Update app.py
Browse files
app.py
CHANGED
|
@@ -2,22 +2,20 @@ import os
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
-
from smolagents import ToolCallingAgent, tool
|
| 6 |
-
from duckduckgo_search import DDGS
|
| 7 |
import math
|
| 8 |
-
from openai import OpenAI
|
| 9 |
|
| 10 |
-
|
| 11 |
-
|
|
|
|
| 12 |
|
| 13 |
# --- Tools ---
|
| 14 |
@tool
|
| 15 |
def web_search(query: str) -> str:
|
| 16 |
"""Performs a web search using DuckDuckGo.
|
| 17 |
-
|
| 18 |
Args:
|
| 19 |
query: The search query string.
|
| 20 |
-
|
| 21 |
Returns:
|
| 22 |
A formatted string with search results.
|
| 23 |
"""
|
|
@@ -34,27 +32,26 @@ def web_search(query: str) -> str:
|
|
| 34 |
@tool
|
| 35 |
def calculate(expression: str) -> str:
|
| 36 |
"""Evaluates mathematical expressions.
|
| 37 |
-
|
| 38 |
Args:
|
| 39 |
expression: The math expression to evaluate.
|
| 40 |
-
|
| 41 |
Returns:
|
| 42 |
The result as a string or error message.
|
| 43 |
"""
|
| 44 |
try:
|
| 45 |
-
safe_dict = {k: v for k, v in math.__dict__.items()
|
| 46 |
-
if not k.startswith("__")}
|
| 47 |
safe_dict.update({
|
| 48 |
'__builtins__': None,
|
| 49 |
'abs': abs,
|
| 50 |
'round': round
|
| 51 |
})
|
| 52 |
-
result = eval(expression, {
|
| 53 |
return str(result)
|
| 54 |
except Exception as e:
|
| 55 |
return f"Calculation error: {str(e)}"
|
| 56 |
|
| 57 |
-
# --- Agent
|
| 58 |
class GAIAAgent:
|
| 59 |
def __init__(self):
|
| 60 |
self.system_prompt = """You are an AI assistant that answers questions using tools:
|
|
@@ -63,13 +60,11 @@ class GAIAAgent:
|
|
| 63 |
- Be concise and accurate"""
|
| 64 |
|
| 65 |
try:
|
| 66 |
-
# Initialize with proper configuration
|
| 67 |
self.agent = ToolCallingAgent(
|
| 68 |
name="GAIA_Agent",
|
| 69 |
description=self.system_prompt,
|
| 70 |
tools=[web_search, calculate],
|
| 71 |
-
model=
|
| 72 |
-
model_kwargs={"model": "gpt-3.5-turbo"}, # Specify model here
|
| 73 |
planning_interval=3
|
| 74 |
)
|
| 75 |
print("✅ Agent initialized successfully")
|
|
@@ -77,35 +72,30 @@ class GAIAAgent:
|
|
| 77 |
raise RuntimeError(f"Agent init failed: {str(e)}")
|
| 78 |
|
| 79 |
def __call__(self, question: str) -> str:
|
| 80 |
-
"""Process a question with proper error handling."""
