ViperEkura
/
AstrAI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
525 Commits 1 Branch 9 Tags 4.1 MiB
Commit Graph

95 Commits

Author SHA1 Message Date
ViperEkura c17aa0dc54 fix: eval script bugs and add missing features 
- evaluate_mmlu: fix double few-shot injection (build_prompt no longer
  adds few-shot, apply_chat handles it once)
- evaluate_humaneval: fix pass@k k-filtering to be per-problem instead
  of using first problem's n globally; reuse ProcessPoolExecutor across
  problems; fix closure UnboundLocalError in test_one; handle None in
  report when k > n
- evaluate_ifd: remove dead code (score_plain/score_messages); add
  multi-file/directory input support with --input_path/--output_dir;
  add summary.json aggregation and --max_samples; add --dtype flag
- evaluate_ppl: add --device and --dtype flags (was hardcoded to cuda)
- evaluate_ifeval: fix docstring path (scripts/tools -> scripts/eval)
- analyze_weights: add --output JSON export; fix dead code filter
  ("_norm" not in r was always True)
 2026-07-17 14:02:58 +08:00
ViperEkura b12b24eadc feat: rewrite evaluate_ppl with token-level loss and multi-file support 
- Support multiple input files, glob patterns, and directory input
- Add --token_level flag: per-record token_ids + log_probs JSONL output
- Add --max_samples for random subsampling per file
- LossAccumulator: streaming mode (histogram-based percentiles, low memory) vs exact mode (full token list)
- Token type analysis (ascii/cjk/non_ascii/special) when token_level=True
- Fix token_ids/log_probs alignment (shift offset)
- Cache frozenset(stop_ids) outside loop for performance
- Aggregate stats: mean/median/ppl/p50/p90/p95/p99
- Summary JSON with all datasets in one file
 2026-07-17 13:14:15 +08:00
ViperEkura 84ed2327f5 feat: add --resume flag to decouple weight loading from training resumption 
- Add --resume bool flag to train.py CLI
- --param_path always loads weights only by default
- --resume restores epoch, consumed_samples, optimizer & scheduler
- Checkpoint.load() now preserves full meta dict
- Update test_early_stopping to use new param_path/resume API
 2026-07-16 14:23:23 +08:00
ViperEkura bb175fda91 fix: resume optimizer LR, step display, and consumed_samples alignment 2026-07-15 08:59:52 +08:00
ViperEkura 2c7a71a9c0 refactor: separate old policy and ref model in GRPO strategy 
- Split single ref_model into old_model (importance sampling ratio) and ref_model (frozen KL regularizer)
- Move ref_model/old_model creation from strategy __init__ to TrainContextBuilder, pass as explicit parameters
- Remove periodic sync_ref_model + sync_interval; add sync_old_model for external rollout loop to call
- DPOStrategy also receives ref_model from builder
- Fix std to use unbiased=False (population std per GRPO paper)
- Remove redundant tests (test_grpo_kl_zero_at_init, test_grpo_no_sync_interval_param)
- Remove --grpo_sync_interval CLI arg
 2026-07-14 20:03:45 +08:00
ViperEkura c8567a6f65 fix: exclude embedding, lm_head, bias, and norm params from Muon optimizer, use AdamW 2026-07-08 19:42:20 +08:00
