Test commit

app.py

@@ -1,7 +1,6 @@
 import gradio as gr
-from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
-import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
+from gradio_leaderboard import ColumnFilter, Leaderboard, SelectColumns
 from huggingface_hub import snapshot_download
 
 from src.about import (
@@ -20,9 +19,9 @@ from src.display.utils import (
     EVAL_TYPES,
     AutoEvalColumn,
     ModelType,
-    fields,
+    Precision,
     WeightType,
-    Precision
+    fields,
 )
 from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
 from src.populate import get_evaluation_queue_df, get_leaderboard_df
@@ -32,18 +31,29 @@ from src.submission.submit import add_new_eval
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
+
 ### Space initialisation
 try:
     print(EVAL_REQUESTS_PATH)
     snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=QUEUE_REPO,
+        local_dir=EVAL_REQUESTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
 try:
     print(EVAL_RESULTS_PATH)
     snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=RESULTS_REPO,
+        local_dir=EVAL_RESULTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
@@ -57,6 +67,7 @@ LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS,
     pending_eval_queue_df,
 ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
 
+
 def init_leaderboard(dataframe):
     if dataframe is None or dataframe.empty:
         raise ValueError("Leaderboard DataFrame is empty or None.")
@@ -80,9 +91,7 @@ def init_leaderboard(dataframe):
                 max=150,
                 label="Select the number of parameters (B)",
             ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
-            ),
+            ColumnFilter(AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True),
         ],
         bool_checkboxgroup_label="Hide models",
         interactive=False,
@@ -201,4 +210,4 @@ with demo:
 scheduler = BackgroundScheduler()
 scheduler.add_job(restart_space, "interval", seconds=1800)
 scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+demo.queue(default_concurrency_limit=40).launch()

src/about.py

@@ -1,6 +1,7 @@
 from dataclasses import dataclass
 from enum import Enum
 
+
 @dataclass
 class Task:
     benchmark: str
@@ -11,7 +12,7 @@ class Task:
 # Select your tasks here
 # ---------------------------------------------------
 class Tasks(Enum):
-    # task_key in the json file, metric_key in the json file, name to display in the leaderboard 
+    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
     task0 = Task("category_mean", "history", "History")
     task1 = Task("category_mean", "grammar", "Grammar")
     task2 = Task("category_mean", "logic", "Logic")
@@ -19,9 +20,9 @@ class Tasks(Enum):
     task4 = Task("category_mean", "spelling", "Spelling")
     task5 = Task("category_mean", "Vocabulary", "Vocabulary")
 
-NUM_FEWSHOT = 0 # Change with your few shot
-# ---------------------------------------------------
 
+NUM_FEWSHOT = 0  # Change with your few shot
+# ---------------------------------------------------
 
 
 # Your leaderboard name
@@ -33,7 +34,7 @@ This leaderboard evaluates the performance of models on the HunBench benchmark.
 """
 
 # Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
+LLM_BENCHMARKS_TEXT = """
 ## How it works
 TODO
 ## Reproducibility

src/display/utils.py

@@ -27,7 +27,7 @@ auto_eval_column_dict = []
 auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
 auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
 # Scores
-auto_eval_column_dict.append(["mean_score", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
+auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
 for task in Tasks:
     auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
 # Model information

src/leaderboard/read_evals.py

@@ -1,6 +1,5 @@
 import glob
 import json
-import math
 import os
 from dataclasses import dataclass
 
@@ -8,28 +7,28 @@ import dateutil
 import numpy as np
 
 from src.display.formatting import make_clickable_model
-from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
+from src.display.utils import AutoEvalColumn, ModelType, Precision, Tasks, WeightType
 from src.submission.check_validity import is_model_on_hub
 
 
 @dataclass
 class EvalResult:
-    """Represents one full evaluation. Built from a combination of the result and request file for a given run.
-    """
-    eval_name: str # org_model_precision (uid)
-    full_model: str # org/model (path on hub)
-    org: str 
+    """Represents one full evaluation. Built from a combination of the result and request file for a given run."""
+
+    eval_name: str  # org_model_precision (uid)
+    full_model: str  # org/model (path on hub)
+    org: str
     model: str
-    revision: str # commit hash, "" if main
+    revision: str  # commit hash, "" if main
     results: dict
     precision: Precision = Precision.Unknown
-    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
-    weight_type: WeightType = WeightType.Original # Original or Adapter
-    architecture: str = "Unknown" 
+    model_type: ModelType = ModelType.Unknown  # Pretrained, fine tuned, ...
+    weight_type: WeightType = WeightType.Original  # Original or Adapter
+    architecture: str = "Unknown"
     license: str = "?"
     likes: int = 0
     num_params: int = 0
-    date: str = "" # submission date of request file
+    date: str = ""  # submission date of request file
     still_on_hub: bool = False
 
