Spaces:
Sleeping
Sleeping
fixes
Browse files
app.py
CHANGED
@@ -3,69 +3,75 @@ import gradio as gr
|
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
|
6 |
-
from tools import
|
7 |
-
from smolagents import CodeAgent
|
8 |
|
9 |
# --- Constants ---
|
10 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
11 |
|
12 |
class BasicAgent:
|
13 |
def __init__(self):
|
14 |
-
#
|
15 |
-
model = OpenAIServerModel(model_id="gpt-4o")
|
16 |
-
final_tool = FinalAnswerTool()
|
17 |
self.agent = CodeAgent(
|
18 |
-
model=
|
19 |
-
tools=[
|
20 |
-
|
21 |
-
|
|
|
22 |
)
|
23 |
|
24 |
def __call__(self, question: str) -> str:
|
25 |
-
#
|
26 |
return self.agent.run(question)
|
27 |
|
28 |
|
29 |
-
def run_and_submit_all(
|
30 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
if not username:
|
32 |
-
return "
|
33 |
|
34 |
-
# Fetch questions
|
|
|
35 |
try:
|
36 |
-
resp = requests.get(
|
37 |
resp.raise_for_status()
|
38 |
questions = resp.json()
|
39 |
except Exception as e:
|
40 |
return f"Error fetching questions: {e}", None
|
41 |
|
42 |
-
# Run agent
|
43 |
agent = BasicAgent()
|
44 |
-
results = []
|
45 |
-
payload = []
|
46 |
for q in questions:
|
47 |
-
|
48 |
-
text = q.get(
|
49 |
-
if not
|
50 |
continue
|
51 |
try:
|
52 |
ans = agent(text)
|
53 |
except Exception as e:
|
54 |
ans = f"ERROR: {e}"
|
55 |
-
results.append({
|
56 |
-
payload.append({
|
57 |
|
58 |
if not payload:
|
59 |
return "Agent returned no answers.", pd.DataFrame(results)
|
60 |
|
61 |
-
# Submit
|
|
|
62 |
submission = {
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
}
|
67 |
try:
|
68 |
-
sub_resp = requests.post(
|
69 |
sub_resp.raise_for_status()
|
70 |
data = sub_resp.json()
|
71 |
status = (
|
@@ -80,30 +86,33 @@ def run_and_submit_all(username):
|
|
80 |
return status, pd.DataFrame(results)
|
81 |
|
82 |
|
83 |
-
def test_random_question(
|
84 |
-
|
85 |
-
|
86 |
-
|
|
|
|
|
87 |
try:
|
88 |
q = requests.get(f"{DEFAULT_API_URL}/random-question", timeout=15).json()
|
89 |
-
|
90 |
-
|
|
|
91 |
except Exception as e:
|
92 |
return f"Error during test: {e}", ""
|
93 |
|
94 |
-
# --- Gradio
|
95 |
with gr.Blocks() as demo:
|
96 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
97 |
gr.Markdown(
|
98 |
"""
|
99 |
**Instructions:**
|
100 |
-
1.
|
101 |
-
2.
|
102 |
-
3. Use **Run Evaluation & Submit All Answers**
|
103 |
"""
|
104 |
)
|
105 |
|
106 |
-
|
107 |
run_btn = gr.Button("Run Evaluation & Submit All Answers")
|
108 |
test_btn = gr.Button("Test Random Question")
|
109 |
|
@@ -112,8 +121,9 @@ with gr.Blocks() as demo:
|
|
112 |
question_out = gr.Textbox(label="Random Question", lines=3, interactive=False)
|
113 |
answer_out = gr.Textbox(label="Agent Answer", lines=3, interactive=False)
|
114 |
|
115 |
-
|
116 |
-
|
|
|
117 |
|
118 |
if __name__ == "__main__":
|
119 |
demo.launch(debug=True, share=False)
|
|
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
|
6 |
+
from tools import AnswerTool
|
7 |
+
from smolagents import CodeAgent
|
8 |
|
9 |
# --- Constants ---
|
10 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
11 |
|
12 |
class BasicAgent:
|
13 |
def __init__(self):
|
14 |
+
# Initialize CodeAgent with a single custom AnswerTool to handle GAIA Level 1 questions
|
|
|
|
|
15 |
self.agent = CodeAgent(
|
16 |
+
model=None,
|
17 |
+
tools=[AnswerTool()],
|
18 |
+
add_base_tools=False,
|
19 |
+
max_steps=1,
|
20 |
+
verbosity_level=0
|
21 |
)
|
22 |
|
23 |
def __call__(self, question: str) -> str:
|
24 |
+
# Directly run the agent on the question (single-step tool invocation)
|
25 |
return self.agent.run(question)
|
26 |
|
27 |
|
28 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
29 |
+
"""
|
30 |
+
Fetch all GAIA Level 1 questions, run the BasicAgent, submit answers, and display results.
