fix/pyaudioop

.github/workflows/main.yaml

@@ -17,4 +17,4 @@ jobs:
       - name: Push to hub
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
-        run: git push --force https://hanhainebula:$HF_TOKEN@huggingface.co/spaces/AIR-Bench/leaderboard main
+        run: git push https://hanhainebula:$HF_TOKEN@huggingface.co/spaces/AIR-Bench/leaderboard main

app.py

@@ -1,55 +1,3 @@
-import sys
-import types
-
-# Python 3.13 removed audioop from stdlib; pydub (a gradio dep) needs it.
-# Provide minimal stubs so the import doesn't crash the leaderboard.
-try:
-    import audioop  # noqa: F811
-except ModuleNotFoundError:
-    _audioop = types.ModuleType("audioop")
-    _func_names = [
-        "add", "adpcm2lin", "alaw2lin", "avg", "avgpp", "bias",
-        "byteswap", "cross", "findfactor", "findfit", "findmax",
-        "getsample", "lin2adpcm", "lin2alaw", "lin2lin", "lin2ulaw",
-        "max", "maxpp", "minmax", "mul", "ratecv", "reverse", "rms",
-        "tomono", "tostereo", "ulaw2lin",
-    ]
-
-    def _make_stub(name):
-        def stub(*args, **kwargs):
-            raise NotImplementedError(f"audioop.{name} is not available on Python 3.13")
-
-        return stub
-
-    for _name in _func_names:
-        setattr(_audioop, _name, _make_stub(_name))
-    _audioop.error = Exception
-    sys.modules["audioop"] = _audioop
-
-# huggingface_hub >= 1.0 removed HfFolder (gradio 4.29.0 oauth.py imports it)
-import huggingface_hub
-
-if not hasattr(huggingface_hub, "HfFolder"):
-
-    class _HfFolder:
-        path = "/root/.huggingface/token"
-
-        @staticmethod
-        def get_token():
-            import os as _os
-
-            return _os.environ.get("HF_TOKEN")
-
-        @staticmethod
-        def save_token(token):
-            pass
-
-        @staticmethod
-        def delete_token():
-            pass
-
-    huggingface_hub.HfFolder = _HfFolder
-
 import os
 
 import gradio as gr

requirements.txt

@@ -4,7 +4,7 @@ click>=8.1.3
 datasets>=2.14.5
 gradio<5.0.0
 gradio_client>=0.16.1
-huggingface-hub>=0.18.0,<1.0.0
+huggingface-hub>=0.18.0
 numpy>=1.24.2
 pandas>=2.0.0
 python-dateutil>=2.8.2

src/columns.py

@@ -1,4 +1,8 @@
-from dataclasses import dataclass, field, make_dataclass
+from dataclasses import dataclass, make_dataclass
+
+
+def _fields(raw_class):
+    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
 
