Upload runner.py with huggingface_hub

runner.py

@@ -8,6 +8,7 @@
 #     "transformers",
 #     "accelerate",
 #     "huggingface_hub",
+#     "inspect_evals",
 # ]
 # ///
 """Runner that downloads an eval script and executes it using inspect CLI."""
@@ -58,7 +59,7 @@ def bundle_and_upload_to_space(log_dir: str, hf_space_id: str, hf_token: str):
 
 if __name__ == "__main__":
     if len(sys.argv) < 4:
-        print("Usage: eval_runner.py <eval_ref> <model> <space_id> [log_dir] [--package-path] [extra_args...]")
+        print("Usage: eval_runner.py <eval_ref> <model> <space_id> [log_dir] [--inspect-evals] [extra_args...]")
         sys.exit(1)
 
     eval_ref = sys.argv[1]
@@ -66,17 +67,17 @@ if __name__ == "__main__":
     hf_space_id = sys.argv[3]
     log_dir = sys.argv[4] if len(sys.argv) > 4 and not sys.argv[4].startswith("--") else "./logs"
 
-    # Check if it's a package path (e.g., inspect_evals/arc_easy, petri/audit)
-    is_package_path = "--package-path" in sys.argv
-    extra_args = [arg for arg in sys.argv[5:] if arg != "--package-path"]
+    # Check if this is an inspect_evals path
+    is_inspect_evals = "--inspect-evals" in sys.argv
+    extra_args = [arg for arg in sys.argv[5:] if arg != "--inspect-evals"]
 
-    if is_package_path:
-        # Package path - use directly with inspect eval
-        print(f"Using package path: {eval_ref}")
+    if is_inspect_evals:
+        # Use inspect_evals path directly
+        print(f"Using inspect_evals: {eval_ref}")
         eval_target = eval_ref
         cleanup_file = None
     else:
-        # URL - download and use
+        # Download custom eval script
         print(f"Downloading eval from {eval_ref}...")
         with urllib.request.urlopen(eval_ref) as response:
             eval_code = response.read().decode('utf-8')