simplified env vars

README.md

@@ -15,7 +15,6 @@ space_ci:
   private: true
   secrets:
   - HF_TOKEN
-  - H4_TOKEN
 tags:
 - leaderboard
 short_description: Track, rank and evaluate open LLMs and chatbots

app.py

@@ -32,9 +32,10 @@ from src.envs import (
     API,
     EVAL_REQUESTS_PATH,
     AGGREGATED_REPO,
-    H4_TOKEN,
+    HF_TOKEN,
     QUEUE_REPO,
     REPO_ID,
+    HF_HOME,
 )
 from src.populate import get_evaluation_queue_df, get_leaderboard_df
 from src.submission.submit import add_new_eval
@@ -48,7 +49,7 @@ enable_space_ci()
 
 
 def restart_space():
-    API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
+    API.restart_space(repo_id=REPO_ID, token=HF_TOKEN)
 
 
 def time_diff_wrapper(func):
@@ -98,7 +99,7 @@ def init_space(full_init: bool = True):
             restart_space()
 
     # Always retrieve the leaderboard DataFrame
-    leaderboard_dataset = datasets.load_dataset(AGGREGATED_REPO, "default", split="train")
+    leaderboard_dataset = datasets.load_dataset(AGGREGATED_REPO, "default", split="train", cache_dir=HF_HOME)
     leaderboard_df = get_leaderboard_df(
         leaderboard_dataset=leaderboard_dataset, 
         cols=COLS,

src/envs.py

@@ -2,7 +2,7 @@ import os
 from huggingface_hub import HfApi
 
 # clone / pull the lmeh eval data
-H4_TOKEN = os.environ.get("H4_TOKEN", None)
+HF_TOKEN = os.environ.get("HF_TOKEN", None)
 
 REPO_ID = "HuggingFaceH4/open_llm_leaderboard"
 QUEUE_REPO = "open-llm-leaderboard/requests"
@@ -29,4 +29,4 @@ RATE_LIMIT_PERIOD = 7
 RATE_LIMIT_QUOTA = 5
 HAS_HIGHER_RATE_LIMIT = ["TheBloke"]
 
-API = HfApi(token=H4_TOKEN)
+API = HfApi(token=HF_TOKEN)

src/submission/submit.py

@@ -8,7 +8,7 @@ from src.display.formatting import styled_error, styled_message, styled_warning
 from src.envs import (
     API,
     EVAL_REQUESTS_PATH,
-    H4_TOKEN,
+    HF_TOKEN,
     QUEUE_REPO,
     RATE_LIMIT_PERIOD,
     RATE_LIMIT_QUOTA,
@@ -76,7 +76,7 @@ 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=H4_TOKEN, test_tokenizer=True
+            model_name=base_model, revision=revision, token=HF_TOKEN, test_tokenizer=True
         )
         if not base_model_on_hub:
             return styled_error(f'Base model "{base_model}" {error}')