Spaces:
Runtime error
Runtime error
Omachoko
commited on
Commit
·
c7fc100
1
Parent(s):
997480e
Conform app.py to required evaluation runner design
Browse files
app.py
CHANGED
@@ -9,20 +9,151 @@ import gradio as gr
|
|
9 |
import json
|
10 |
from datetime import datetime
|
11 |
from gaia_agent import ModularGAIAAgent
|
|
|
|
|
|
|
12 |
|
13 |
agent = ModularGAIAAgent()
|
14 |
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
for r in results:
|
19 |
-
answers += f"Task ID: {r['task_id']}\nAnswer: {r['answer']}\nReasoning Trace: {' | '.join(r['reasoning_trace'])}\n\n"
|
20 |
-
return answers
|
21 |
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
def show_help():
|
28 |
return (
|
@@ -74,4 +205,25 @@ with demo:
|
|
74 |
help_md = gr.Markdown(show_help())
|
75 |
|
76 |
if __name__ == "__main__":
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
import json
|
10 |
from datetime import datetime
|
11 |
from gaia_agent import ModularGAIAAgent
|
12 |
+
import requests
|
13 |
+
import inspect
|
14 |
+
import pandas as pd
|
15 |
|
16 |
agent = ModularGAIAAgent()
|
17 |
|
18 |
+
# (Keep Constants as is)
|
19 |
+
# --- Constants ---
|
20 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
|
|
|
|
|
|
21 |
|
22 |
+
# --- Basic Agent Definition ---
|
23 |
+
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
|
24 |
+
class BasicAgent:
|
25 |
+
def __init__(self):
|
26 |
+
print("BasicAgent (GAIA Modular Agent) initialized.")
|
27 |
+
self.agent = ModularGAIAAgent()
|
28 |
+
def __call__(self, question: str) -> str:
|
29 |
+
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
30 |
+
try:
|
31 |
+
answer, trace = self.agent.answer_question({"task_id": "manual", "question": question, "file_name": ""})
|
32 |
+
print(f"Agent returning answer: {answer}")
|
33 |
+
return answer
|
34 |
+
except Exception as e:
|
35 |
+
print(f"Agent error: {e}")
|
36 |
+
return f"AGENT ERROR: {e}"
|
37 |
+
|
38 |
+
def run_and_submit_all( profile: gr.OAuthProfile | None):
|
39 |
+
"""
|
40 |
+
Fetches all questions, runs the BasicAgent on them, submits all answers,
|
41 |
+
and displays the results.
|
42 |
+
"""
|
43 |
+
# --- Determine HF Space Runtime URL and Repo URL ---
|
44 |
+
space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
|
45 |
+
|
46 |
+
if profile:
|
47 |
+
username= f"{profile.username}"
|
48 |
+
print(f"User logged in: {username}")
|
49 |
+
else:
|
50 |
+
print("User not logged in.")
|
51 |
+
return "Please Login to Hugging Face with the button.", None
|
52 |
+
|
53 |
+
api_url = DEFAULT_API_URL
|
54 |
+
questions_url = f"{api_url}/questions"
|
55 |
+
submit_url = f"{api_url}/submit"
|
56 |
+
|
57 |
+
# 1. Instantiate Agent ( modify this part to create your agent)
|
58 |
+
try:
|
59 |
+
agent = BasicAgent()
|
60 |
+
except Exception as e:
|
61 |
+
print(f"Error instantiating agent: {e}")
|
62 |
+
return f"Error initializing agent: {e}", None
|
63 |
+
# 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)
|
64 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
65 |
+
print(agent_code)
|
66 |
+
|
67 |
+
# 2. Fetch Questions
|
68 |
+
print(f"Fetching questions from: {questions_url}")
|
69 |
+
try:
|
70 |
+
response = requests.get(questions_url, timeout=15)
|
71 |
+
response.raise_for_status()
|
72 |
+
questions_data = response.json()
|
73 |
+
if not questions_data:
|
74 |
+
print("Fetched questions list is empty.")
|
75 |
+
return "Fetched questions list is empty or invalid format.", None
|
76 |
+
print(f"Fetched {len(questions_data)} questions.")