 
 # These classes are for user facing column names,
@@ -15,43 +19,37 @@ class ColumnContent:
 
 def get_default_auto_eval_column_dict():
     auto_eval_column_dict = []
-    auto_eval_column_dict.append(
-        ["rank", ColumnContent, field(default_factory=lambda: ColumnContent(COL_NAME_RANK, "number", True))]
-    )
+    auto_eval_column_dict.append(["rank", ColumnContent, ColumnContent(COL_NAME_RANK, "number", True)])
     auto_eval_column_dict.append(
         [
             "retrieval_model",
             ColumnContent,
-            field(default_factory=lambda: ColumnContent(COL_NAME_RETRIEVAL_MODEL, "markdown", True, never_hidden=True)),
+            ColumnContent(COL_NAME_RETRIEVAL_MODEL, "markdown", True, never_hidden=True),
         ]
     )
     auto_eval_column_dict.append(
         [
             "reranking_model",
             ColumnContent,
-            field(default_factory=lambda: ColumnContent(COL_NAME_RERANKING_MODEL, "markdown", True, never_hidden=True)),
+            ColumnContent(COL_NAME_RERANKING_MODEL, "markdown", True, never_hidden=True),
         ]
     )
     auto_eval_column_dict.append(
-        ["revision", ColumnContent, field(default_factory=lambda: ColumnContent(COL_NAME_REVISION, "markdown", True, never_hidden=True))]
-    )
-    auto_eval_column_dict.append(
-        ["timestamp", ColumnContent, field(default_factory=lambda: ColumnContent(COL_NAME_TIMESTAMP, "date", True, never_hidden=True))]
+        ["revision", ColumnContent, ColumnContent(COL_NAME_REVISION, "markdown", True, never_hidden=True)]
     )
     auto_eval_column_dict.append(
-        ["average", ColumnContent, field(default_factory=lambda: ColumnContent(COL_NAME_AVG, "number", True))]
+        ["timestamp", ColumnContent, ColumnContent(COL_NAME_TIMESTAMP, "date", True, never_hidden=True)]
     )
+    auto_eval_column_dict.append(["average", ColumnContent, ColumnContent(COL_NAME_AVG, "number", True)])
     auto_eval_column_dict.append(
         [
             "retrieval_model_link",
             ColumnContent,
-            field(
-                default_factory=lambda: ColumnContent(
-                    COL_NAME_RETRIEVAL_MODEL_LINK,
-                    "markdown",
-                    False,
-                    hidden=True,
-                )
+            ColumnContent(
+                COL_NAME_RETRIEVAL_MODEL_LINK,
+                "markdown",
+                False,
+                hidden=True,
             ),
         ]
     )
@@ -59,18 +57,16 @@ def get_default_auto_eval_column_dict():
         [
             "reranking_model_link",
             ColumnContent,
-            field(
-                default_factory=lambda: ColumnContent(
-                    COL_NAME_RERANKING_MODEL_LINK,
-                    "markdown",
-                    False,
-                    hidden=True,
-                )
+            ColumnContent(
+                COL_NAME_RERANKING_MODEL_LINK,
+                "markdown",
+                False,
+                hidden=True,
             ),
         ]
     )
     auto_eval_column_dict.append(
-        ["is_anonymous", ColumnContent, field(default_factory=lambda: ColumnContent(COL_NAME_IS_ANONYMOUS, "bool", False, hidden=True))]
+        ["is_anonymous", ColumnContent, ColumnContent(COL_NAME_IS_ANONYMOUS, "bool", False, hidden=True)]
     )
     return auto_eval_column_dict
 
@@ -80,7 +76,7 @@ def make_autoevalcolumn(cls_name, benchmarks):
     # Leaderboard columns
     for benchmark in list(benchmarks.value):
         auto_eval_column_dict.append(
-            [benchmark.name, ColumnContent, field(default_factory=lambda b=benchmark: ColumnContent(b.value.col_name, "number", True))]
+            [benchmark.name, ColumnContent, ColumnContent(benchmark.value.col_name, "number", True)]
         )
 
     # We use make dataclass to dynamically fill the scores from Tasks
@@ -89,19 +85,14 @@ def make_autoevalcolumn(cls_name, benchmarks):
 
 def get_default_col_names_and_types(benchmarks):
     AutoEvalColumn = make_autoevalcolumn("AutoEvalColumn", benchmarks)
-    col_names = []
-    col_types = []
-    for f in AutoEvalColumn.__dataclass_fields__.values():
-        col = f.default_factory()
-        if not col.hidden:
-            col_names.append(col.name)
-            col_types.append(col.type)
+    col_names = [c.name for c in _fields(AutoEvalColumn) if not c.hidden]
+    col_types = [c.type for c in _fields(AutoEvalColumn) if not c.hidden]
     return col_names, col_types
 
 
 def get_fixed_col_names_and_types():
     fixed_cols = get_default_auto_eval_column_dict()[:-3]
-    return [c.default_factory().name for _, _, c in fixed_cols], [c.default_factory().type for _, _, c in fixed_cols]
+    return [c.name for _, _, c in fixed_cols], [c.type for _, _, c in fixed_cols]
 
 
 COL_NAME_AVG = "Average ⬆️"

src/models.py

@@ -1,6 +1,6 @@
 import json
 from collections import defaultdict
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from enum import Enum
 from typing import List
 
@@ -142,10 +142,10 @@ class LeaderboardDataStore:
     version: str
     slug: str
     raw_data: list = None
-    qa_raw_df: pd.DataFrame = field(default_factory=pd.DataFrame)
-    doc_raw_df: pd.DataFrame = field(default_factory=pd.DataFrame)
-    qa_fmt_df: pd.DataFrame = field(default_factory=pd.DataFrame)
-    doc_fmt_df: pd.DataFrame = field(default_factory=pd.DataFrame)
+    qa_raw_df: pd.DataFrame = pd.DataFrame()
+    doc_raw_df: pd.DataFrame = pd.DataFrame()
+    qa_fmt_df: pd.DataFrame = pd.DataFrame()
+    doc_fmt_df: pd.DataFrame = pd.DataFrame()
     reranking_models: list = None
     qa_types: list = None
     doc_types: list = None