scripts/convert_rl_to_sft_bandit.py

#!/usr/bin/env python3
import argparse
import json
import re
from pathlib import Path
from typing import List, Tuple, Optional

try:
    import yaml  # type: ignore
except Exception:
    yaml = None

DEFAULT_BANDIT_INSTRUCTION = (
    """Turn 1:
State:
You are playing a bandit game. Goal: Maximize your total reward by choosing which arm to pull. 
Game Rules: 
""")


def load_env_instruction_and_cfg(repo_root: Path) -> Tuple[str, int, bool]:
    instruction = DEFAULT_BANDIT_INSTRUCTION
    max_tokens = 100
    enable_think = True

    if yaml is None:
        return instruction, max_tokens, enable_think

    envs_yaml = repo_root / "config" / "envs.yaml"
    if envs_yaml.exists():
        try:
            with open(envs_yaml, "r", encoding="utf-8") as f:
                envs = yaml.safe_load(f)
            custom_envs = envs.get("custom_envs", {}) if isinstance(envs, dict) else {}
            if isinstance(custom_envs, dict):
                # Prefer Bandit, fallback to BanditTest
                for key in ["Bandit", "BanditTest"]:
                    if key in custom_envs:
                        cfg = custom_envs[key]
                        instruction = cfg.get("env_instruction", instruction) or instruction
                        max_tokens = int(cfg.get("max_tokens", max_tokens))
                        break
        except Exception:
            pass

    base_yaml = repo_root / "config" / "base.yaml"
    if base_yaml.exists():
        try:
            with open(base_yaml, "r", encoding="utf-8") as f:
                base_cfg = yaml.safe_load(f)
            ap = base_cfg.get("agent_proxy", {}) if isinstance(base_cfg, dict) else {}
            enable_think = bool(ap.get("enable_think", enable_think))
        except Exception:
            pass

    return instruction, max_tokens, enable_think


def parse_names_from_text(text: str) -> Optional[Tuple[str, str]]:
    # Heuristic similar to BanditWrapper: find segment after "named " and split by " and "
    try:
        anchor = "named "
        if anchor in text:
            segment = text.split(anchor, 1)[1]
            segment = segment.split("\n", 1)[0]
            parts = segment.split(" and ")
            if len(parts) >= 2:
                name_a = parts[0].strip().strip(' .!?,')
                name_b = parts[1].strip().strip(' .!?,')
                if name_a and name_b:
                    return name_a, name_b
    except Exception:
        pass
    # Secondary regex attempt: capture two capitalized tokens joined by and
    m = re.search(r"([A-Za-z][\w]*)\s+and\s+([A-Za-z][\w]*)", text)
    if m:
        return m.group(1), m.group(2)
    return None


def build_messages_for_episode(
    names: Tuple[str, str],
    actions: List[int],
    rewards: List[float],
    instruction: str,
    max_tokens: int,
    enable_think: bool,
) -> List[dict]:
    name_a, name_b = names
    # Enrich instruction with the two arm names so the SFT sample is self-contained
    enriched_instr = (
        f"{instruction}\n"
        f"1. There are 2 arms, named {name_a} and {name_b}\n"
        f"""2. Each arm has its own reward distribution, related to their names. 
3. Analyze the symbolic meaning of each arm's name to guess how their reward distribution might behave.\n"""
        f"4. Based on the symbolic meaning of their names, which arm do you think is more likely to give higher rewards on average? Choose between {name_a} and {name_b}, and output like <answer> {name_a} </answer> or <answer> {name_b} </answer>.\n"
    )
    messages = [
        {"role": "system", "content": "You're a helpful assistant. "},
        {"role": "user", "content": enriched_instr},
    ]

    # Bandit is single-step in our setup; still handle lists robustly
    turns = max(len(actions), 1)
    for t in range(turns):
        # Only one turn state block; reiterate arms for clarity
        length_prompt = f"Max response length: {max_tokens} words (tokens)."
        fmt = "<answer> [your answer] </answer>" if not enable_think else "<think> [Your thoughts] </think> <answer> [your answer] </answer>"
        messages[-1]["content"] += (
            f"\nYou have 1 action left. Always output: {fmt} with no extra text. Strictly follow this format. {length_prompt}"
        )
        if t < len(actions):
            act = int(actions[t])
            # RL action space is {0,1}; env expects {1,2}, but for chat we emit the chosen name
            chosen = name_a if act == 0 else name_b
            assistant_text = f"<answer>{chosen}</answer>" if not enable_think else f"<think></think><answer>{chosen}</answer>"
            messages.append({"role": "assistant", "content": assistant_text})
            r = rewards[t] if t < len(rewards) else 0.0
            messages.append({"role": "user", "content": f"Reward:\n{r}\n"})

