert result["tool_calls"][1]["function"]["name"] == "get_time"
+
+
+def test_parse_complete_complex_real_world():
+    parser = SimpleJsonToolParser()
+    body = (
+        '{"name": "send_email", "arguments": {'
+        '"to": ["a@b.com", "c@d.com"], "cc": null, '
+        '"subject": "Hello World", "body": "This is a test email.", '
+        '"priority": 1, "attachments": false}}'
+    )
+    result = parser.parse_complete(body)
+    assert result is not None
+    tc = result["tool_calls"][0]
+    assert tc["function"]["name"] == "send_email"
+    args = tc["function"]["arguments"]
+    assert '"to"' in args
+    assert "a@b.com" in args
+    assert "null" in args
+    assert "false" in args
+
+
+def test_parse_complete_content_with_multiple_tool_calls():
+    parser = SimpleJsonToolParser()
+    body = 'I will do two things. {"name": "f1", "arguments": {"a": 1}}{"name": "f2", "arguments": {"b": 2}}'
+    result = parser.parse_complete(body)
+    assert result is not None
+    assert result["content"] == "I will do two things."
+    assert len(result["tool_calls"]) == 2
+
+
+def test_parse_complete_no_arguments_field():
+    parser = SimpleJsonToolParser()
+    result = parser.parse_complete('{"name": "ping"}')
+    assert result is not None
+    assert result["tool_calls"][0]["function"]["name"] == "ping"
+    assert result["tool_calls"][0]["function"]["arguments"] == ""
+
+
+def test_parse_complete_content_is_none_when_pure_tool_call():
+    parser = SimpleJsonToolParser()
+    result = parser.parse_complete('{"name": "f", "arguments": {"x": 1}}')
+    assert result is not None
+    assert result["content"] is None
+
+
+def test_parse_complete_tool_calls_have_ids():
+    parser = SimpleJsonToolParser()
+    result = parser.parse_complete(
+        '{"name": "f1", "arguments": {}}{"name": "f2", "arguments": {}}'
+    )
+    assert result is not None
+    ids = [tc["id"] for tc in result["tool_calls"]]
+    assert len(ids) == 2
+    assert all(isinstance(i, str) and i.startswith("call_") for i in ids)
+    assert ids[0] != ids[1]
+
+
+def test_feed_then_parse_complete_same_instance():
+    parser = SimpleJsonToolParser()
+    parser.feed('{"name": "get_weather", "arguments": {"city": "Beijing"}}')
+    result = parser.parse_complete(
+        '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    )
+    assert result is not None
+    assert result["tool_calls"][0]["function"]["name"] == "get_weather"
+    assert parser.has_tool_calls
+
+
+@pytest.mark.parametrize(
+    "text,matches",
+    [
+        ('{"name": "f"}', True),
+        ('{ "name" : "f"}', True),
+        ('{"other": 1}', False),
+        ('prefix {"name": "f", "args": {}}', True),
+        ('{"name": "f"}', True),  # match at start
+        ('   {"name": "f"}', True),
+    ],
+)
+def test_pattern_regex(text, matches):
+    result = _TOOL_CALL_HEAD_RE.search(text)
+    if matches:
+        assert result is not None
+    else:
+        assert result is None
+
+
+def test_pattern_name_at_start():
+    assert _TOOL_CALL_HEAD_RE.match('{"name": "f"}')
+
+
+def test_factory_register_and_create():
+    parser = ToolParserFactory.create("simple_json")
+    assert isinstance(parser, BaseToolParser)
+    assert isinstance(parser, SimpleJsonToolParser)
+
+
+def test_factory_create_passes_tools():
+    parser = ToolParserFactory.create(
+        "simple_json", tools=[{"type": "function"}], tool_choice="required"
+    )
+    assert parser.tool_choice == "required"
+
+
+def test_factory_list_registered():
+    assert "simple_json" in ToolParserFactory.list_registered()
+
+
+def test_factory_create_with_no_extra_kwargs():
+    assert isinstance(ToolParserFactory.create("simple_json"), BaseToolParser)
+
+
+def test_factory_create_with_tools_only():
+    tools = [
+        {
+            "type": "function",
+            "function": {"name": "test", "parameters": {"type": "object"}},
+        }
+    ]
+    parser = ToolParserFactory.create("simple_json", tools=tools)
+    assert parser.tools == tools
+    assert parser.tool_choice == "auto"
+
+
+def test_feed_accepts_token_ids_and_ignores_them():
+    parser = SimpleJsonToolParser()
+    text = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    deltas_with = parser.feed(text, current_token_ids=[123, 456], delta_token_ids=[456])
+    assert len(deltas_with) > 0
+
+
+def test_feed_token_ids_do_not_affect_parsing():
+    parser_no_ids = SimpleJsonToolParser()
+    parser_with_ids = SimpleJsonToolParser()
+    text = '{"name": "get_weather", "arguments": {"city": "Beijing"}}'
+    result_no = parser_no_ids.feed(text)
+    result_with = parser_with_ids.feed(
+        text, current_token_ids=[1, 2, 3], delta_token_ids=[3]
+    )
+    assert len(result_no) == len(result_with)
+    assert (
+        result_no[0]["tool_calls"][0]["function"]["name"]
+        == result_with[0]["tool_calls"][0]["function"]["name"]
+    )
+
+
+def test_parser_uses_token_ids_for_detection():
+    class TokenIdParser(BaseToolParser):
+        def __init__(self, tools=None, tool_choice="auto"):
+            super().__init__(tools, tool_choice)
+            self._detections = 0
+
+        def feed(self, body, current_token_ids=None, delta_token_ids=None):
+            if current_token_ids and 999 in current_token_ids:
+                self._detections += 1
+            return []
+
+        def parse_complete(self, body):
+            return None
+
+        @property
+        def has_tool_calls(self):
+            return self._detections > 0
+
+    parser = TokenIdParser()
+    parser.feed("hello", current_token_ids=[1, 999, 3])
+    assert parser.has_tool_calls

