Update app.py
Browse files
app.py
CHANGED
@@ -2,128 +2,96 @@ import os
|
|
2 |
import gradio as gr
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
-
|
6 |
-
from
|
7 |
-
from
|
8 |
-
from smolagents.tools import DuckDuckGoSearchTool, WebSearchTool, WikipediaSearchTool
|
9 |
|
10 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
11 |
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
web_tool = WebSearchTool()
|
17 |
-
agent = ToolCallingAgent(
|
18 |
-
tools=[wiki_tool, duck_tool, web_tool],
|
19 |
-
model=model
|
20 |
-
)
|
21 |
-
return agent
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
|
|
|
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
33 |
|
|
|
34 |
try:
|
35 |
-
|
|
|
|
|
|
|
36 |
except Exception as e:
|
37 |
-
return f"Error
|
38 |
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
try:
|
42 |
-
|
43 |
-
response.raise_for_status()
|
44 |
-
questions_data = response.json()
|
45 |
-
if not questions_data:
|
46 |
-
return "Fetched questions list is empty or invalid format.", None
|
47 |
except Exception as e:
|
48 |
-
return f"Error
|
|
|
|
|
|
|
49 |
|
50 |
-
results_log = []
|
51 |
-
answers_payload = []
|
52 |
for item in questions_data:
|
53 |
task_id = item.get("task_id")
|
54 |
question_text = item.get("question")
|
55 |
-
if
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
60 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
61 |
-
except Exception as e:
|
62 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"(Agent error: {e})"})
|
63 |
-
|
64 |
-
if not answers_payload:
|
65 |
-
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
66 |
-
|
67 |
-
submission_data = {
|
68 |
-
"username": username.strip(),
|
69 |
-
"agent_code": agent_code,
|
70 |
-
"answers": answers_payload
|
71 |
-
}
|
72 |
|
73 |
-
|
74 |
-
|
75 |
-
response.raise_for_status()
|
76 |
-
result_data = response.json()
|
77 |
-
final_status = (
|
78 |
-
f"Submission Successful!\n"
|
79 |
-
f"User: {result_data.get('username')}\n"
|
80 |
-
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
81 |
-
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
82 |
-
f"Message: {result_data.get('message', 'No message received.')}"
|
83 |
-
)
|
84 |
-
results_df = pd.DataFrame(results_log)
|
85 |
-
return final_status, results_df
|
86 |
-
except Exception as e:
|
87 |
-
status_message = f"Submission Failed: {e}"
|
88 |
-
results_df = pd.DataFrame(results_log)
|
89 |
-
return status_message, results_df
|
90 |
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
""
|
95 |
-
|
96 |
-
|
97 |
-
- Log in to your Hugging Face account with the button below.
|
98 |
-
- Click 'Run Evaluation & Submit All Answers' to begin.
|
99 |
-
Disclaimer: Submission may take a while depending on the number of questions and agent speed.
|
100 |
-
"""
|
101 |
)
|
|
|
|
|
|
|
|
|
|
|
102 |
gr.LoginButton()
|
103 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
104 |
-
status_output = gr.Textbox(label="Run Status
|
105 |
-
results_table = gr.DataFrame(label="Questions and
|
106 |
|
107 |
-
run_button.click(
|
108 |
-
fn=run_and_submit_all,
|
109 |
-
outputs=[status_output, results_table]
|
110 |
-
)
|
111 |
|
112 |
if __name__ == "__main__":
|
113 |
-
|
114 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
115 |
-
space_id_startup = os.getenv("SPACE_ID")
|
116 |
-
if space_host_startup:
|
117 |
-
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
118 |
-
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
119 |
-
else:
|
120 |
-
print("ℹ️ SPACE_HOST not found (running locally?)")
|
121 |
-
if space_id_startup:
|
122 |
-
print(f"✅ SPACE_ID found: {space_id_startup}")
|
123 |
-
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
|
124 |
-
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
125 |
-
else:
|
126 |
-
print("ℹ️ SPACE_ID not found")
|
127 |
-
print("-"*(60 + len(" App Starting ")) + "\n")
|
128 |
-
print("Launching Gradio Interface for SmolAgent Evaluation...")