    return messages[:-1]


def convert_file(step_dir: Path, output_dir: Path, repo_root: Path, include_failed: bool = False) -> Path:
    traj_path = step_dir / "trajectories.jsonl"
    metrics_path = step_dir / "metrics.json"
    if not traj_path.exists():
        raise FileNotFoundError(f"Missing trajectories.jsonl at {traj_path}")

    instruction, max_tokens, enable_think = load_env_instruction_and_cfg(repo_root)

    output_dir.mkdir(parents=True, exist_ok=True)
    out_path = output_dir / f"{step_dir.name}_sft.jsonl"

    global_step = None
    if metrics_path.exists():
        try:
            with open(metrics_path, "r", encoding="utf-8") as f:
                m = json.load(f)
            global_step = m.get("global_step")
        except Exception:
            pass

    written = 0
    with open(traj_path, "r", encoding="utf-8") as fin, open(out_path, "w", encoding="utf-8") as fout:
        for line in fin:
            line = line.strip()
            if not line:
                continue
            traj = json.loads(line)
            ep_success = bool(traj.get("episode_success", False))
            if (not include_failed) and (not ep_success):
                continue

            # Try to obtain the two arm names from the trajectory
            names_tuple: Optional[Tuple[str, str]] = None
            if isinstance(traj.get("prompt"), str):
                names_tuple = parse_names_from_text(traj["prompt"])  # if prompt field exists
            if names_tuple is None and isinstance(traj.get("states_text"), list) and len(traj["states_text"]) > 0:
                names_tuple = parse_names_from_text(traj["states_text"][0])
            if names_tuple is None and isinstance(traj.get("names"), list) and len(traj["names"]) >= 2:
                names_tuple = (str(traj["names"][0]), str(traj["names"][1]))
            if names_tuple is None:
                # Fallback placeholders; this loses semantic meaning but still yields a valid SFT record
                names_tuple = ("ArmA", "ArmB")

            actions = traj.get("actions", []) or []
            rewards = traj.get("rewards", []) or []

            messages = build_messages_for_episode(
                names=names_tuple,
                actions=actions,
                rewards=rewards,
                instruction=instruction,
                max_tokens=max_tokens,
                enable_think=enable_think,
            )

            record = {
                "messages": messages,
                "meta": {
                    "episode_return": traj.get("episode_return", None),
                    "episode_success": ep_success,
                    "global_step": global_step,
                },
            }
            fout.write(json.dumps(record, ensure_ascii=False) + "\n")
            written += 1

    if written == 0:
        with open(out_path, "w", encoding="utf-8") as f:
            pass
    return out_path


def find_latest_step_dir(traj_root: Path) -> Path:
    step_dirs = [p for p in traj_root.iterdir() if p.is_dir() and p.name.startswith("step_")]
    if not step_dirs:
        raise FileNotFoundError(f"No step_* directories under {traj_root}")
    step_dirs.sort(key=lambda p: int(p.name.split("_")[-1]))
    return step_dirs[-1]


def main():
    parser = argparse.ArgumentParser(description="Convert Bandit RL trajectories to LLM SFT chat JSONL")
    parser.add_argument("run_dir", help="Path to the run directory (contains trajectories/)")
    parser.add_argument("--step", default=None, help="Specific step directory name (e.g., step_10000)")
    parser.add_argument("--include_failed", action="store_true", help="Include failed episodes in SFT data")
    args = parser.parse_args()

    repo_root = Path(__file__).resolve().parents[1]
    run_dir = Path(args.run_dir)
    traj_root = run_dir / "trajectories"
    if not traj_root.exists():
        raise FileNotFoundError(f"Not found trajectories directory: {traj_root}")

    step_dir = traj_root / args.step if args.step else find_latest_step_dir(traj_root)
    output_dir = run_dir / "sft"
    out_path = convert_file(step_dir=step_dir, output_dir=output_dir, repo_root=repo_root, include_failed=args.include_failed)
    print(f"SFT data written to: {out_path}")


if __name__ == "__main__":
    main()