Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
# app.py
|
2 |
import os
|
3 |
import gradio as gr
|
4 |
import requests
|
@@ -6,36 +5,35 @@ import pandas as pd
|
|
6 |
|
7 |
from tools import AnswerTool
|
8 |
from smolagents import CodeAgent, OpenAIServerModel
|
9 |
-
from smolagents import DuckDuckGoSearchTool
|
10 |
|
11 |
-
#
|
12 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
13 |
|
14 |
class BasicAgent:
|
15 |
def __init__(self):
|
16 |
-
# Initialize CodeAgent with GPT-4o, AnswerTool
|
17 |
-
model
|
18 |
answer_tool = AnswerTool()
|
19 |
web_tool = DuckDuckGoSearchTool()
|
|
|
20 |
self.agent = CodeAgent(
|
21 |
model=model,
|
22 |
-
tools=[answer_tool, web_tool],
|
23 |
-
add_base_tools=
|
24 |
-
max_steps=2,
|
25 |
verbosity_level=0
|
26 |
)
|
27 |
|
28 |
def __call__(self, question: str) -> str:
|
29 |
-
#
|
30 |
return self.agent.run(question)
|
31 |
|
32 |
-
|
33 |
def run_and_submit_all(username):
|
34 |
-
# Username fornito manualmente
|
35 |
if not username:
|
36 |
return "Please enter your Hugging Face username.", None
|
37 |
|
38 |
-
# Fetch questions
|
39 |
try:
|
40 |
resp = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15)
|
41 |
if resp.status_code == 429:
|
@@ -45,30 +43,30 @@ def run_and_submit_all(username):
|
|
45 |
except Exception as e:
|
46 |
return f"Error fetching questions: {e}", None
|
47 |
|
48 |
-
# Run agent on all questions
|
49 |
-
agent
|
50 |
results = []
|
51 |
payload = []
|
52 |
for q in questions:
|
53 |
-
tid
|
54 |
-
text = q.get(
|
55 |
if not (tid and text):
|
56 |
continue
|
57 |
try:
|
58 |
ans = agent(text)
|
59 |
except Exception as e:
|
60 |
ans = f"ERROR: {e}"
|
61 |
-
results.append({
|
62 |
-
payload.append({
|
63 |
|
64 |
if not payload:
|
65 |
return "Agent returned no answers.", pd.DataFrame(results)
|
66 |
|
67 |
-
# Submit answers
|
68 |
submission = {
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
}
|
73 |
try:
|
74 |
sub_resp = requests.post(f"{DEFAULT_API_URL}/submit", json=submission, timeout=60)
|
@@ -85,20 +83,18 @@ def run_and_submit_all(username):
|
|
85 |
|
86 |
return status, pd.DataFrame(results)
|
87 |
|
88 |
-
|
89 |
def test_random_question(username):
|
90 |
if not username:
|
91 |
return "Please enter your Hugging Face username.", ""
|
92 |
try:
|
93 |
q = requests.get(f"{DEFAULT_API_URL}/random-question", timeout=15).json()
|
94 |
-
question = q.get(
|
95 |
-
ans
|
96 |
return question, ans
|
97 |
except Exception as e:
|
98 |
return f"Error during test: {e}", ""
|
99 |
|
100 |
-
|
101 |
-
# --- Gradio UI ---
|
102 |
with gr.Blocks() as demo:
|
103 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
104 |
gr.Markdown(
|
@@ -111,8 +107,8 @@ with gr.Blocks() as demo:
|
|
111 |
)
|
112 |
|
113 |
username_input = gr.Textbox(label="Hugging Face Username", placeholder="your-username")
|
114 |
-
run_btn
|
115 |
-
test_btn
|
116 |
|
117 |
status_out = gr.Textbox(label="Status / Result", lines=5, interactive=False)
|
118 |
table_out = gr.DataFrame(label="Full Results Table", wrap=True)
|
|
|
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
import requests
|
|
|
5 |
|
6 |
from tools import AnswerTool
|
7 |