ViperEkura 8035be9b1f fix: make MuonMix inherit from torch.optim.Optimizer 2026-07-08 17:03:05 +08:00
ViperEkura c50adbaac0 feat : replace AdamW with MuonMix (Muon + AdamW) optimizer 
- Muon for 2D matrix params, AdamW for 1D (norm/bias/embed)
- MuonMix wrapper handles combined step/zero_grad/state_dict
- New CLI args: weight_decay, muon_momentum, muon_nesterov, muon_ns_steps, muon_adjust_lr
- Removed adamw_beta1/adamw_beta2/adamw_weight_decay
- Moved optimizer/strategy params from signature to **kwargs
 2026-07-07 14:10:36 +08:00
ViperEkura f0cd0134c6 fix : update benchmark for v1.3.8 cache API, add argparse and cache type switch 
- Replaced old KVCage API with PageCache/ContiguousCache
- Added --cache contiguous|paged switch for decoding comparison
- Added argparse for all params (batch/prompt/gen/device/dtype)
- Fixed PageCache decode crash by extending pages for full sequence
 2026-07-05 20:30:26 +08:00
ViperEkura db9b39b084 fix: resolve IFD token-set asymmetry and support single-token answers 
- Sentinel-anchored unconditional pass: both branches now predict the same N response tokens
- Single-token responses (rl=1) fully supported
- ctx_len tracked per sample; skip_reason replaces silent None
- --per_token flag for per-token IFD breakdown
 2026-07-05 17:48:26 +08:00
ViperEkura 599a51f4f7 fix: reliable test timeout, separate generate/test phases, dynamic pass@k 
- Replace SIGALRM+exec() with subprocess.run(timeout=) for test execution
- Add --test_only flag to skip generation and test existing completions
- Add --generate_only flag for generation-only runs
- Derive pass@k values from num_samples (filter k > n)
- Support loading completions from array JSON (not just JSONL)
 2026-07-05 08:47:30 +08:00
ViperEkura 17d6eaa2f2 refactor: rewrite humaneval evaluation with functional pipeline design 
- fix KeyError race condition in inference cache touch()
- EvalConfig dataclass for centralized configuration
- load->generate->extract->test->score->report pipeline
- two-phase generation+testing for max GPU utilization
- signal-based SIGALRM timeout protection for code exec
- suppress subprocess stdout/stderr pollution
 2026-07-05 07:58:28 +08:00
ViperEkura 2d908639e9 feat : add ROUGE evaluation script (manual impl, no deps) 
- ROUGE-1/2 via n-gram overlap (Counter)
- ROUGE-L via LCS (DP)
- CLI: python scripts/eval/evaluate_rouge.py --data_path ... --output ...
- Library: compute_rouge(ref, cand) -> dict of precision/recall/f1
 2026-07-05 01:15:01 +08:00
ViperEkura c7158418dd perf: add BFD bin-packing and custom attention mask to IFD batch scoring 2026-07-04 18:58:13 +08:00
ViperEkura 4d3c9341c1 refactor: rewrite IFD evaluation with clean three-layer architecture 2026-07-04 18:33:51 +08:00
ViperEkura 1adca39cd8 fix: handle long sequences and optimize IFD computation 2026-07-04 08:35:45 +08:00
ViperEkura 204873fa2f fix: handle long sequences and optimize IFD computation 2026-07-04 08:23:32 +08:00
ViperEkura a5c1de6b1b feat: add model_path temperature top_p top_k max_tokens system_prompt args to stream_chat 2026-07-04 07:33:32 +08:00
