added mkdirs and changed some env

src/envs.py

@@ -27,21 +27,21 @@ if not os.access(HF_HOME, os.W_OK):
 else:
     print("Write access confirmed for HF_HOME")
 
-EVAL_REQUESTS_PATH = os.path.join(HF_HOME, "eval-queue")
-EVAL_RESULTS_PATH = os.path.join(HF_HOME, "eval-results")
+EVAL_RESULTS_PATH = os.path.join(HF_HOME, "openbench")
+
+RESET_JUDGEMENT_ENV = "RESET_JUDGEMENT"
+
+API = HfApi(token=H4_TOKEN)
+
+# useless env
+EVAL_REQUESTS_PATH = os.path.join(HF_HOME, "data/eval-queue")
 DYNAMIC_INFO_PATH = os.path.join(HF_HOME, "dynamic-info")
 DYNAMIC_INFO_FILE_PATH = os.path.join(DYNAMIC_INFO_PATH, "model_infos.json")
-
 EVAL_REQUESTS_PATH_PRIVATE = "eval-queue-private"
 EVAL_RESULTS_PATH_PRIVATE = "eval-results-private"
-
 PATH_TO_COLLECTION = "open-llm-leaderboard/llm-leaderboard-best-models-652d6c7965a4619fb5c27a03"
 
 # Rate limit variables
 RATE_LIMIT_PERIOD = 7
 RATE_LIMIT_QUOTA = 5
 HAS_HIGHER_RATE_LIMIT = ["TheBloke"]
-
-RESET_JUDGEMENT_ENV = "RESET_JUDGEMENT"
-
-API = HfApi(token=H4_TOKEN)

src/leaderboard/build_leaderboard.py

@@ -53,6 +53,11 @@ def download_dataset(repo_id, local_dir, repo_type="dataset", max_attempts=3, ba
 
 def download_openbench():
     """Downloads pre generated data"""
+    os.makedirs("data/arena-hard-v0.1/model_answer/internal/", exist_ok=True)
+    os.makedirs("data/arena-hard-v0.1/model_answer/external/", exist_ok=True)
+    os.makedirs("data/arena-hard-v0.1/model_answer/model_judgement/", exist_ok=True)
+    os.makedirs(EVAL_RESULTS_PATH, exist_ok=True)
+
     # download answers of different models that we trust
     download_dataset("Vikhrmodels/openbench-eval", EVAL_RESULTS_PATH)