Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,7 +2,7 @@ import gradio as gr
|
|
| 2 |
import json
|
| 3 |
import os
|
| 4 |
import requests
|
| 5 |
-
|
| 6 |
from agent import solve_task, load_tasks
|
| 7 |
from mistral_hf_wrapper import MistralInference
|
| 8 |
|
|
@@ -22,47 +22,43 @@ def run_and_submit_all():
|
|
| 22 |
|
| 23 |
answers = []
|
| 24 |
for i, task in enumerate(tasks):
|
| 25 |
-
print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task.get('question_id', '
|
| 26 |
-
|
| 27 |
try:
|
| 28 |
result = solve_task(task, model)
|
| 29 |
-
|
| 30 |
-
if not result.get("submitted_answer"):
|
| 31 |
-
print("[WARN] Empty model response detected")
|
| 32 |
result["submitted_answer"] = "ERROR: Empty model response"
|
| 33 |
-
else:
|
| 34 |
-
print(f"Answer: {result['submitted_answer'][:100]}...")
|
| 35 |
-
|
| 36 |
answers.append(result)
|
| 37 |
-
|
| 38 |
except Exception as e:
|
| 39 |
-
|
|
|
|
| 40 |
answers.append({
|
| 41 |
"question_id": task.get("question_id", "UNKNOWN"),
|
| 42 |
"submitted_answer": f"ERROR: {str(e)}"
|
| 43 |
})
|
| 44 |
|
| 45 |
if not answers:
|
| 46 |
-
return "No answers generated. Check model response."
|
| 47 |
-
|
| 48 |
-
print("[INFO] Submitting answers to GAIA benchmark API...")
|
| 49 |
-
res = requests.post(
|
| 50 |
-
"https://agents-course-unit4-scoring.hf.space/submit",
|
| 51 |
-
headers={"Content-Type": "application/json"},
|
| 52 |
-
json={
|
| 53 |
-
"username": USERNAME,
|
| 54 |
-
"agent_code": CODE_LINK,
|
| 55 |
-
"answers": answers
|
| 56 |
-
},
|
| 57 |
-
)
|
| 58 |
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
-
|
| 64 |
-
|
|
|
|
|
|
|
| 65 |
|
|
|
|
|
|
|
| 66 |
# Gradio interface
|
| 67 |
gr.Interface(
|
| 68 |
fn=run_and_submit_all,
|
|
|
|
| 2 |
import json
|
| 3 |
import os
|
| 4 |
import requests
|
| 5 |
+
import traceback
|
| 6 |
from agent import solve_task, load_tasks
|
| 7 |
from mistral_hf_wrapper import MistralInference
|
| 8 |
|
|
|
|
| 22 |
|
| 23 |
answers = []
|
| 24 |
for i, task in enumerate(tasks):
|
| 25 |
+
print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task.get('question_id', 'UNKNOWN')}")
|
|
|
|
| 26 |
try:
|
| 27 |
result = solve_task(task, model)
|
| 28 |
+
print(f"[INFO]Answer: {result['submitted_answer'][:100]}...")
|
| 29 |
+
if not result.get("submitted_answer"):
|
|
|
|
| 30 |
result["submitted_answer"] = "ERROR: Empty model response"
|
|
|
|
|
|
|
|
|
|
| 31 |
answers.append(result)
|
|
|
|
| 32 |
except Exception as e:
|
| 33 |
+
tb = traceback.format_exc()
|
| 34 |
+
print(f"[ERROR] Task failed: {e}\n{tb}")
|
| 35 |
answers.append({
|
| 36 |
"question_id": task.get("question_id", "UNKNOWN"),
|
| 37 |
"submitted_answer": f"ERROR: {str(e)}"
|
| 38 |
})
|
| 39 |
|
| 40 |
if not answers:
|
| 41 |
+
return "No answers generated. Check model response or network."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
|
| 43 |
+
try:
|
| 44 |
+
res = requests.post(
|
| 45 |
+
"https://agents-course-unit4-scoring.hf.space/submit",
|
| 46 |
+
headers={"Content-Type": "application/json"},
|
| 47 |
+
json={
|
| 48 |
+
"username": USERNAME,
|
| 49 |
+
"agent_code": CODE_LINK,
|
| 50 |
+
"answers": answers
|
| 51 |
+
},
|
| 52 |
+
timeout=60 # increase timeout here too
|
| 53 |
+
)
|
| 54 |
|
| 55 |
+
if res.ok:
|
| 56 |
+
return json.dumps(res.json(), indent=2)
|
| 57 |
+
else:
|
| 58 |
+
return f"Error submitting: {res.status_code} - {res.text}"
|
| 59 |
|
| 60 |
+
except Exception as e:
|
| 61 |
+
return f"[SUBMISSION ERROR] {str(e)}"
|
| 62 |
# Gradio interface
|
| 63 |
gr.Interface(
|
| 64 |
fn=run_and_submit_all,
|