|
| 81 |
try:
|
| 82 |
response = self.agent.run(question)
|
| 83 |
return str(response) if response else "No answer generated"
|
| 84 |
except Exception as e:
|
| 85 |
return f"Error: {str(e)}"
|
| 86 |
|
| 87 |
-
|
| 88 |
-
# --- Gradio Interface ---
|
| 89 |
def submit_answers(profile: gr.OAuthProfile | None):
|
| 90 |
if not profile:
|
| 91 |
return "Please login to Hugging Face", None
|
| 92 |
|
| 93 |
try:
|
| 94 |
agent = GAIAAgent()
|
| 95 |
-
response = requests.get(
|
| 96 |
-
"https://agents-course-unit4-scoring.hf.space/questions",
|
| 97 |
-
timeout=20
|
| 98 |
-
)
|
| 99 |
questions = response.json() or []
|
| 100 |
-
|
| 101 |
answers = []
|
| 102 |
results = []
|
| 103 |
-
|
| 104 |
-
for item in questions[:15]:
|
| 105 |
task_id = item.get("task_id", "")
|
| 106 |
question = item.get("question", "")
|
| 107 |
answer = agent(question)
|
| 108 |
-
|
| 109 |
answers.append({
|
| 110 |
"task_id": task_id,
|
| 111 |
"submitted_answer": answer[:1000]
|
|
@@ -115,7 +105,7 @@ def submit_answers(profile: gr.OAuthProfile | None):
|
|
| 115 |
"Question": question[:100],
|
| 116 |
"Answer": answer[:200]
|
| 117 |
})
|
| 118 |
-
|
| 119 |
submit_response = requests.post(
|
| 120 |
"https://agents-course-unit4-scoring.hf.space/submit",
|
| 121 |
json={
|
|
@@ -126,24 +116,25 @@ def submit_answers(profile: gr.OAuthProfile | None):
|
|
| 126 |
timeout=60
|
| 127 |
)
|
| 128 |
data = submit_response.json()
|
| 129 |
-
|
| 130 |
return (
|
| 131 |
f"Submitted {len(answers)} answers\n"
|
| 132 |
f"Score: {data.get('score', 'N/A')}%\n"
|
| 133 |
f"Correct: {data.get('correct_count', 0)}/{len(answers)}",
|
| 134 |
pd.DataFrame(results)
|
| 135 |
)
|
| 136 |
-
|
| 137 |
except Exception as e:
|
| 138 |
return f"Error: {str(e)}", None
|
| 139 |
|
|
|
|
| 140 |
with gr.Blocks() as demo:
|
| 141 |
-
gr.Markdown("# GAIA Agent")
|
| 142 |
gr.LoginButton()
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
|
| 148 |
if __name__ == "__main__":
|
| 149 |
demo.launch()
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
|
|
|
|
|
|
| 5 |
import math
|
|
|
|
| 6 |
|
| 7 |
+
from smolagents import ToolCallingAgent, tool
|
| 8 |
+
from instructor import OpenAIChat
|
| 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 |
"""
|
|
|
|
| 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: {str(e)}"
|
| 53 |
|
| 54 |
+
# --- Agent ---
|
| 55 |
class GAIAAgent:
|
| 56 |
def __init__(self):
|
| 57 |
self.system_prompt = """You are an AI assistant that answers questions using tools:
|
|
|
|
| 60 |
- Be concise and accurate"""
|
| 61 |
|
| 62 |
try:
|
|
|
|
| 63 |
self.agent = ToolCallingAgent(
|
| 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")
|
|
|
|
| 72 |
raise RuntimeError(f"Agent init failed: {str(e)}")
|
| 73 |
|
| 74 |
def __call__(self, question: str) -> str:
|
|
|
|
| 75 |
try:
|
| 76 |
response = self.agent.run(question)
|
| 77 |
return str(response) if response else "No answer generated"
|
| 78 |
except Exception as e:
|
| 79 |
return f"Error: {str(e)}"
|
| 80 |
|
| 81 |
+
# --- Gradio Submission Logic ---
|
|
|
|
| 82 |
def submit_answers(profile: gr.OAuthProfile | None):
|
| 83 |
if not profile:
|
| 84 |
return "Please login to Hugging Face", None
|
| 85 |
|
| 86 |
try:
|
| 87 |
agent = GAIAAgent()
|
| 88 |
+
response = requests.get("https://agents-course-unit4-scoring.hf.space/questions", timeout=20)
|
|
|
|
|
|
|
|
|
|
| 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 |
answers.append({
|
| 100 |
"task_id": task_id,
|
| 101 |
"submitted_answer": answer[:1000]
|
|
|
|
| 105 |
"Question": question[:100],
|
| 106 |
"Answer": answer[:200]
|
| 107 |
})
|
| 108 |
+
|
| 109 |
submit_response = requests.post(
|
| 110 |
"https://agents-course-unit4-scoring.hf.space/submit",
|
| 111 |
json={
|
|
|
|
| 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 |
except Exception as e:
|
| 128 |
return f"Error: {str(e)}", None
|
| 129 |
|
| 130 |
+
# --- Gradio Interface ---
|
| 131 |
with gr.Blocks() as demo:
|
| 132 |
+
gr.Markdown("# GAIA Agent with DuckDuckGo + Calculator")
|
| 133 |
gr.LoginButton()
|
| 134 |
+
btn = gr.Button("Run Evaluation", variant="primary")
|
| 135 |
+
status = gr.Textbox(label="Results")
|
| 136 |
+
df = gr.DataFrame(label="Evaluation Details")
|
| 137 |
+
btn.click(submit_answers, outputs=[status, df])
|
| 138 |
|
| 139 |
if __name__ == "__main__":
|
| 140 |
demo.launch()
|