Spaces:
Sleeping
Sleeping
add already_answered write
Browse files- simuGAIA.py +26 -0
simuGAIA.py
CHANGED
@@ -232,6 +232,7 @@ def run_simuGAIA_all( profile: gr.OAuthProfile | None, submit: Optional[bool] =
|
|
232 |
# 3. Run your Agent
|
233 |
results_log = []
|
234 |
answers_payload = []
|
|
|
235 |
print(f"Running agent on {len(questions_data)} questions...")
|
236 |
for item in questions_data:
|
237 |
task_id = item.get("task_id")
|
@@ -243,11 +244,34 @@ def run_simuGAIA_all( profile: gr.OAuthProfile | None, submit: Optional[bool] =
|
|
243 |
if question_text.startswith(".rewsna eht sa"): # ("How many studio albums"): <--- REMOVE THAT FOR ALL QUESTIONS
|
244 |
print(f"First question detected. INVOKING AGENT! Be careful!")
|
245 |
submitted_answer = agent.invoke(question_text)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
246 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
247 |
else:
|
248 |
submitted_answer = "NO AGENT INVOKED"
|
249 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
250 |
|
|
|
251 |
except Exception as e:
|
252 |
print(f"Error running agent on task {task_id}: {e}")
|
253 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
@@ -259,11 +283,13 @@ def run_simuGAIA_all( profile: gr.OAuthProfile | None, submit: Optional[bool] =
|
|
259 |
if not submit:
|
260 |
return "Run finished. No submission done, as asked.", pd.DataFrame(results_log)
|
261 |
|
|
|
262 |
# 4. Prepare Submission
|
263 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
264 |
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
265 |
print(status_update)
|
266 |
|
|
|
267 |
# 5. Submit
|
268 |
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
269 |
try:
|
|
|
232 |
# 3. Run your Agent
|
233 |
results_log = []
|
234 |
answers_payload = []
|
235 |
+
|
236 |
print(f"Running agent on {len(questions_data)} questions...")
|
237 |
for item in questions_data:
|
238 |
task_id = item.get("task_id")
|
|
|
244 |
if question_text.startswith(".rewsna eht sa"): # ("How many studio albums"): <--- REMOVE THAT FOR ALL QUESTIONS
|
245 |
print(f"First question detected. INVOKING AGENT! Be careful!")
|
246 |
submitted_answer = agent.invoke(question_text)
|
247 |
+
|
248 |
+
# Save answer, task_id, and question_text to already_answered.json
|
249 |
+
record = {
|
250 |
+
"task_id": task_id,
|
251 |
+
"question_text": question_text,
|
252 |
+
"submitted_answer": submitted_answer
|
253 |
+
}
|
254 |
+
json_path = "already_answered.json"
|
255 |
+
try:
|
256 |
+
if os.path.exists(json_path):
|
257 |
+
with open(json_path, "r", encoding="utf-8") as f:
|
258 |
+
data = json.load(f)
|
259 |
+
if not isinstance(data, list):
|
260 |
+
data = []
|
261 |
+
else:
|
262 |
+
data = []
|
263 |
+
data.append(record)
|
264 |
+
with open(json_path, "w", encoding="utf-8") as f:
|
265 |
+
json.dump(data, f, ensure_ascii=False, indent=2)
|
266 |
+
except Exception as e:
|
267 |
+
print(f"Error saving to {json_path}: {e}")
|
268 |
+
|
269 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
270 |
else:
|
271 |
submitted_answer = "NO AGENT INVOKED"
|
272 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
273 |
|
274 |
+
|
275 |
except Exception as e:
|
276 |
print(f"Error running agent on task {task_id}: {e}")
|
277 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
|
|
283 |
if not submit:
|
284 |
return "Run finished. No submission done, as asked.", pd.DataFrame(results_log)
|
285 |
|
286 |
+
|
287 |
# 4. Prepare Submission
|
288 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
289 |
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
290 |
print(status_update)
|
291 |
|
292 |
+
|
293 |
# 5. Submit
|
294 |
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
295 |
try:
|