scripts/train_grpo.py

from __future__ import annotations

import os
import sys
import json
import argparse
from pathlib import Path

import requests
import torch
from datasets import Dataset
from trl import GRPOConfig, GRPOTrainer
from unsloth import FastLanguageModel, PatchFastRL

sys.path.insert(0, str(Path(__file__).resolve().parent))
from agent_prompt import SYSTEM_PROMPT, get_agent_prompt

PatchFastRL("GRPO", FastLanguageModel)

# --- Configuration ---
MODEL_NAME = os.getenv("MODEL_NAME", "meta-llama/Llama-3.2-3B-Instruct")
ENV_URL = os.getenv("ENV_URL", "http://localhost:8000")
OUTPUT_DIR = os.getenv("OUTPUT_DIR", "outputs/commitguard-llama-3b")
WANDB_PROJECT = os.getenv("WANDB_PROJECT", "commitguard")


# --- Reward: one reset + verdict per completion ---
def get_reward_from_env(prompts, completions, **kwargs) -> list[float]:
    rewards = []
    for prompt, completion in zip(prompts, completions):
        try:
            # Reset to get a fresh episode
            r = requests.post(f"{ENV_URL}/reset", json={}, timeout=10)
            if r.status_code != 200:
                rewards.append(-0.5)
                continue
            # Send the model's completion as the action
            text = completion[-1]["content"] if isinstance(completion, list) else str(completion)
            r = requests.post(f"{ENV_URL}/step", json={"action": text}, timeout=10)
            if r.status_code == 200:
                rewards.append(float(r.json().get("reward", 0.0)))
            else:
                rewards.append(-0.5)
        except Exception:
            rewards.append(-1.0)
    return rewards


def build_dataset(n_samples: int) -> Dataset:
    print(f"Fetching {n_samples} training prompts from {ENV_URL}...")
    samples = []
    for i in range(n_samples):
        try:
            r = requests.post(f"{ENV_URL}/reset", json={}, timeout=10)
            if r.status_code != 200:
                continue
            obs = r.json()["observation"]
            user_msg = get_agent_prompt(
                obs["diff"], obs["available_files"], obs.get("step_idx", 0)
            )
            samples.append({
                "prompt": [
                    {"role": "system", "content": SYSTEM_PROMPT},
                    {"role": "user", "content": user_msg},
                ],
            })
        except Exception:
            continue
        if (i + 1) % 50 == 0:
            print(f"  fetched {i + 1}/{n_samples}")
    print(f"Built dataset with {len(samples)} samples.")
    return Dataset.from_list(samples)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--samples", type=int, default=200)
    ap.add_argument("--max-steps", type=int, default=300)
    ap.add_argument("--save-steps", type=int, default=50)
    ap.add_argument("--num-generations", type=int, default=4)
    ap.add_argument("--batch-size", type=int, default=1)
    ap.add_argument("--grad-accum", type=int, default=4)
    ap.add_argument("--lr", type=float, default=5e-6)
    ap.add_argument("--no-wandb", action="store_true")
    args = ap.parse_args()

    # 1. Load Model
    print(f"Loading {MODEL_NAME} with Unsloth 4-bit...")
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name=MODEL_NAME,
        max_seq_length=2048,
        load_in_4bit=True,
        fast_inference=True,
        max_lora_rank=16,
    )

    model = FastLanguageModel.get_peft_model(
        model,
        r=8,
        target_modules=[
            "q_proj", "k_proj", "v_proj", "o_proj",
            "gate_proj", "up_proj", "down_proj",
        ],
        lora_alpha=16,
        lora_dropout=0,
        bias="none",
        use_gradient_checkpointing="unsloth",
        random_state=3407,
    )

    # 2. Build dataset from live env
    dataset = build_dataset(args.samples)

    # 3. GRPO config
    training_args = GRPOConfig(
        output_dir=OUTPUT_DIR,
        num_generations=args.num_generations,
        max_completion_length=512,
        per_device_train_batch_size=args.batch_size,
        gradient_accumulation_steps=args.grad_accum,
        learning_rate=args.lr,
        logging_steps=1,
        save_steps=args.save_steps,
        max_steps=args.max_steps,
        report_to="none" if args.no_wandb else "wandb",
        bf16=torch.cuda.is_bf16_supported(),
        fp16=not torch.cuda.is_bf16_supported(),
    )

    # 4. Train
    trainer = GRPOTrainer(
        model=model,
        processing_class=tokenizer,
        reward_funcs=[get_reward_from_env],
        args=training_args,
        train_dataset=dataset,
    )

    print("Starting GRPO training...")
    trainer.train()

    # 5. Save
    final_dir = f"{OUTPUT_DIR}/final"
    model.save_pretrained_merged(final_dir, tokenizer, save_method="lora")
    print(f"Training complete. LoRA adapter saved to {final_dir}")


if __name__ == "__main__":
    main()