Update app.py
Browse files
app.py
CHANGED
@@ -5,20 +5,13 @@ import pandas as pd
|
|
5 |
import math
|
6 |
|
7 |
from smolagents import ToolCallingAgent, tool
|
8 |
-
from
|
9 |
from duckduckgo_search import DDGS
|
10 |
|
11 |
# --- Tools ---
|
12 |
@tool
|
13 |
def web_search(query: str) -> str:
|
14 |
-
"""Performs a web search using DuckDuckGo.
|
15 |
-
|
16 |
-
Args:
|
17 |
-
query: The search query string.
|
18 |
-
|
19 |
-
Returns:
|
20 |
-
A formatted string with search results.
|
21 |
-
"""
|
22 |
try:
|
23 |
with DDGS() as ddgs:
|
24 |
results = ddgs.text(query, max_results=3)
|
@@ -27,114 +20,76 @@ def web_search(query: str) -> str:
|
|
27 |
for r in results
|
28 |
) if results else "No results found."
|
29 |
except Exception as e:
|
30 |
-
return f"Search error: {
|
31 |
|
32 |
@tool
|
33 |
def calculate(expression: str) -> str:
|
34 |
-
"""Evaluates mathematical expressions.
|
35 |
-
|
36 |
-
Args:
|
37 |
-
expression: The math expression to evaluate.
|
38 |
-
|
39 |
-
Returns:
|
40 |
-
The result as a string or error message.
|
41 |
-
"""
|
42 |
try:
|
43 |
safe_dict = {k: v for k, v in math.__dict__.items() if not k.startswith("__")}
|
44 |
-
safe_dict.update({
|
45 |
-
'__builtins__': None,
|
46 |
-
'abs': abs,
|
47 |
-
'round': round
|
48 |
-
})
|
49 |
result = eval(expression, {"__builtins__": None}, safe_dict)
|
50 |
return str(result)
|
51 |
except Exception as e:
|
52 |
-
return f"Calculation error: {
|
53 |
|
54 |
# --- Agent ---
|
55 |
class GAIAAgent:
|
56 |
def __init__(self):
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
name="GAIA_Agent",
|
65 |
-
description=self.system_prompt,
|
66 |
-
tools=[web_search, calculate],
|
67 |
-
model=OpenAIChat(model="gpt-3.5-turbo"), # ✅ Fixed model wrapper
|
68 |
-
planning_interval=3
|
69 |
-
)
|
70 |
-
print("✅ Agent initialized successfully")
|
71 |
-
except Exception as e:
|
72 |
-
raise RuntimeError(f"Agent init failed: {str(e)}")
|
73 |
|
74 |
def __call__(self, question: str) -> str:
|
75 |
try:
|
76 |
-
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
except Exception as e:
|
79 |
-
return f"Error: {
|
80 |
|
81 |
-
# --- Gradio Submission
|
82 |
def submit_answers(profile: gr.OAuthProfile | None):
|
83 |
if not profile:
|
84 |
return "Please login to Hugging Face", None
|
85 |
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
questions = response.json() or []
|
90 |
-
|
91 |
-
answers = []
|
92 |
-
results = []
|
93 |
-
|
94 |
-
for item in questions[:15]: # Limit to 15 for speed
|
95 |
-
task_id = item.get("task_id", "")
|
96 |
-
question = item.get("question", "")
|
97 |
-
answer = agent(question)
|
98 |
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
"Question": question[:100],
|
106 |
-
"Answer": answer[:200]
|
107 |
-
})
|
108 |
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
},
|
116 |
-
timeout=60
|
117 |
-
)
|
118 |
-
data = submit_response.json()
|
119 |
-
|
120 |
-
return (
|
121 |
-
f"Submitted {len(answers)} answers\n"
|
122 |
-
f"Score: {data.get('score', 'N/A')}%\n"
|
123 |
-
f"Correct: {data.get('correct_count', 0)}/{len(answers)}",
|
124 |
-
pd.DataFrame(results)
|
125 |
-
)
|
126 |
|
127 |
-
|
128 |
-
|
129 |
|
130 |
-
# --- Gradio Interface ---
|
131 |
with gr.Blocks() as demo:
|
132 |
-
gr.Markdown("# GAIA Agent
|
133 |
gr.LoginButton()
|
134 |
btn = gr.Button("Run Evaluation", variant="primary")
|
135 |
status = gr.Textbox(label="Results")
|
136 |
-
df = gr.DataFrame(label="
|
137 |
btn.click(submit_answers, outputs=[status, df])
|
138 |
|
139 |
if __name__ == "__main__":
|
140 |
-
demo.launch()
|
|
|
5 |
import math
|
6 |
|
7 |
from smolagents import ToolCallingAgent, tool
|
8 |
+
from smolagents.models import OpenAIServerModel
|
9 |
from duckduckgo_search import DDGS
|
10 |
|
11 |
# --- Tools ---
|
12 |
@tool
|
13 |
def web_search(query: str) -> str:
|
14 |
+
"""Performs a web search using DuckDuckGo."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
try:
|
16 |
with DDGS() as ddgs:
|
17 |
results = ddgs.text(query, max_results=3)
|
|
|
20 |
for r in results
|
21 |
) if results else "No results found."
