step_train/scripts_train/10_run_dapo.py

"""Step 10: Run DAPO for the shared Capability LoRA.

Single GPU:
    python scripts/train/10_run_dapo.py
    python scripts/train/10_run_dapo.py --config configs/train/dapo.yaml

8-GPU data parallel (TensorBoard auto-detected via $TENSORBOARD_LOG_PATH):
    torchrun --standalone --nproc_per_node=8 scripts/train/10_run_dapo.py
    # or use the launcher: bash scripts/train/run_dapo_8gpu.sh
"""

import argparse
import os
import sys

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
os.environ.setdefault("CARTRIDGES_DIR", os.path.join(PROJECT_ROOT, "cartridges-lib"))
os.environ.setdefault("CARTRIDGES_OUTPUT_DIR", os.path.join(PROJECT_ROOT, "checkpoints/cartridge"))

from src.train.rl.dapo_trainer import DAPOTrainer
from src.utils import cleanup_distributed, load_yaml, set_seed, setup_distributed, setup_logger

logger = setup_logger(__name__)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--config", default="configs/train/dapo.yaml")
    ap.add_argument("--epochs", type=int, default=None)
    ap.add_argument("--batch-anchors", type=int, default=None)
    ap.add_argument("--output-dir", default=None)
    ap.add_argument("--resume", action="store_true",
                    help="continue from output_dir/latest.txt (policy LoRA + optimizer + step)")
    args = ap.parse_args()

    cfg_path = args.config if os.path.isabs(args.config) else os.path.join(PROJECT_ROOT, args.config)
    cfg = load_yaml(cfg_path)
    if args.epochs is not None:
        cfg["epochs"] = args.epochs
    if args.batch_anchors is not None:
        cfg["batch_anchors"] = args.batch_anchors
    if args.output_dir is not None:
        cfg["output_dir"] = args.output_dir
    if args.resume:
        cfg["resume"] = True

    dist_ctx = setup_distributed()
    set_seed(cfg.get("seed", 42))
    if dist_ctx["is_distributed"]:
        logger.info(f"Distributed: rank {dist_ctx['rank']}/{dist_ctx['world_size']} "
                    f"local_rank={dist_ctx['local_rank']}")

    try:
        trainer = DAPOTrainer(cfg, PROJECT_ROOT)
        logger.info(f"Loaded {len(trainer.anchors)} anchors; K={trainer.K}, batch={trainer.batch_anchors}")
        trainer.run()
    finally:
        cleanup_distributed()


if __name__ == "__main__":
    main()