Handle hf_token for local runs

yourbench_space/app.py

@@ -146,16 +146,19 @@ def run_evaluation_pipeline(oauth_token: gr.OAuthToken | None, org_name, eval_na
     repo_id = f"{org_name}/leaderboard_yourbench_{eval_ds_name.replace('/', '_')}"
     folder_path = str(Path(PATH) / "yourbench_space" / "leaderboard_space")
 
+    new_env = os.environ.copy()
+    if oauth_token:
+        hf_token = oauth_token.token
+        new_env["HF_TOKEN"] = hf_token
+    else:
+        hf_token = os.environ.get("HF_TOKEN")
+
     try:
-        load_dataset(eval_ds_name, name=config_name, streaming=True, token=oauth_token.token)
+        load_dataset(eval_ds_name, name=config_name, streaming=True, token=hf_token)
     except Exception as e:
         logger.error(f"Failed to load dataset '{eval_ds_name}': {e}")
         return "❌ Failed: Dataset loading error"
 
-    new_env = os.environ.copy()
-    if oauth_token:
-        new_env["HF_TOKEN"] = oauth_token.token
-
     try:
         create_eval_file(eval_ds_name)
         status = asyncio.run(run_evaluations(org=org_name, eval_ds_name=eval_ds_name, custom_env=new_env))
@@ -171,18 +174,18 @@ def run_evaluation_pipeline(oauth_token: gr.OAuthToken | None, org_name, eval_na
             repo_id=repo_id,
             repo_type="space",
             space_sdk="gradio",
-            token=oauth_token.token,
+            token=hf_token,
         )
     except Exception as e:
         if "409" in str(e) and "already created this space repo" in str(e):
             logger.info(f"Space '{repo_id}' already exists. Deleting and regenerating it.")
             try:
-                api.delete_repo(repo_id=repo_id, repo_type="space", token=oauth_token.token)
+                api.delete_repo(repo_id=repo_id, repo_type="space", token=hf_token)
                 api.create_repo(
                     repo_id=repo_id,
                     repo_type="space",
                     space_sdk="gradio",
-                    token=oauth_token.token,
+                    token=hf_token,
                 )
                 space_was_regenerated = True
             except Exception as delete_err:
@@ -197,16 +200,16 @@ def run_evaluation_pipeline(oauth_token: gr.OAuthToken | None, org_name, eval_na
             repo_id=repo_id,
             repo_type="space",
             folder_path=folder_path,
-            token=oauth_token.token,
+            token=hf_token,
         )
         api.add_space_secret(
             repo_id=repo_id,
             key="HF_TOKEN",
-            value=oauth_token.token,
-            token=oauth_token.token,
+            value=hf_token,
+            token=hf_token,
         )
-        api.add_space_variable(repo_id=repo_id, key="TASK", value=eval_ds_name, token=oauth_token.token)
-        api.add_space_variable(repo_id=repo_id, key="ORG_NAME", value=org_name, token=oauth_token.token)
+        api.add_space_variable(repo_id=repo_id, key="TASK", value=eval_ds_name, token=hf_token)
+        api.add_space_variable(repo_id=repo_id, key="ORG_NAME", value=org_name, token=hf_token)
     except Exception as e:
         logger.error(f"Failed during space setup: {e}")
         return f"✅ Evaluation succeeded\n❌ Failed: Space setup error\n{e}"
@@ -247,7 +250,7 @@ def init_session(profile: gr.OAuthProfile | None):
 btn_launch_evals = gr.Button(
     "🚀 Launch Evaluation",
     visible=True,
-    interactive=False,  # Start non-interactive
+    interactive=True,  # Start non-interactive
     variant="primary",
 )