|
129 |
-
demo.launch(debug=True, share=False)
|
|
|
2 |
import gradio as gr
|
3 |
import requests
|
4 |
import pandas as pd
|
5 |
+
from huggingface_hub import InferenceClient
|
6 |
+
from duckduckgo_search import DDGS
|
7 |
+
from datasets import load_dataset
|
|
|
8 |
|
9 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
10 |
|
11 |
+
# Hugging Face Token (set in environment)
|
12 |
+
HF_TOKEN = os.environ.get("HF_TOKEN")
|
13 |
+
deepseek_model = "deepseek-ai/DeepSeek-R1"
|
14 |
+
hf_client = InferenceClient(model=deepseek_model, token=HF_TOKEN)
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
+
# Load Wikipedia dataset (small subset for efficient retrieval)
|
17 |
+
wiki_dataset = load_dataset("wikipedia", "20220301.en", split="train[:10000]")
|
18 |
+
|
19 |
+
def search_wikipedia(question):
|
20 |
+
results = wiki_dataset.filter(lambda x: question.lower() in x["text"].lower())
|
21 |
+
if len(results):
|
22 |
+
return results[0]["text"][:1000] # limit to first 1000 chars
|
23 |
+
return "No relevant information found on Wikipedia."
|
24 |
|
25 |
+
def duckduckgo_search(query):
|
26 |
+
with DDGS() as ddgs:
|
27 |
+
results = [r["body"] for r in ddgs.text(query, max_results=3)]
|
28 |
+
return "\n".join(results) if results else "No results found."
|
29 |
|
30 |
+
def ask_deepseek(prompt, max_tokens=512):
|
31 |
try:
|
32 |
+
response = hf_client.text_generation(
|
33 |
+
prompt, max_new_tokens=max_tokens, temperature=0.2, repetition_penalty=1.1
|
34 |
+
)
|
35 |
+
return response
|
36 |
except Exception as e:
|
37 |
+
return f"DeepSeek Error: {e}"
|
38 |
|
39 |
+
class SmartAgent:
|
40 |
+
def __call__(self, question: str) -> str:
|
41 |
+
q_lower = question.lower()
|
42 |
+
if any(term in q_lower for term in ["current", "latest", "2024", "2025", "recent", "live", "today", "now"]):
|
43 |
+
return duckduckgo_search(question)
|
44 |
+
deepseek_response = ask_deepseek(question)
|
45 |
+
if "DeepSeek Error" not in deepseek_response and deepseek_response.strip():
|
46 |
+
return deepseek_response
|
47 |
+
# fallback to Wikipedia if DeepSeek fails
|
48 |
+
return search_wikipedia(question)
|
49 |
|
50 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
51 |
+
if not profile:
|
52 |
+
return "Please Login to Hugging Face with the button.", None
|
53 |
+
username = profile.username
|
54 |
+
questions_url = f"{DEFAULT_API_URL}/questions"
|
55 |
+
submit_url = f"{DEFAULT_API_URL}/submit"
|
56 |
+
agent_code = f"https://huggingface.co/spaces/{os.getenv('SPACE_ID')}/tree/main"
|
57 |
+
|
58 |
try:
|
59 |
+
agent = SmartAgent()
|
|
|
|
|
|
|
|
|
60 |
except Exception as e:
|
61 |
+
return f"Agent Error: {e}", None
|
62 |
+
|
63 |
+
questions_data = requests.get(questions_url).json()
|
64 |
+
results_log, answers_payload = [], []
|
65 |
|
|
|
|
|
66 |
for item in questions_data:
|
67 |
task_id = item.get("task_id")
|
68 |
question_text = item.get("question")
|
69 |
+
if task_id and question_text:
|
70 |
+
answer = agent(question_text)
|
71 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
72 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": answer})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
+
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
75 |
+
response = requests.post(submit_url, json=submission_data).json()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
+
final_status = (
|
78 |
+
f"Submission Successful!\n"
|
79 |
+
f"User: {response.get('username')}\n"
|
80 |
+
f"Overall Score: {response.get('score', 'N/A')}%\n"
|
81 |
+
f"({response.get('correct_count', '?')}/{response.get('total_attempted', '?')} correct)\n"
|
82 |
+
f"Message: {response.get('message', 'No message received.')}"
|
|
|
|
|
|
|
|
|
83 |
)
|
84 |
+
|
85 |
+
return final_status, pd.DataFrame(results_log)
|
86 |
+
|
87 |
+
with gr.Blocks() as demo:
|
88 |
+
gr.Markdown("# Smart Agent Evaluation Runner")
|
89 |
gr.LoginButton()
|
90 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
91 |
+
status_output = gr.Textbox(label="Run Status", lines=5, interactive=False)
|
92 |
+
results_table = gr.DataFrame(label="Questions and Answers")
|
93 |
|
94 |
+
run_button.click(run_and_submit_all, outputs=[status_output, results_table])
|
|
|
|
|
|
|
95 |
|
96 |
if __name__ == "__main__":
|
97 |
+
demo.launch(debug=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|