|
31 |
+
"""
|
32 |
+
space_id = os.getenv("SPACE_ID")
|
33 |
+
if not profile:
|
34 |
+
return "Please login to Hugging Face with the login button.", None
|
35 |
+
username = getattr(profile, "username", None) or getattr(profile, "name", None)
|
36 |
if not username:
|
37 |
+
return "Login error: username not found.", None
|
38 |
|
39 |
+
# 1. Fetch questions
|
40 |
+
questions_url = f"{DEFAULT_API_URL}/questions"
|
41 |
try:
|
42 |
+
resp = requests.get(questions_url, timeout=15)
|
43 |
resp.raise_for_status()
|
44 |
questions = resp.json()
|
45 |
except Exception as e:
|
46 |
return f"Error fetching questions: {e}", None
|
47 |
|
48 |
+
# 2. Run agent on each question
|
49 |
agent = BasicAgent()
|
50 |
+
results, payload = [], []
|
|
|
51 |
for q in questions:
|
52 |
+
task_id = q.get("task_id")
|
53 |
+
text = q.get("question")
|
54 |
+
if not task_id or not text:
|
55 |
continue
|
56 |
try:
|
57 |
ans = agent(text)
|
58 |
except Exception as e:
|
59 |
ans = f"ERROR: {e}"
|
60 |
+
results.append({"Task ID": task_id, "Question": text, "Answer": ans})
|
61 |
+
payload.append({"task_id": task_id, "submitted_answer": ans})
|
62 |
|
63 |
if not payload:
|
64 |
return "Agent returned no answers.", pd.DataFrame(results)
|
65 |
|
66 |
+
# 3. Submit answers
|
67 |
+
submit_url = f"{DEFAULT_API_URL}/submit"
|
68 |
submission = {
|
69 |
+
"username": username.strip(),
|
70 |
+
"agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
|
71 |
+
"answers": payload
|
72 |
}
|
73 |
try:
|
74 |
+
sub_resp = requests.post(submit_url, json=submission, timeout=60)
|
75 |
sub_resp.raise_for_status()
|
76 |
data = sub_resp.json()
|
77 |
status = (
|
|
|
86 |
return status, pd.DataFrame(results)
|
87 |
|
88 |
|
89 |
+
def test_random_question(profile: gr.OAuthProfile | None):
|
90 |
+
"""
|
91 |
+
Fetch a single random GAIA question and return the agent's answer.
|
92 |
+
"""
|
93 |
+
if not profile:
|
94 |
+
return "Please login to Hugging Face with the login button.", ""
|
95 |
try:
|
96 |
q = requests.get(f"{DEFAULT_API_URL}/random-question", timeout=15).json()
|
97 |
+
question = q.get("question", "")
|
98 |
+
ans = BasicAgent()(question)
|
99 |
+
return question, ans
|
100 |
except Exception as e:
|
101 |
return f"Error during test: {e}", ""
|
102 |
|
103 |
+
# --- Gradio Interface ---
|
104 |
with gr.Blocks() as demo:
|
105 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
106 |
gr.Markdown(
|
107 |
"""
|
108 |
**Instructions:**
|
109 |
+
1. Clone this space and define your agent logic in `tools.py`.
|
110 |
+
2. Log in with your Hugging Face account using the login button below.
|
111 |
+
3. Use **Run Evaluation & Submit All Answers** or **Test Random Question**.
|
112 |
"""
|
113 |
)
|
114 |
|
115 |
+
login = gr.LoginButton()
|
116 |
run_btn = gr.Button("Run Evaluation & Submit All Answers")
|
117 |
test_btn = gr.Button("Test Random Question")
|
118 |
|
|
|
121 |
question_out = gr.Textbox(label="Random Question", lines=3, interactive=False)
|
122 |
answer_out = gr.Textbox(label="Agent Answer", lines=3, interactive=False)
|
123 |
|
124 |
+
# Wire buttons to callbacks; LoginButton auto-passes profile
|
125 |
+
run_btn.click(fn=run_and_submit_all, inputs=[login], outputs=[status_out, table_out])
|
126 |
+
test_btn.click(fn=test_random_question, inputs=[login], outputs=[question_out, answer_out])
|
127 |
|
128 |
if __name__ == "__main__":
|
129 |
demo.launch(debug=True, share=False)
|