| """Explicit V4-C1 entry; --plan never starts CUDA, W&B or training.""" |
| import argparse |
| import json |
| from pathlib import Path |
| import sys |
| import tomllib |
|
|
| ROOT = Path(__file__).resolve().parents[2] |
| sys.path.insert(0, str(ROOT / "src")) |
|
|
|
|
| def main(): |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument("--config", type=Path, default=ROOT / "configs/posttrain-v4-c1.toml") |
| parser.add_argument("--resume", action="store_true") |
| parser.add_argument("--plan", action="store_true") |
| parser.add_argument("--offline", action="store_true", help="W&B local logs only; credentials never supplied as CLI args") |
| args = parser.parse_args() |
| config = tomllib.loads(args.config.read_text(encoding="utf-8")) |
| if args.plan: |
| from vimeml.posttraining.train import build_plan |
| plan = build_plan(config) |
| print(json.dumps({k: v for k, v in plan.items() if k not in {"code_sha256", "required_files"}}, indent=2)) |
| return |
| output = ROOT / config["output_dir"] |
| logs = ROOT / config["log_dir"] |
| if args.resume and not (output / "last.pt").exists(): |
| parser.error("No C1 last.pt checkpoint to resume") |
| if not args.resume and (output.exists() and any(output.iterdir()) or logs.exists() and any(logs.iterdir())): |
| parser.error("Fresh C1 output/log directories must be empty") |
| tracking = config.get("tracking", {}) |
| if not tracking.get("enabled", True): |
| from vimeml.posttraining.train import run |
| return run(config, resume=args.resume) |
| import wandb |
| directory = ROOT / "artifacts/tracking" |
| directory.mkdir(parents=True, exist_ok=True) |
| mode = "offline" if args.offline else tracking.get("mode", "online") |
| |
| |
| with wandb.init(project=tracking["project"], entity=tracking.get("entity"), name=output.name, |
| group=output.name, mode=mode, dir=str(directory), sync_tensorboard=True, |
| save_code=False, force=mode == "online", config=config, |
| settings=wandb.Settings(disable_git=True, init_timeout=30)) as cloud: |
| from vimeml.training.data import write_json |
| write_json(directory / f"{output.name}-live.json", {"id": cloud.id, "url": cloud.url, "resume": args.resume}) |
| print(f"W&B run: {cloud.url}" if cloud.url else "W&B offline", flush=True) |
| from vimeml.posttraining.train import run |
| result = run(config, resume=args.resume) |
| cloud.summary.update({"training_result": result}) |
| return result |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
