Spaces:
Running
Running
File size: 1,973 Bytes
877add7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 | #!/usr/bin/env python3
"""Explicit TRL SFT entrypoint for small/scale profiles."""
from __future__ import annotations
import argparse
import json
from pathlib import Path
import sys
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
from app.training.sft_trl import SFTRunConfig, run_sft_trl
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Train SFT adapter with TRL + Unsloth.")
parser.add_argument("--model-id", default="Qwen/Qwen2.5-1.5B-Instruct")
parser.add_argument("--dataset-path", default="data/processed/sft_examples.json")
parser.add_argument("--output-dir", default="checkpoints")
parser.add_argument("--epochs", type=int, default=1)
parser.add_argument("--batch-size", type=int, default=2)
parser.add_argument("--max-steps", type=int, default=30)
parser.add_argument("--max-seq-len", type=int, default=1024)
parser.add_argument("--learning-rate", type=float, default=2e-5)
parser.add_argument("--use-unsloth", action="store_true")
parser.add_argument("--allow-fallback", action="store_true")
return parser.parse_args()
def main() -> None:
args = parse_args()
root = Path(__file__).resolve().parents[1]
cfg = SFTRunConfig(
model_id=args.model_id,
output_dir=root / args.output_dir,
dataset_path=root / args.dataset_path,
epochs=args.epochs,
batch_size=args.batch_size,
max_steps=args.max_steps,
max_seq_len=args.max_seq_len,
learning_rate=args.learning_rate,
use_unsloth=args.use_unsloth,
allow_fallback=args.allow_fallback,
)
result = run_sft_trl(cfg)
out = root / "outputs" / "reports"
out.mkdir(parents=True, exist_ok=True)
(out / "sft_trl_run.json").write_text(json.dumps(result, ensure_ascii=True, indent=2), encoding="utf-8")
print("sft_trl_done")
if __name__ == "__main__":
main()
|