ViperEkura 70c0e5de90 refactor: merge validation into MetricCallback, simplify progress bar to optimizer steps 
- Remove separate ValidationCallback, merge into MetricCallback
- Progress bar now tracks optimizer steps instead of micro-steps
- Remove unused log_interval config field and CLI flag
- Fix validation all_reduce: use SUM(loss, count) instead of AVG
- Simplify metric logging: always log every optimizer step
- Add grad_norm display to progress bar
 2026-07-03 21:43:08 +08:00
ViperEkura aabb0d83e9 refactor : replace iteration with consumed_samples 
- Replace context.iteration with consumed_samples (global sample count)
- Add optimizer_step property derived from consumed_samples
- Checkpoint meta.json stores consumed_samples, drops iteration
- CLI --start_batch renamed to --start_samples (per-rank samples)
- Checkpoint dir naming: epoch_X_step_Y instead of epoch_X_iter_Y
- Metric log entries use step and consumed_samples fields
- Backward compat removed (old iteration checkpoints unsupported)
 2026-06-30 18:42:42 +08:00
ViperEkura 44579ea6dc refactor : metric 日志改为以 optimizer step 为单位，默认每步记录 
- log_interval 默认 100 -> 1，语义从 batch iteration 改为 optimizer step
- step 指标从 on_batch_end 移到 on_optimizer_step，不受梯度累积影响
- JSONL 条目新增 step 字段，保留 iter
- flush 落盘仍在 on_batch_end
 2026-06-30 15:12:31 +08:00
ViperEkura 0f1fcb079f refactor : grad_norm 指标简化，clip_grad_norm 移至 executor 
- metrics 默认加入 grad_norm，移除 grad_std/max/min/mean/nan_num
- grad_norm 默认返回总 L2 范数，per_param=True 返回各参数范数
- clip_grad_norm 从 callback 移至 BaseExecutor/FSDPExecutor
- FSDPExecutor 覆盖为 model.clip_grad_norm_() 保证分布式正确
- ctx_get_grad_norm 改为读取 context.grad_norm
 2026-06-30 14:59:43 +08:00
ViperEkura 84d4769163 feat: SVD 有效秩/权重统计分析脚本 2026-06-29 21:39:22 +08:00
ViperEkura bf09a35c95 feat: optimizer 参数分组，bias/norm 不做 weight decay 2026-06-27 16:30:34 +08:00
ViperEkura b4587c5d08 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
 2026-06-25 17:18:20 +08:00
ViperEkura 01d2da2893 feat : 训练支持 --schedule_type 及对应调度器参数 
- --schedule_type 可选 cosine/sgdr/wsd，默认 cosine
- --min_rate 统一控制最小 LR 比率
- --cycle_length / --t_mult 用于 sgdr
- --stable_steps / --decay_steps 用于 wsd，自动计算默认值
 2026-06-22 10:35:56 +08:00
ViperEkura 39985840c7 refactor : neftune_alpha 在 Embedding 构造时传入，由模型配置链路负责 
- BaseModelConfig 添加 neftune_alpha 字段 (默认 0.0)
- Embedding.__init__ 接受 neftune_alpha 参数，不再外部 set
- AutoRegressiveLM / EmbeddingEncoder 从 config 传入 neftune_alpha
- train.py 将 CLI 参数注入 config 后再创建模型
- TrainContextBuilder 移除 neftune 设置（不再是其职责）
 2026-06-19 14:23:27 +08:00
ViperEkura 7348bac6ab fix: 规范 generate.py 命令行接口 
- generate.py 清理描述文字，help 统一标注默认值
- max_tokens 默认改为 None，回退 model config max_len
- evaluate_ppl.py 同步清理描述文字
- params.md 同步 max_tokens 默认值
 2026-06-19 14:03:02 +08:00
ViperEkura 376e9eba80 feat: IFEval 使用 chat template 格式化 prompt，添加 model.eval() 
- generate_one 用 tokenizer.apply_chat_template 包 user 消息
- 新增 model.eval() 关闭 dropout，确保确定性输出
 2026-06-18 16:45:16 +08:00
ViperEkura a62c2e11a2 feat: IFD 默认使用 chat template，支持裸文本模式 