tests/module/test_encoder.py

@@ -1,6 +1,13 @@
+import json
+import os
+import tempfile
+
+import pytest
+import safetensors.torch as st
 import torch
 
 from astrai.config.model_config import EncoderConfig
+from astrai.model.automodel import AutoModel
 from astrai.model.encoder import EmbeddingEncoder
 
 TINY_CONFIG = dict(
@@ -14,92 +21,56 @@ TINY_CONFIG = dict(
     norm_eps=1e-5,
 )
 
+_device = "cuda" if torch.cuda.is_available() else "cpu"
 
-def test_encoder_forward_mean():
+
-    config = EncoderConfig(**TINY_CONFIG)
+def _make_model(**kwargs):
-    device = "cuda" if torch.cuda.is_available() else "cpu"
+    config = EncoderConfig(**{**TINY_CONFIG, **kwargs})
-    model = EmbeddingEncoder(config).to(device=device)
+    return EmbeddingEncoder(config).to(device=_device)
+
+
+@pytest.mark.parametrize("pooling_type", ["mean", "cls", "last"])
+def test_encoder_forward_pooling(pooling_type):
+    model = _make_model(pooling_type=pooling_type)
     model.eval()
 
     batch_size, seq_len = 2, 8
     input_ids = torch.randint(
-        0, config.vocab_size, (batch_size, seq_len), device=device
+        0, TINY_CONFIG["vocab_size"], (batch_size, seq_len), device=_device
     )
 
     with torch.no_grad():
         output = model(input_ids)
 
