Spaces:
Running
Running
Create scripts/sync_lmarena.py
Browse files- scripts/sync_lmarena.py +81 -0
scripts/sync_lmarena.py
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# scripts/sync_lmarena.py
|
| 2 |
+
from __future__ import annotations
|
| 3 |
+
|
| 4 |
+
import json
|
| 5 |
+
import os
|
| 6 |
+
import re
|
| 7 |
+
from datetime import datetime, timezone
|
| 8 |
+
from pathlib import Path
|
| 9 |
+
from typing import Optional, Tuple
|
| 10 |
+
|
| 11 |
+
from huggingface_hub import HfApi, hf_hub_download
|
| 12 |
+
|
| 13 |
+
UPSTREAM_SPACE = "lmarena-ai/lmarena-leaderboard" # source Space
|
| 14 |
+
LOCAL_DIR = Path("data/lmarena")
|
| 15 |
+
LOCAL_CSV = LOCAL_DIR / "leaderboard_table_latest.csv"
|
| 16 |
+
LOCAL_META = LOCAL_DIR / "sync_meta.json"
|
| 17 |
+
|
| 18 |
+
LEADERBOARD_RE = re.compile(r"^leaderboard_table_(\d{8})\.csv$")
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
def _pick_latest_leaderboard_file(files: list[str]) -> Optional[Tuple[str, str]]:
|
| 22 |
+
"""
|
| 23 |
+
Returns (filename, yyyymmdd) for the newest leaderboard_table_YYYYMMDD.csv found.
|
| 24 |
+
"""
|
| 25 |
+
candidates: list[Tuple[str, str]] = []
|
| 26 |
+
for f in files:
|
| 27 |
+
m = LEADERBOARD_RE.match(f)
|
| 28 |
+
if m:
|
| 29 |
+
candidates.append((f, m.group(1)))
|
| 30 |
+
|
| 31 |
+
if not candidates:
|
| 32 |
+
return None
|
| 33 |
+
|
| 34 |
+
# Sort by date string; YYYYMMDD sorts lexicographically correctly
|
| 35 |
+
candidates.sort(key=lambda x: x[1])
|
| 36 |
+
return candidates[-1]
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
def main() -> int:
|
| 40 |
+
token = os.getenv("HF_TOKEN") # optional for public, but recommended for rate limits
|
| 41 |
+
api = HfApi(token=token)
|
| 42 |
+
|
| 43 |
+
# list files from the upstream *space repo*
|
| 44 |
+
files = api.list_repo_files(repo_id=UPSTREAM_SPACE, repo_type="space")
|
| 45 |
+
|
| 46 |
+
latest = _pick_latest_leaderboard_file(files)
|
| 47 |
+
if not latest:
|
| 48 |
+
raise RuntimeError(
|
| 49 |
+
f"No leaderboard_table_YYYYMMDD.csv found in upstream Space: {UPSTREAM_SPACE}"
|
| 50 |
+
)
|
| 51 |
+
|
| 52 |
+
filename, yyyymmdd = latest
|
| 53 |
+
|
| 54 |
+
# Download the raw file to a temp location (hub cache) then copy to our repo path
|
| 55 |
+
downloaded_path = hf_hub_download(
|
| 56 |
+
repo_id=UPSTREAM_SPACE,
|
| 57 |
+
repo_type="space",
|
| 58 |
+
filename=filename,
|
| 59 |
+
token=token,
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
LOCAL_DIR.mkdir(parents=True, exist_ok=True)
|
| 63 |
+
|
| 64 |
+
# Copy file contents to our tracked path
|
| 65 |
+
Path(downloaded_path).replace(LOCAL_CSV) if False else LOCAL_CSV.write_bytes(Path(downloaded_path).read_bytes())
|
| 66 |
+
|
| 67 |
+
meta = {
|
| 68 |
+
"source_space": UPSTREAM_SPACE,
|
| 69 |
+
"source_filename": filename,
|
| 70 |
+
"source_date": yyyymmdd,
|
| 71 |
+
"synced_at_utc": datetime.now(timezone.utc).isoformat(),
|
| 72 |
+
}
|
| 73 |
+
LOCAL_META.write_text(json.dumps(meta, indent=2), encoding="utf-8")
|
| 74 |
+
|
| 75 |
+
print(f"[OK] Synced {filename} -> {LOCAL_CSV}")
|
| 76 |
+
print(json.dumps(meta, indent=2))
|
| 77 |
+
return 0
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
if __name__ == "__main__":
|
| 81 |
+
raise SystemExit(main())
|