|
77 |
+
except requests.exceptions.RequestException as e:
|
78 |
+
print(f"Error fetching questions: {e}")
|
79 |
+
return f"Error fetching questions: {e}", None
|
80 |
+
except requests.exceptions.JSONDecodeError as e:
|
81 |
+
print(f"Error decoding JSON response from questions endpoint: {e}")
|
82 |
+
print(f"Response text: {response.text[:500]}")
|
83 |
+
return f"Error decoding server response for questions: {e}", None
|
84 |
+
except Exception as e:
|
85 |
+
print(f"An unexpected error occurred fetching questions: {e}")
|
86 |
+
return f"An unexpected error occurred fetching questions: {e}", None
|
87 |
+
|
88 |
+
# 3. Run your Agent
|
89 |
+
results_log = []
|
90 |
+
answers_payload = []
|
91 |
+
print(f"Running agent on {len(questions_data)} questions...")
|
92 |
+
for item in questions_data:
|
93 |
+
task_id = item.get("task_id")
|
94 |
+
question_text = item.get("question")
|
95 |
+
if not task_id or question_text is None:
|
96 |
+
print(f"Skipping item with missing task_id or question: {item}")
|
97 |
+
continue
|
98 |
+
try:
|
99 |
+
submitted_answer = agent(question_text)
|
100 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
101 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
102 |
+
except Exception as e:
|
103 |
+
print(f"Error running agent on task {task_id}: {e}")
|
104 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
105 |
+
|
106 |
+
if not answers_payload:
|
107 |
+
print("Agent did not produce any answers to submit.")
|
108 |
+
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
109 |
+
|
110 |
+
# 4. Prepare Submission
|
111 |
+
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
112 |
+
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
113 |
+
print(status_update)
|
114 |
+
|
115 |
+
# 5. Submit
|
116 |
+
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
117 |
+
try:
|
118 |
+
response = requests.post(submit_url, json=submission_data, timeout=60)
|
119 |
+
response.raise_for_status()
|
120 |
+
result_data = response.json()
|
121 |
+
final_status = (
|
122 |
+
f"Submission Successful!\n"
|
123 |
+
f"User: {result_data.get('username')}\n"
|
124 |
+
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
125 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
126 |
+
f"Message: {result_data.get('message', 'No message received.')}"
|
127 |
+
)
|
128 |
+
print("Submission successful.")
|
129 |
+
results_df = pd.DataFrame(results_log)
|
130 |
+
return final_status, results_df
|
131 |
+
except requests.exceptions.HTTPError as e:
|
132 |
+
error_detail = f"Server responded with status {e.response.status_code}."
|
133 |
+
try:
|
134 |
+
error_json = e.response.json()
|
135 |
+
error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
|
136 |
+
except requests.exceptions.JSONDecodeError:
|
137 |
+
error_detail += f" Response: {e.response.text[:500]}"
|
138 |
+
status_message = f"Submission Failed: {error_detail}"
|
139 |
+
print(status_message)
|
140 |
+
results_df = pd.DataFrame(results_log)
|
141 |
+
return status_message, results_df
|
142 |
+
except requests.exceptions.Timeout:
|
143 |
+
status_message = "Submission Failed: The request timed out."
|
144 |
+
print(status_message)
|
145 |
+
results_df = pd.DataFrame(results_log)
|
146 |
+
return status_message, results_df
|
147 |
+
except requests.exceptions.RequestException as e:
|
148 |
+
status_message = f"Submission Failed: Network error - {e}"
|
149 |
+
print(status_message)
|
150 |
+
results_df = pd.DataFrame(results_log)
|
151 |
+
return status_message, results_df
|
152 |
+
except Exception as e:
|
153 |
+
status_message = f"An unexpected error occurred during submission: {e}"
|
154 |
+
print(status_message)
|
155 |
+
results_df = pd.DataFrame(results_log)
|
156 |
+
return status_message, results_df
|
157 |
|
158 |
def show_help():
|
159 |
return (
|
|
|
205 |
help_md = gr.Markdown(show_help())
|
206 |
|
207 |
if __name__ == "__main__":
|
208 |
+
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
209 |
+
# Check for SPACE_HOST and SPACE_ID at startup for information
|
210 |
+
space_host_startup = os.getenv("SPACE_HOST")
|
211 |
+
space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
|
212 |
+
|
213 |
+
if space_host_startup:
|
214 |
+
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
215 |
+
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
216 |
+
else:
|
217 |
+
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
218 |
+
|
219 |
+
if space_id_startup: # Print repo URLs if SPACE_ID is found
|
220 |
+
print(f"✅ SPACE_ID found: {space_id_startup}")
|
221 |
+
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
222 |
+
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
223 |
+
else:
|
224 |
+
print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
|
225 |
+
|
226 |
+
print("-"*(60 + len(" App Starting ")) + "\n")
|
227 |
+
|
228 |
+
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
229 |
+
demo.launch(debug=True, share=False)
|