from smolagents import CodeAgent, OpenAIServerModel
|
8 |
+
from smolagents import DuckDuckGoSearchTool, WikipediaSearchTool
|
9 |
|
10 |
+
# Constants
|
11 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
12 |
|
13 |
class BasicAgent:
|
14 |
def __init__(self):
|
15 |
+
# Initialize CodeAgent with GPT-4o, custom AnswerTool, DuckDuckGo and Wikipedia tools
|
16 |
+
model = OpenAIServerModel(model_id="gpt-4o")
|
17 |
answer_tool = AnswerTool()
|
18 |
web_tool = DuckDuckGoSearchTool()
|
19 |
+
wiki_tool = WikipediaSearchTool()
|
20 |
self.agent = CodeAgent(
|
21 |
model=model,
|
22 |
+
tools=[answer_tool, web_tool, wiki_tool],
|
23 |
+
add_base_tools=True,
|
24 |
+
max_steps=2,
|
25 |
verbosity_level=0
|
26 |
)
|
27 |
|
28 |
def __call__(self, question: str) -> str:
|
29 |
+
# Run the agent on the question
|
30 |
return self.agent.run(question)
|
31 |
|
|
|
32 |
def run_and_submit_all(username):
|
|
|
33 |
if not username:
|
34 |
return "Please enter your Hugging Face username.", None
|
35 |
|
36 |
+
# 1. Fetch questions
|
37 |
try:
|
38 |
resp = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15)
|
39 |
if resp.status_code == 429:
|
|
|
43 |
except Exception as e:
|
44 |
return f"Error fetching questions: {e}", None
|
45 |
|
46 |
+
# 2. Run agent on all questions
|
47 |
+
agent = BasicAgent()
|
48 |
results = []
|
49 |
payload = []
|
50 |
for q in questions:
|
51 |
+
tid = q.get("task_id")
|
52 |
+
text = q.get("question")
|
53 |
if not (tid and text):
|
54 |
continue
|
55 |
try:
|
56 |
ans = agent(text)
|
57 |
except Exception as e:
|
58 |
ans = f"ERROR: {e}"
|
59 |
+
results.append({"Task ID": tid, "Question": text, "Answer": ans})
|
60 |
+
payload.append({"task_id": tid, "submitted_answer": ans})
|
61 |
|
62 |
if not payload:
|
63 |
return "Agent returned no answers.", pd.DataFrame(results)
|
64 |
|
65 |
+
# 3. Submit answers
|
66 |
submission = {
|
67 |
+
"username": username,
|
68 |
+
"agent_code": f"https://huggingface.co/spaces/{os.getenv('SPACE_ID')}/tree/main",
|
69 |
+
"answers": payload
|
70 |
}
|
71 |
try:
|
72 |
sub_resp = requests.post(f"{DEFAULT_API_URL}/submit", json=submission, timeout=60)
|
|
|
83 |
|
84 |
return status, pd.DataFrame(results)
|
85 |
|
|
|
86 |
def test_random_question(username):
|
87 |
if not username:
|
88 |
return "Please enter your Hugging Face username.", ""
|
89 |
try:
|
90 |
q = requests.get(f"{DEFAULT_API_URL}/random-question", timeout=15).json()
|
91 |
+
question = q.get("question", "")
|
92 |
+
ans = BasicAgent()(question)
|
93 |
return question, ans
|
94 |
except Exception as e:
|
95 |
return f"Error during test: {e}", ""
|
96 |
|
97 |
+
# Build Gradio interface
|
|
|
98 |
with gr.Blocks() as demo:
|
99 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
100 |
gr.Markdown(
|
|
|
107 |
)
|
108 |
|
109 |
username_input = gr.Textbox(label="Hugging Face Username", placeholder="your-username")
|
110 |
+
run_btn = gr.Button("Run Evaluation & Submit All Answers")
|
111 |
+
test_btn = gr.Button("Test Random Question")
|
112 |
|
113 |
status_out = gr.Textbox(label="Status / Result", lines=5, interactive=False)
|
114 |
table_out = gr.DataFrame(label="Full Results Table", wrap=True)
|