Upload kaggle/run_exp3_multi.py with huggingface_hub
Browse files- kaggle/run_exp3_multi.py +63 -0
kaggle/run_exp3_multi.py
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
================================================================================
|
| 4 |
+
LITM v4 — Experiment 3: Multi-Needle Retrieval
|
| 5 |
+
Standalone Kaggle-ready script.
|
| 6 |
+
================================================================================
|
| 7 |
+
"""
|
| 8 |
+
import argparse
|
| 9 |
+
import logging
|
| 10 |
+
import os
|
| 11 |
+
import sys
|
| 12 |
+
|
| 13 |
+
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
| 14 |
+
|
| 15 |
+
from experiments.multi_needle import run_multi_needle
|
| 16 |
+
|
| 17 |
+
logging.basicConfig(
|
| 18 |
+
format="%(asctime)s - %(levelname)s - %(message)s",
|
| 19 |
+
level=logging.INFO,
|
| 20 |
+
stream=sys.stdout,
|
| 21 |
+
)
|
| 22 |
+
logger = logging.getLogger(__name__)
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
def parse_args():
|
| 26 |
+
p = argparse.ArgumentParser(description="LITM v4 - Exp 3: Multi-Needle")
|
| 27 |
+
p.add_argument("--model", default="Qwen/Qwen2.5-1.5B-Instruct")
|
| 28 |
+
p.add_argument("--n-examples", type=int, default=30)
|
| 29 |
+
p.add_argument("--n-sentences", type=int, default=300)
|
| 30 |
+
p.add_argument("--output", default="/kaggle/working/litm_results")
|
| 31 |
+
return p.parse_args()
|
| 32 |
+
|
| 33 |
+
|
| 34 |
+
def main():
|
| 35 |
+
args = parse_args()
|
| 36 |
+
out_dir = os.path.join(args.output, "exp3_multi")
|
| 37 |
+
os.makedirs(out_dir, exist_ok=True)
|
| 38 |
+
|
| 39 |
+
logger.info("=" * 60)
|
| 40 |
+
logger.info("LITM v4 - Experiment 3: Multi-Needle Retrieval")
|
| 41 |
+
logger.info(f"Model: {args.model}")
|
| 42 |
+
logger.info(f"Examples: {args.n_examples} | Sentences: {args.n_sentences}")
|
| 43 |
+
logger.info(f"Output: {out_dir}")
|
| 44 |
+
logger.info("=" * 60)
|
| 45 |
+
|
| 46 |
+
result = run_multi_needle(
|
| 47 |
+
model_name=args.model,
|
| 48 |
+
num_sentences=args.n_sentences,
|
| 49 |
+
num_examples=args.n_examples,
|
| 50 |
+
out_dir=out_dir,
|
| 51 |
+
)
|
| 52 |
+
|
| 53 |
+
logger.info("\n" + "=" * 60)
|
| 54 |
+
logger.info("EXPERIMENT 3 COMPLETE")
|
| 55 |
+
logger.info(f"Start accuracy: {result.get('start', 'N/A')}")
|
| 56 |
+
logger.info(f"Middle accuracy: {result.get('middle', 'N/A')}")
|
| 57 |
+
logger.info(f"End accuracy: {result.get('end', 'N/A')}")
|
| 58 |
+
logger.info(f"Results saved to: {out_dir}")
|
| 59 |
+
logger.info("=" * 60)
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
if __name__ == "__main__":
|
| 63 |
+
main()
|