Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -227,3 +227,43 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 227 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 228 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 229 |
except Exception as e:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 228 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 229 |
except Exception as e:
|
| 230 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
| 231 |
+
|
| 232 |
+
# Prepare submission
|
| 233 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 234 |
+
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
| 235 |
+
|
| 236 |
+
# Submit answers
|
| 237 |
+
try:
|
| 238 |
+
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 239 |
+
response.raise_for_status()
|
| 240 |
+
result_data = response.json()
|
| 241 |
+
final_status = (
|
| 242 |
+
f"Submission Successful!\n"
|
| 243 |
+
f"User: {result_data.get('username')}\n"
|
| 244 |
+
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
| 245 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 246 |
+
f"Message: {result_data.get('message', 'No message received.')}"
|
| 247 |
+
)
|
| 248 |
+
results_df = pd.DataFrame(results_log)
|
| 249 |
+
return final_status, results_df
|
| 250 |
+
except Exception as e:
|
| 251 |
+
results_df = pd.DataFrame(results_log)
|
| 252 |
+
return f"Submission failed: {e}", results_df
|
| 253 |
+
|
| 254 |
+
# ------------------ GRADIO INTERFACE ------------------
|
| 255 |
+
|
| 256 |
+
with gr.Blocks() as demo:
|
| 257 |
+
gr.Markdown("# GAIA Agent Evaluation Runner")
|
| 258 |
+
gr.LoginButton()
|
| 259 |
+
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 260 |
+
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 261 |
+
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 262 |
+
|
| 263 |
+
run_button.click(
|
| 264 |
+
fn=run_and_submit_all,
|
| 265 |
+
outputs=[status_output, results_table]
|
| 266 |
+
)
|
| 267 |
+
|
| 268 |
+
if __name__ == "__main__":
|
| 269 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|