train.py

import argparse
import os
import sys
from typing import List

from datasets import load_dataset
from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
    DataCollatorForLanguageModeling,
    TrainingArguments,
    Trainer,
)

def parse_args():
    p = argparse.ArgumentParser()
    p.add_argument("--dataset", required=True, help="Path to a JSON/JSONL (or .gz) file.")
    p.add_argument("--output", default="trained_model", help="Folder to save the fine-tuned model.")
    p.add_argument("--model_name", default="distilgpt2", help="Base model.")
    p.add_argument("--epochs", type=float, default=1.0)
    p.add_argument("--batch_size", type=int, default=2)
    p.add_argument("--block_size", type=int, default=256)
    p.add_argument("--learning_rate", type=float, default=5e-5)
    return p.parse_args()

def main():
    args = parse_args()
    print(f"📥 Loading dataset: {args.dataset}", flush=True)
    ds = load_dataset("json", data_files=args.dataset, split="train")

    cols = ds.column_names
    print(f"🧾 Columns: {cols}", flush=True)

    print(f"🧠 Loading model & tokenizer: {args.model_name}", flush=True)
    tokenizer = AutoTokenizer.from_pretrained(args.model_name)
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token  # GPT-2 family has no pad token

    model = AutoModelForCausalLM.from_pretrained(args.model_name)

    def build_texts(batch) -> List[str]:
        if "text" in batch:
            return [str(t) for t in batch["text"]]
        if "prompt" in batch and "completion" in batch:
            # simple join: prompt + newline + completion
            return [f"{str(p).rstrip()}\n{str(c)}" for p, c in zip(batch["prompt"], batch["completion"])]
        raise ValueError("Dataset must contain 'text' OR both 'prompt' and 'completion' fields.")

    def tokenize(batch):
        texts = build_texts(batch)
        return tokenizer(
            texts,
            padding="max_length",
            truncation=True,
            max_length=args.block_size,
        )

    print("🔁 Tokenizing…", flush=True)
    tokenized = ds.map(
        tokenize,
        batched=True,
        remove_columns=cols,  # keep only tokenized fields
    )

    data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)

    print("⚙ Preparing Trainer…", flush=True)
    training_args = TrainingArguments(
        output_dir=args.output,
        overwrite_output_dir=True,
        per_device_train_batch_size=args.batch_size,
        num_train_epochs=args.epochs,
        learning_rate=args.learning_rate,
        logging_steps=10,
        save_steps=200,           # frequent-ish checkpoints (kept to 1)
        save_total_limit=1,
        report_to=[],
        gradient_accumulation_steps=1,
        fp16=False,               # CPU-friendly; enable if GPU has fp16
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized,
        tokenizer=tokenizer,
        data_collator=data_collator,
    )

    print("🚀 Training…", flush=True)
    trainer.train()

    print(f"💾 Saving to: {args.output}", flush=True)
    os.makedirs(args.output, exist_ok=True)
    trainer.save_model(args.output)
    tokenizer.save_pretrained(args.output)
    print("✅ Done.", flush=True)

if __name__ == "__main__":
    try:
        main()
    except Exception as e:
        # Make sure a failure returns non-zero so your app can detect it
        print(f"❌ Training failed: {e}", flush=True)
        sys.exit(1)