- 新增 _compute_ifd_with_template，用 tokenizer chat template 格式化后计算 IFD
- 默认开启 chat template，可通过 --no_chat_template 切换回裸拼接
- chat template 缺失时给出 RuntimeError 提示
 2026-06-18 16:35:05 +08:00
ViperEkura 1818d06576 feat: 新增 IFD 数据质量评分工具, 移动 ppl 至 eval 
- 计算指令遵循难度分数用于数据筛选
- IFD = 条件交叉熵 / 无条件交叉熵
- perplexity 移至 scripts/eval/
 2026-06-16 22:03:45 +08:00
ViperEkura 4e8d1ee24e feat: 新增 IFEval 指令遵循评测 
- 实现 25 种正则约束 verifier
- 将评测脚本从 scripts/tools/ 移至 scripts/eval/
 2026-06-16 21:57:34 +08:00
ViperEkura 445378667f feat : NEFTune 噪声注入 + label_smoothing 默认值修正 
- Embedding.forward 训练时注入 randn 噪声，缩放系数 neftune_noise_alpha / sqrt(seq_len)
- TrainConfig.neftune_alpha 通过 config 传递（默认 0=关闭）
- TrainContextBuilder 将 config.neftune_alpha 写入 embed_tokens
- --neftune_alpha CLI 参数（典型值 5.0）
- label_smoothing 默认值 0.05 -> 0.0
 2026-06-11 15:32:43 +08:00
ViperEkura 6ae1828449 refactor : 清理工厂和配置系统中的死代码与冗余抽象 
- 删除 Registry 中未使用的 category/priority 字段，_entries 简化为直接存储类引用
- 修正 __init_subclass__ 避免叶子类（AutoRegressiveLM 等）创建空注册表
- 删除 5 个工厂的薄 create() 覆写，统一使用 BaseFactory.create(name, *args, **kwargs)
- 删除 3 处零调用的 available_types/available_strategies 别名死代码
- 删除零调用的 BaseModelConfig.to_file 死代码
- 将 BaseConfig.from_json/to_json 重命名为 from_file/to_file，消除与子类重复
- 移除两个 inference builder 中总是被覆写的 prompt_tokens=0
 2026-06-07 11:39:50 +08:00
ViperEkura 5e73ca20aa feat : train CLI 新增 val_split/val_step/metrics/log 参数 
- --val_split 从训练集按比例切分验证集
- --val_step 控制验证间隔 optimizer step 数
- --metrics 自定义日志指标列表，默认 loss lr
- --log_dir / --log_interval 控制日志输出目录和频率
 2026-06-03 14:31:22 +08:00
ViperEkura 438dc10391 fix : MMLU eval 使用 chat template 格式匹配 SFT 训练数据 
- 原 prompt 为纯文本格式，与 SFT chat template 不匹配导致模型输出随机
- 新增 apply_chat() 将 MMLU prompt 包装为 user/assistant 对话格式
- choice_text 改为单字母（去掉空格前缀）适配模板输出
- 5-shot 时 few-shot 示例作为独立 user/assistant 轮次插入
 2026-06-03 11:59:42 +08:00
ViperEkura 615ba5d8ef feat : 新增 HumanEval pass@k 代码生成评测 
- InferenceEngine.generate() 批量生成 n 个补全
- 正则提取函数体 + 停止符截断
- multiprocessing sandbox 执行 + timeout 保护
- 标准无偏 pass@k 公式 (1, 10, 100)
 2026-06-03 10:52:32 +08:00
ViperEkura 9b416c1bbb refactor : 并行启动 Strategy 模式重构，local_rank 解耦 
- setup_parallel 接收 local_rank 参数，不再读环境变量推导
- TorchrunStrategy 从 env 读取 LOCAL_RANK，LocalStrategy 用 rank
- _detect_launcher() 分级检测替代内联 RANK 检查
- _run_single_rank 统一入口，消除 _run_single/_run_multi 重复
- 优雅退出：except BaseException 终止子进程并 re-join
- gradient_checkpointing_modules 判定提取到外部变量
 2026-06-02 11:22:24 +08:00
yegroup001 0deee48602 feat : 训练脚本新增 gradient_checkpointing 与多机 DDP 参数 2026-06-02 01:01:00 +08:00
ViperEkura 69207e2c57 refactor : 基于声明式 JSON 配置的预处理管线重构 
- 用工厂注册的 MaskBuilder（chat/instruction/text）替换硬编码的 _transform_* 方法
- mask 规则以 role-to-action 映射声明在配置中，与 chat_template 完全解耦
- 单次编码 + role-span 追踪替代两次编码 + 长度差计算 mask 的方式
- 支持多轮对话训练：所有 assistant 轮次参与训练，而非仅最后一轮
- 新建 astrai.preprocessing 包（builder.py + pipeline.py），删除 astrai/preprocess.py
- CLI 精简为 --config 参数，所有参数通过 PipelineConfig JSON 配置
- 新增 PipelineConfig、InputConfig、ProcessingConfig、OutputConfig dataclass
- 文档：assets/docs/preprocessing.md
- 27 个测试覆盖 mask builder、pipeline、配置序列化、工厂注册
 2026-05-30 20:45:09 +08:00