     @classmethod
@@ -85,10 +84,10 @@ class EvalResult:
             org=org,
             model=model,
             results=results,
-            precision=precision,  
-            revision= config.get("model_sha", ""),
+            precision=precision,
+            revision=config.get("model_sha", ""),
             still_on_hub=still_on_hub,
-            architecture=architecture
+            architecture=architecture,
         )
 
     def update_with_request_file(self, requests_path):
@@ -105,7 +104,9 @@ class EvalResult:
             self.num_params = request.get("params", 0)
             self.date = request.get("submitted_time", "")
         except Exception:
-            print(f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}")
+            print(
+                f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}"
+            )
 
     def to_dict(self):
         """Converts the Eval Result to a dict compatible with our dataframe display"""
@@ -146,10 +147,7 @@ def get_request_file_for_model(requests_path, model_name, precision):
     for tmp_request_file in request_files:
         with open(tmp_request_file, "r") as f:
             req_content = json.load(f)
-            if (
-                req_content["status"] in ["FINISHED"]
-                and req_content["precision"] == precision.split(".")[-1]
-            ):
+            if req_content["status"] in ["FINISHED"] and req_content["precision"] == precision.split(".")[-1]:
                 request_file = tmp_request_file
     return request_file
 
@@ -188,7 +186,7 @@ def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResu
     results = []
     for v in eval_results.values():
         try:
-            v.to_dict() # we test if the dict version is complete
+            v.to_dict()  # we test if the dict version is complete
             results.append(v)
         except KeyError:  # not all eval values present
             continue

src/populate.py

@@ -14,7 +14,7 @@ def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchm
     all_data_json = [v.to_dict() for v in raw_data]
 
     df = pd.DataFrame.from_records(all_data_json)
-    df = df.sort_values(by=[AutoEvalColumn.mean_score.name], ascending=False)
+    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
     df = df[cols].round(decimals=2)
 
     # filter out if any of the benchmarks have not been produced

src/submission/check_validity.py

@@ -1,8 +1,6 @@
 import json
 import os
-import re
 from collections import defaultdict
-from datetime import datetime, timedelta, timezone
 
 import huggingface_hub
 from huggingface_hub import ModelCard
@@ -10,6 +8,7 @@ from huggingface_hub.hf_api import ModelInfo
 from transformers import AutoConfig
 from transformers.models.auto.tokenization_auto import AutoTokenizer
 
+
 def check_model_card(repo_id: str) -> tuple[bool, str]:
     """Checks if the model card and license exist and have been filled"""
     try:
@@ -31,31 +30,38 @@ def check_model_card(repo_id: str) -> tuple[bool, str]:
 
     return True, ""
 
-def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
+
+def is_model_on_hub(
+    model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False
+) -> tuple[bool, str]:
     """Checks if the model model_name is on the hub, and whether it (and its tokenizer) can be loaded with AutoClasses."""
     try:
-        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+        config = AutoConfig.from_pretrained(
+            model_name, revision=revision, trust_remote_code=trust_remote_code, token=token
+        )
         if test_tokenizer:
             try:
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+                tk = AutoTokenizer.from_pretrained(
+                    model_name, revision=revision, trust_remote_code=trust_remote_code, token=token
+                )
             except ValueError as e:
+                return (False, f"uses a tokenizer which is not in a transformers release: {e}", None)
+            except Exception:
                 return (
                     False,
-                    f"uses a tokenizer which is not in a transformers release: {e}",
-                    None
+                    "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?",
+                    None,
                 )
-            except Exception as e:
-                return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
         return True, None, config
 
     except ValueError:
         return (
             False,
             "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
-            None
+            None,
         )
 
-    except Exception as e:
+    except Exception:
         return False, "was not found on hub!", None
 
 
@@ -70,10 +76,12 @@ def get_model_size(model_info: ModelInfo, precision: str):
     model_size = size_factor * model_size
     return model_size
 
+
 def get_model_arch(model_info: ModelInfo):
     """Gets the model architecture from the configuration"""
     return model_info.config.get("architectures", "Unknown")
 
+
 def already_submitted_models(requested_models_dir: str) -> set[str]:
     """Gather a list of already submitted models to avoid duplicates"""
     depth = 1

src/submission/submit.py

@@ -3,17 +3,13 @@ import os
 from datetime import datetime, timezone
 
 from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
-from src.submission.check_validity import (
-    already_submitted_models,
-    check_model_card,
-    get_model_size,
-    is_model_on_hub,
-)
+from src.envs import API, EVAL_REQUESTS_PATH, QUEUE_REPO, TOKEN
+from src.submission.check_validity import already_submitted_models, check_model_card, get_model_size, is_model_on_hub
 
 REQUESTED_MODELS = None
 USERS_TO_SUBMISSION_DATES = None
 
+
 def add_new_eval(
     model: str,
     base_model: str,
@@ -45,7 +41,9 @@ def add_new_eval(
 
     # Is the model on the hub?
     if weight_type in ["Delta", "Adapter"]:
-        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
+        base_model_on_hub, error, _ = is_model_on_hub(
+            model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True
+        )
         if not base_model_on_hub:
             return styled_error(f'Base model "{base_model}" {error}')