Fixing token issue

src/leaderboard/read_evals.py

@@ -10,6 +10,7 @@ import numpy as np
 from src.display.formatting import make_clickable_model
 from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
 from src.submission.check_validity import is_model_on_hub
+from src.envs import TOKEN
 
 
 @dataclass
@@ -56,7 +57,7 @@ class EvalResult:
         full_model = "/".join(org_and_model)
 
         still_on_hub, _, model_config = is_model_on_hub(
-            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
+            full_model, config.get("model_sha", "main"), token=TOKEN, trust_remote_code=True, test_tokenizer=False
         )
         architecture = "?"
         if model_config is not None:

src/submission/check_validity.py

@@ -1,14 +1,13 @@
 import json
 import os
-import re
 from collections import defaultdict
-from datetime import datetime, timedelta, timezone
 
 import huggingface_hub
 from huggingface_hub import ModelCard
 from huggingface_hub.hf_api import ModelInfo
 from transformers import AutoConfig
-from transformers.models.auto.tokenization_auto import tokenizer_class_from_name, get_tokenizer_config
+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"""
@@ -33,7 +32,7 @@ def check_model_card(repo_id: str) -> 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]:
     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, token=token, trust_remote_code=trust_remote_code)
         if test_tokenizer:
             try:
                 print("TOKEN: %s" % token)
@@ -41,7 +40,7 @@ def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_rem
                 print("revision: %s" % revision)
                 print("trust remote code: %s" % trust_remote_code)
 
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+                _ = AutoTokenizer.from_pretrained(model_name, revision=revision, token=token, trust_remote_code=trust_remote_code)
             except ValueError as e:
                 return (
                     False,
@@ -49,7 +48,7 @@ def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_rem
                     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 (False, f"'s tokenizer cannot be loaded. This most commonly happens when your tokenizer class is not in a stable transformers release or not correctly configured?\nExact error is: {e}", None)
         return True, None, config
 
     except ValueError:

src/submission/submit.py

@@ -50,7 +50,7 @@ def add_new_eval(
             return styled_error(f'Base model "{base_model}" {error}')
 
     if not weight_type == "Adapter":
-        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, test_tokenizer=True)
+        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, token=TOKEN, test_tokenizer=True)
         if not model_on_hub:
             return styled_error(f'Model "{model}" {error}')