|
22 |
except Exception as e:
|
23 |
+
return f"Search error: {e}"
|
24 |
|
25 |
@tool
|
26 |
def calculate(expression: str) -> str:
|
27 |
+
"""Evaluates mathematical expressions."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
try:
|
29 |
safe_dict = {k: v for k, v in math.__dict__.items() if not k.startswith("__")}
|
30 |
+
safe_dict.update({'abs': abs, 'round': round})
|
|
|
|
|
|
|
|
|
31 |
result = eval(expression, {"__builtins__": None}, safe_dict)
|
32 |
return str(result)
|
33 |
except Exception as e:
|
34 |
+
return f"Calculation error: {e}"
|
35 |
|
36 |
# --- Agent ---
|
37 |
class GAIAAgent:
|
38 |
def __init__(self):
|
39 |
+
model = OpenAIServerModel(model_id="gpt-3.5-turbo", api_key=os.getenv("OPENAI_API_KEY"))
|
40 |
+
self.agent = ToolCallingAgent(
|
41 |
+
name="GAIA_Agent",
|
42 |
+
description="AI assistant using web_search and calculate tools.",
|
43 |
+
tools=[web_search, calculate],
|
44 |
+
model=model,
|
45 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
|
47 |
def __call__(self, question: str) -> str:
|
48 |
try:
|
49 |
+
return self.agent.run(
|
50 |
+
question,
|
51 |
+
planning_interval=3,
|
52 |
+
system_prompt=(
|
53 |
+
"You are an AI assistant. Use web_search for factual queries "
|
54 |
+
"and calculate for math problems. Be concise and accurate."
|
55 |
+
)
|
56 |
+
)
|
57 |
except Exception as e:
|
58 |
+
return f"Error: {e}"
|
59 |
|
60 |
+
# --- Gradio UI + Submission ---
|
61 |
def submit_answers(profile: gr.OAuthProfile | None):
|
62 |
if not profile:
|
63 |
return "Please login to Hugging Face", None
|
64 |
|
65 |
+
agent = GAIAAgent()
|
66 |
+
resp = requests.get("https://agents-course-unit4-scoring.hf.space/questions", timeout=20)
|
67 |
+
questions = resp.json() or []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
+
answers, rows = [], []
|
70 |
+
for item in questions[:15]:
|
71 |
+
tid, q = item.get("task_id"), item.get("question")
|
72 |
+
ans = agent(q)
|
73 |
+
answers.append({"task_id": tid, "submitted_answer": ans[:1000]})
|
74 |
+
rows.append({"Task ID": tid, "Question": q[:100], "Answer": ans[:200]})
|
|
|
|
|
|
|
75 |
|
76 |
+
post = requests.post("https://agents-course-unit4-scoring.hf.space/submit", json={
|
77 |
+
"username": profile.username,
|
78 |
+
"agent_code": f"https://huggingface.co/spaces/{os.getenv('SPACE_ID')}",
|
79 |
+
"answers": answers
|
80 |
+
}, timeout=60)
|
81 |
+
data = post.json()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
|
83 |
+
out = f"Submitted {len(answers)} answers\nScore: {data.get('score', 'N/A')}%\nCorrect: {data.get('correct_count',0)}/{len(answers)}"
|
84 |
+
return out, pd.DataFrame(rows)
|
85 |
|
|
|
86 |
with gr.Blocks() as demo:
|
87 |
+
gr.Markdown("# GAIA Agent")
|
88 |
gr.LoginButton()
|
89 |
btn = gr.Button("Run Evaluation", variant="primary")
|
90 |
status = gr.Textbox(label="Results")
|
91 |
+
df = gr.DataFrame(label="Details")
|
92 |
btn.click(submit_answers, outputs=[status, df])
|
93 |
|
94 |
if __name__ == "__main__":
|
95 |
+
demo.launch()
|