Spaces:
Sleeping
Sleeping
adding run_simuGAIA_all with no submission by default
Browse files- simuGAIA.py +127 -0
simuGAIA.py
CHANGED
@@ -13,6 +13,10 @@ import traceback
|
|
13 |
# Import our agent
|
14 |
from agent import QAgent
|
15 |
|
|
|
|
|
|
|
|
|
16 |
# Simulation of GAIA benchmark questions
|
17 |
SAMPLE_QUESTIONS = [
|
18 |
{
|
@@ -171,3 +175,126 @@ def run_GAIA_questions_simu():
|
|
171 |
|
172 |
return results
|
173 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
# Import our agent
|
14 |
from agent import QAgent
|
15 |
|
16 |
+
# (Keep Constants as is)
|
17 |
+
# --- Constants ---
|
18 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
19 |
+
|
20 |
# Simulation of GAIA benchmark questions
|
21 |
SAMPLE_QUESTIONS = [
|
22 |
{
|
|
|
175 |
|
176 |
return results
|
177 |
|
178 |
+
|
179 |
+
|
180 |
+
def run_simuGAIA_all( profile: gr.OAuthProfile | None, submit: Optional[bool] = False):
|
181 |
+
"""
|
182 |
+
Fetches all questions, runs the QAgent on them,
|
183 |
+
and displays the results.
|
184 |
+
"""
|
185 |
+
# --- Determine HF Space Runtime URL and Repo URL for submission ---
|
186 |
+
# space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
|
187 |
+
|
188 |
+
if profile:
|
189 |
+
username= f"{profile.username}"
|
190 |
+
print(f"User logged in: {username}")
|
191 |
+
else:
|
192 |
+
print("User not logged in.")
|
193 |
+
return "Please Login to Hugging Face with the button.", None
|
194 |
+
|
195 |
+
api_url = DEFAULT_API_URL
|
196 |
+
questions_url = f"{api_url}/questions"
|
197 |
+
# submit_url = f"{api_url}/submit"
|
198 |
+
|
199 |
+
# 1. Instantiate and init Agent ( modify this part to create your agent)
|
200 |
+
agent = init_agent()
|
201 |
+
|
202 |
+
|
203 |
+
# In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
|
204 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
205 |
+
print(agent_code)
|
206 |
+
|
207 |
+
# 2. Fetch Questions
|
208 |
+
print(f"Fetching questions from: {questions_url}")
|
209 |
+
try:
|
210 |
+
response = requests.get(questions_url, timeout=15)
|
211 |
+
response.raise_for_status()
|
212 |
+
questions_data = response.json()
|
213 |
+
if not questions_data:
|
214 |
+
print("Fetched questions list is empty.")
|
215 |
+
return "Fetched questions list is empty or invalid format.", None
|
216 |
+
print(f"Fetched {len(questions_data)} questions.")
|
217 |
+
except requests.exceptions.RequestException as e:
|
218 |
+
print(f"Error fetching questions: {e}")
|
219 |
+
return f"Error fetching questions: {e}", None
|
220 |
+
except requests.exceptions.JSONDecodeError as e:
|
221 |
+
print(f"Error decoding JSON response from questions endpoint: {e}")
|
222 |
+
print(f"Response text: {response.text[:500]}")
|
223 |
+
return f"Error decoding server response for questions: {e}", None
|
224 |
+
except Exception as e:
|
225 |
+
print(f"An unexpected error occurred fetching questions: {e}")
|
226 |
+
return f"An unexpected error occurred fetching questions: {e}", None
|
227 |
+
|
228 |
+
# 3. Run your Agent
|
229 |
+
results_log = []
|
230 |
+
answers_payload = []
|
231 |
+
print(f"Running agent on {len(questions_data)} questions...")
|
232 |
+
for item in questions_data:
|
233 |
+
task_id = item.get("task_id")
|
234 |
+
question_text = item.get("question")
|
235 |
+
if not task_id or question_text is None:
|
236 |
+
print(f"Skipping item with missing task_id or question: {item}")
|
237 |
+
continue
|
238 |
+
try:
|
239 |
+
# submitted_answer = agent(question_text)
|
240 |
+
# answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
241 |
+
results_log.append({"Task ID": task_id, "Question": question_text}) # , "Submitted Answer": submitted_answer})
|
242 |
+
except Exception as e:
|
243 |
+
print(f"Error running agent on task {task_id}: {e}")
|
244 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
245 |
+
|
246 |
+
if not answers_payload:
|
247 |
+
print("Agent did not produce any answers to submit.")
|
248 |
+
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
249 |
+
|
250 |
+
if not submit:
|
251 |
+
return "Run finished. No submission done, as asked.", pd.DataFrame(results_log)
|
252 |
+
|
253 |
+
# 4. Prepare Submission
|
254 |
+
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
255 |
+
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
256 |
+
print(status_update)
|
257 |
+
|
258 |
+
# 5. Submit
|
259 |
+
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
260 |
+
try:
|
261 |
+
response = requests.post(submit_url, json=submission_data, timeout=60)
|
262 |
+
response.raise_for_status()
|
263 |
+
result_data = response.json()
|
264 |
+
final_status = (
|
265 |
+
f"Submission Successful!\n"
|
266 |
+
f"User: {result_data.get('username')}\n"
|
267 |
+
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
268 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
269 |
+
f"Message: {result_data.get('message', 'No message received.')}"
|
270 |
+
)
|
271 |
+
print("Submission successful.")
|
272 |
+
results_df = pd.DataFrame(results_log)
|
273 |
+
return final_status, results_df
|
274 |
+
except requests.exceptions.HTTPError as e:
|
275 |
+
error_detail = f"Server responded with status {e.response.status_code}."
|
276 |
+
try:
|
277 |
+
error_json = e.response.json()
|
278 |
+
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
279 |
+
except requests.exceptions.JSONDecodeError:
|
280 |
+
error_detail += f" Response: {e.response.text[:500]}"
|
281 |
+
status_message = f"Submission Failed: {error_detail}"
|
282 |
+
print(status_message)
|
283 |
+
results_df = pd.DataFrame(results_log)
|
284 |
+
return status_message, results_df
|
285 |
+
except requests.exceptions.Timeout:
|
286 |
+
status_message = "Submission Failed: The request timed out."
|
287 |
+
print(status_message)
|
288 |
+
results_df = pd.DataFrame(results_log)
|
289 |
+
return status_message, results_df
|
290 |
+
except requests.exceptions.RequestException as e:
|
291 |
+
status_message = f"Submission Failed: Network error - {e}"
|
292 |
+
print(status_message)
|
293 |
+
results_df = pd.DataFrame(results_log)
|
294 |
+
return status_message, results_df
|
295 |
+
except Exception as e:
|
296 |
+
status_message = f"An unexpected error occurred during submission: {e}"
|
297 |
+
print(status_message)
|
298 |
+
results_df = pd.DataFrame(results_log)
|
299 |
+
return status_message, results_df
|
300 |
+
|