-    assert output.shape == (batch_size, config.dim)
+    assert output.shape == (batch_size, TINY_CONFIG["dim"])
-    assert not torch.isnan(output).any()
-
-
-def test_encoder_forward_cls():
-    config = EncoderConfig(**{**TINY_CONFIG, "pooling_type": "cls"})
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model = EmbeddingEncoder(config).to(device=device)
-    model.eval()
-
-    batch_size, seq_len = 2, 8
-    input_ids = torch.randint(
-        0, config.vocab_size, (batch_size, seq_len), device=device
-    )
-
-    with torch.no_grad():
-        output = model(input_ids)
-
-    assert output.shape == (batch_size, config.dim)
-    assert not torch.isnan(output).any()
-
-
-def test_encoder_forward_last():
-    config = EncoderConfig(**{**TINY_CONFIG, "pooling_type": "last"})
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model = EmbeddingEncoder(config).to(device=device)
-    model.eval()
-
-    batch_size, seq_len = 2, 8
-    input_ids = torch.randint(
-        0, config.vocab_size, (batch_size, seq_len), device=device
-    )
-
-    with torch.no_grad():
-        output = model(input_ids)
-
-    assert output.shape == (batch_size, config.dim)
     assert not torch.isnan(output).any()
 
 
 def test_encoder_forward_with_padding():
-    config = EncoderConfig(**TINY_CONFIG)
+    model = _make_model()
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model = EmbeddingEncoder(config).to(device=device)
     model.eval()
 
     batch_size, seq_len = 2, 8
     input_ids = torch.randint(
-        0, config.vocab_size, (batch_size, seq_len), device=device
+        0, TINY_CONFIG["vocab_size"], (batch_size, seq_len), device=_device
     )
-    input_mask = torch.ones(batch_size, seq_len, dtype=torch.bool, device=device)
+    input_mask = torch.ones(batch_size, seq_len, dtype=torch.bool, device=_device)
     input_mask[:, 4:] = False
 
     with torch.no_grad():
         output = model(input_ids, input_mask=input_mask)
 
-    assert output.shape == (batch_size, config.dim)
+    assert output.shape == (batch_size, TINY_CONFIG["dim"])
     assert not torch.isnan(output).any()
 
 
 def test_encoder_normalize():
-    config = EncoderConfig(
+    model = _make_model(pooling_type="mean", normalize_embeddings=True)
-        **{**TINY_CONFIG, "pooling_type": "mean", "normalize_embeddings": True}
-    )
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model = EmbeddingEncoder(config).to(device=device)
     model.eval()
 
     batch_size, seq_len = 2, 8
     input_ids = torch.randint(
-        0, config.vocab_size, (batch_size, seq_len), device=device
+        0, TINY_CONFIG["vocab_size"], (batch_size, seq_len), device=_device
     )
 
     with torch.no_grad():
@@ -110,24 +81,19 @@ def test_encoder_normalize():
 
 
 def test_encoder_register():
-    from astrai.model.automodel import AutoModel
-
     assert AutoModel.is_registered("embedding")
     cls = AutoModel.get_component_class("embedding")
     assert cls is EmbeddingEncoder
 
 
 def test_encoder_from_transformer_checkpoint():
-    config = EncoderConfig(**TINY_CONFIG)
+    model = _make_model()
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model = EmbeddingEncoder(config).to(device=device)
-
     state_dict = model.state_dict()
     state_dict["lm_head.weight"] = torch.randn(
-        config.vocab_size, config.dim, device=device
+        TINY_CONFIG["vocab_size"], TINY_CONFIG["dim"], device=_device
     )
 