ViperEkura 138c5bcc08 feat : 添加 JSONL 预处理管线 
- Pipeline 模板, Reader 加 transform 加 Writer 可组合
- 自动检测 JSONL 格式, 支持 messages 文本 prompt 加 response 三种
- chat 数据通过 apply_chat_template 适配, 自动生成 loss_mask
- 输出对齐 Store 和 DatasetFactory, 直接用于训练
- 默认 bin 格式, CLI 入口 scripts/tools/preprocess.py
 2026-05-30 17:12:42 +08:00
ViperEkura a923e0a23a fix : 修复 MMLU 评测脚本数据源和依赖 
- 数据源改为 Berkeley data.tar（GitHub zip 不含数据文件）
- urllib 替换为 requests，支持代理下载
- zip 解压替换为 tar，增加目录 flatten 逻辑
- 添加 model.eval() 确保推理模式正确
 2026-05-30 16:51:24 +08:00
ViperEkura 2d5dc93b3d fix : 修正类型标注与统一 CLI 参数命名 
- AutoRegressiveLM.forward 返回类型标注 -> Dict[str, Tensor]
- EmbeddingEncoder 移除冗余 position_ids 自动创建
- CLI 脚本模型目录参数统一为 --param_path
 2026-05-27 20:49:44 +08:00
ViperEkura 4145d35e3c refactor: 检查点加载重构，路径替代对象传递 
- model: nn.Module -> model_fn 工厂函数，spawn 边界只传字符串
- Trainer.train(resume_dir=path) — Checkpoint 不再通过 pickle 传递
- TrainContextBuilder.with_resume_dir(path) — 自动检测 meta.json 分流 resume/from-scratch
- CheckpointCallback: 拆分 state_dict 收集（全 rank）与磁盘写入（rank-0），修复 FSDP 死锁
- serialization: load_torch 支持 broadcast，消除 _load_extra/_load_torch_broadcast
- optimizer/scheduler 恢复逻辑内联到 build()，在 executor.prepare() 之后执行
- pyproject.toml: ruff exclude build/ 避免 CI 扫描构建产物
 2026-05-27 20:15:29 +08:00
ViperEkura 34c6c45bd6 feat: 初步实现 MMLU 评测脚本 
- 支持 few-shot (log-likelihood ranking) 与 zero-shot
- 自动下载 Hendrycks MMLU 数据集
- --device / --dtype 可配置，默认 GPU bf16
 2026-05-26 20:23:31 +08:00
ViperEkura e9def84ce7 fix : perplexity.py left padding 导致 batch>1 时 PPL 计算错误 2026-05-26 19:59:57 +08:00
ViperEkura 836e02a166 docs: 同步 architecture/inference/training 文档至实际代码，CLI 补充 fsdp 选项 
- 修正 ProtocolHandler 架构：concrete + ResponseBuilder(ABC) 策略模式
- 修正训练循环 scheduler.step() 在 sync_gradients 块内
- 修正组合/聚合关系：注入组件改为 o--，删除不持有引用的关联
- --parallel_mode CLI choices 加入 fsdp
- nprocs > 1 且 parallel_mode=none 时 raise error
 2026-05-26 19:37:00 +08:00
ViperEkura 1d26aa2e93 fix: 禁用DDP static_graph避免PyTorch 2.7.1下no_sync与backward冲突 
- static_graph=True时DDP.no_sync() + loss.backward()触发expect_autograd_hooks_内部断言
- PyTorch 2.7.1中no_sync上下文切换与静态图hook状态管理存在兼容性bug
- 将static_graph设为False恢复梯度累积正常执行
- find_unused_parameters保持False（模型无不参与计算的参数）
 2026-05-26 15:08:01 +08:00
ViperEkura a548d4553e fix: 断点续训恢复优化器/调度器状态及采样器剩余长度 
- 使用Checkpoint.load()替代手动加载model.safetensors，恢复optimizer/scheduler状态
- TrainContextBuilder从checkpoint.extra恢复优化器和调度器state_dict
- ResumableDistributedSampler.__len__返回剩余样本数而非总数
- 训练前对state_dict置空避免mp.spawn pickle 7GB大对象
 2026-05-26 13:50:25 +08:00
ViperEkura 3ab4f237e5 refactor: 重构训练后端为 Executor 模式 
- backend.py → executor.py，BaseTrainingBackend → BaseExecutor
- 新增 NoneExecutor（单卡）和 DDPExecutor（DDP，world_size=1 自动降级）
- 新增 GradientState 分离梯度同步状态，AccumOptimizer/AccumScheduler 包裹拦截
- 新增 astrai/protocols.py：OptimizerProtocol/SchedulerProtocol 结构子类型
- TrainContext.backend → executor，TrainConfig 移除 parallel_wrapper/state_dict_fn，新增 parallel_mode/executor_kwargs
- 训练循环用 accumulate() 包裹，on_optimizer_step 命名约定=gate
- scripts/tools/train.py 移除 ddp_wrap/prepare_checkpoint，新增 --parallel_mode
 2026-05-24 20:35:44 +08:00