src/schemas/config.py

import json
from dataclasses import dataclass, fields
from pathlib import Path
from transformers import TrainingArguments

VALID_MODES = {"marker", "qa_m", "qa_b"}


@dataclass
class TrainingConfig:
    mode: str = "marker"
    data_path: str = "data/data_augmented_256.jsonl"
    output_dir: str = "models/"
    model_name: str = "distilbert-base-uncased"
    max_len: int = 256
    num_train_epochs: int = 5
    per_device_train_batch_size: int = 32
    per_device_eval_batch_size: int = 64
    gradient_accumulation_steps: int = 1
    learning_rate: float = 2e-5
    warmup_ratio: float = 0.1
    weight_decay: float = 0.01
    val_split: float = 0.1
    test_split: float = 0.1
    early_stopping_patience: int = 3
    fp16: bool = True
    seed: int = 42
    logging_steps: int = 50
    save_total_limit: int = 2
    loss_fn: str = "cross_entropy"
    focal_gamma: float = 2.0

    def __post_init__(self):
        if self.mode not in VALID_MODES:
            raise ValueError(f"mode must be one of {VALID_MODES}, got '{self.mode}'")
        if self.loss_fn not in ("cross_entropy", "focal"):
            raise ValueError(f"loss_fn must be 'cross_entropy' or 'focal', got '{self.loss_fn}'")

    @classmethod
    def from_json(cls, path: str | Path) -> "TrainingConfig":
        with open(path, "r", encoding="utf-8") as f:
            raw = json.load(f)
        known = {f.name for f in fields(cls)}
        unknown = set(raw.keys()) - known
        if unknown:
            print(f"Warning: unknown config keys ignored: {unknown}")
        filtered = {k: v for k, v in raw.items() if k in known}
        return cls(**filtered)

    def to_training_arguments(self) -> TrainingArguments:
        return TrainingArguments(
            output_dir=self.output_dir,
            num_train_epochs=self.num_train_epochs,
            per_device_train_batch_size=self.per_device_train_batch_size,
            per_device_eval_batch_size=self.per_device_eval_batch_size,
            gradient_accumulation_steps=self.gradient_accumulation_steps,
            learning_rate=self.learning_rate,
            warmup_ratio=self.warmup_ratio,
            weight_decay=self.weight_decay,
            eval_strategy="epoch",
            save_strategy="epoch",
            load_best_model_at_end=True,
            metric_for_best_model="macro_f1",
            greater_is_better=True,
            save_total_limit=self.save_total_limit,
            logging_dir=f"{self.output_dir}/logs",
            logging_steps=self.logging_steps,
            report_to="none",
            fp16=self.fp16,
            seed=self.seed,
            data_seed=self.seed,
        )