refactor: migrate config system to Pydantic dataclasses

- Replace hand-rolled BaseConfig (from_dict/to_dict/_coerce/_unwrap_optional) with pydantic.dataclasses
- from_dict now uses cls(**d), to_dict uses dataclasses.asdict + json.dumps filter
- TrainConfig: required fields are now truly required (no default=None), delete manual validate()/__post_init__
- Remove dead required() helper and metadata={'help': ...} annotations
- Fix gradient_checkpointing_modules type from List[str] to List[type]
- Add pydantic>=2.0 as direct dependency in pyproject.toml
- Add numpy-style Parameters docstrings to all config classes
- Enable use_attribute_docstrings in BaseConfig for schema generation
- LoRAConfig also migrated to pydantic dataclass

astrai/config/preprocess_config.py

@@ -5,9 +5,11 @@ modes, both driven declaratively through ``input.sections`` or
 ``input.sources``.
 """
 
-from dataclasses import dataclass, field
+from dataclasses import field
 from typing import Dict, List, Optional
 
+from pydantic.dataclasses import dataclass
+
 from astrai.config.base import BaseConfig
 
 
@@ -25,6 +27,10 @@ class InputConfig(BaseConfig):
             "chosen": {"sections": [{"field": "chosen", ...}]},
             "rejected": {"sections": [{"field": "rejected", ...}]},
         }}}
+
+    Args:
+        sections (Optional[List[Dict]]): Section list for single-output mode. Defaults to None.
+        sources (Optional[Dict[str, Dict]]): Source map for multi-output mode, DPO/GRPO. Defaults to None.
     """
 
     sections: Optional[List[Dict]] = None
@@ -33,34 +39,17 @@ class InputConfig(BaseConfig):
 
 @dataclass
 class ProcessingConfig(BaseConfig):
-    """Processing configuration.
+    """Processing configuration for tokenization and packing.
 
-    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).
-    batch_size : int
-        Number of records tokenized together (default: 256).
-    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.
+    Args:
+        max_seq_len (int): Maximum sequence length. Defaults to 2048.
+        min_chars (int): Minimum number of characters to keep. Defaults to 50.
+        max_chars (int): Maximum number of characters to keep. Defaults to 2_000_000.
+        max_items (Optional[int]): Maximum number of items to process, None=unlimited. Defaults to None.
+        batch_size (int): Number of records tokenized together. Defaults to 256.
+        packing_strategy (str): How to pack sequences: 'simple', 'bfd', or 'bfd_split'. Defaults to "simple".
+        max_packed_len (int): Maximum length of a packed bin. Defaults to 8192.
+        truncation_mode (str): How to truncate over-length sequences: 'keep_start' or 'keep_end'. Defaults to "keep_start".
     """
 
     max_seq_len: int = 2048
@@ -75,24 +64,14 @@ class ProcessingConfig(BaseConfig):
 
 @dataclass
 class OutputConfig(BaseConfig):
-    """Output configuration.
+    """Output configuration for storage.
 
-    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).
+    Args:
+        domain_key (Optional[str]): Domain key for the output store. Defaults to None.
+        storage_format (str): Storage format: 'bin' or 'jsonl'. Defaults to "bin".
+        max_tokens_per_shard (int): Maximum tokens per shard before splitting. Defaults to 100_000_000.
+        dtype (Dict[str, str]): Per-key dtype overrides, e.g. {"input_ids": "int32"}. Defaults to {}.
+        position_ids_mode (str): Position ids mode: 'none', 'doc_reset', or 'continuous'. Defaults to "doc_reset".
     """
 
     domain_key: Optional[str] = None
@@ -104,6 +83,17 @@ class OutputConfig(BaseConfig):
 
 @dataclass
 class PipelineConfig(BaseConfig):
+    """Top-level preprocessing pipeline config.
+
+    Args:
+        version (int): Config schema version. Defaults to 1.
+        input (InputConfig): Input mapping config.
+        mask (Dict[str, str]): Per-field mask labels, e.g. {"system": "mask", "assistant": "train"}. Defaults to {}.
+        mask_default (str): Default mask label for unlisted fields. Defaults to "mask".
+        preprocessing (ProcessingConfig): Processing config.
+        output (OutputConfig): Output config.
+    """
+
     version: int = 1
     input: InputConfig = field(default_factory=InputConfig)
     mask: Dict[str, str] = field(default_factory=dict)