-    new_model = EmbeddingEncoder(config).to(device=device)
+    new_model = _make_model()
     new_model.load_state_dict(state_dict, strict=True)
 
     for key in model.state_dict():
@@ -135,12 +101,6 @@ def test_encoder_from_transformer_checkpoint():
 
 
 def test_encoder_save_load():
-    import json
-    import os
-    import tempfile
-
-    import safetensors.torch as st
-
     test_dir = tempfile.mkdtemp(prefix="encoder_test_")
     config_path = os.path.join(test_dir, "config.json")
     weights_path = os.path.join(test_dir, "model.safetensors")

tests/trainer/conftest.py

@@ -65,7 +65,7 @@ def create_train_config(
 
     def scheduler_fn(optim):
         return SchedulerFactory.create(
-            optim, "cosine", warmup_steps=10, lr_decay_steps=10, min_rate=0.05
+            "cosine", optim, warmup_steps=10, lr_decay_steps=10, min_rate=0.05
         )
 
     return TrainConfig(

tests/trainer/test_callbacks.py

@@ -102,7 +102,7 @@ def test_gradient_checkpointing_trainer_integration(base_test_env, random_datase
 
     def scheduler_fn(optim):
         return SchedulerFactory.create(
-            optim, "cosine", warmup_steps=10, lr_decay_steps=10, min_rate=0.05
+            "cosine", optim, warmup_steps=10, lr_decay_steps=10, min_rate=0.05
         )
 
     train_config = TrainConfig(
@@ -136,7 +136,7 @@ def test_callback_integration(base_test_env, random_dataset):
 
     def scheduler_fn(optim):
         return SchedulerFactory.create(
-            optim, "cosine", warmup_steps=10, lr_decay_steps=10, min_rate=0.05
+            "cosine", optim, warmup_steps=10, lr_decay_steps=10, min_rate=0.05
         )
 
     train_config = TrainConfig(

tests/trainer/test_early_stopping.py

@@ -16,7 +16,7 @@ def test_early_stopping_simulation(base_test_env, early_stopping_dataset):
 
     def scheduler_fn(optim):
         return SchedulerFactory.create(
-            optim, "cosine", warmup_steps=10, lr_decay_steps=10, min_rate=0.05
+            "cosine", optim, warmup_steps=10, lr_decay_steps=10, min_rate=0.05
         )
 
     train_config = TrainConfig(

tests/trainer/test_train_strategy.py

@@ -36,8 +36,8 @@ def test_schedule_factory_random_configs():
                 min_rate = params["min_rate"]
                 lr_decay_steps = total_steps - warmup_steps
                 scheduler = SchedulerFactory.create(
-                    optimizer,
                     schedule_type,
+                    optimizer,
                     warmup_steps=warmup_steps,
                     lr_decay_steps=lr_decay_steps,
                     min_rate=min_rate,
@@ -52,8 +52,8 @@ def test_schedule_factory_random_configs():
                 t_mult = params["t_mult"]
                 min_rate = params["min_rate"]
                 scheduler = SchedulerFactory.create(
-                    optimizer,
                     schedule_type,
+                    optimizer,
                     warmup_steps=warmup_steps,
                     cycle_length=cycle_length,
                     t_mult=t_mult,
@@ -103,8 +103,8 @@ def test_schedule_factory_edge_cases():
         min_rate = params["min_rate"]
         lr_decay_steps = total_steps - warmup_steps
         scheduler = SchedulerFactory.create(
-            optimizer,
             "cosine",
+            optimizer,
             warmup_steps=warmup_steps,
             lr_decay_steps=lr_decay_steps,
             min_rate=min_rate,
@@ -129,8 +129,8 @@ def test_schedule_factory_state_persistence():
     min_rate = 0.1
     lr_decay_steps = total_steps - warmup_steps
     scheduler = SchedulerFactory.create(
-        optimizer,
         "cosine",
+        optimizer,
         warmup_steps=warmup_steps,
         lr_decay_steps=lr_decay_steps,
         min_rate=min_rate,
@@ -146,8 +146,8 @@ def test_schedule_factory_state_persistence():
 
     # Create new scheduler with same parameters
     new_scheduler = SchedulerFactory.create(
-        optimizer,
         "cosine",
+        optimizer,
         warmup_steps=warmup_steps,
         lr_decay_steps=lr_decay_steps,
         min_rate=min_rate,