Spaces:
Sleeping
Sleeping
test3
Browse files
app.py
CHANGED
|
@@ -2,131 +2,110 @@ import os
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
-
import datetime
|
| 6 |
-
from openai import OpenAI
|
| 7 |
|
|
|
|
| 8 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 9 |
|
| 10 |
-
|
|
|
|
| 11 |
def __init__(self):
|
| 12 |
-
|
| 13 |
-
if not api_key:
|
| 14 |
-
raise ValueError("OPENAI_API_KEY is not set.")
|
| 15 |
-
self.client = OpenAI(api_key=api_key)
|
| 16 |
-
print("ToolEnhancedAgent initialized.")
|
| 17 |
-
|
| 18 |
-
def use_tool(self, tool_name: str, input_text: str) -> str:
|
| 19 |
-
try:
|
| 20 |
-
if tool_name == "calculator":
|
| 21 |
-
return str(eval(input_text))
|
| 22 |
-
elif tool_name == "date":
|
| 23 |
-
return str(datetime.datetime.now().date())
|
| 24 |
-
elif tool_name == "wikipedia":
|
| 25 |
-
res = requests.get(f"https://en.wikipedia.org/api/rest_v1/page/summary/{input_text}")
|
| 26 |
-
if res.status_code == 200:
|
| 27 |
-
return res.json().get("extract", "No summary found.")
|
| 28 |
-
else:
|
| 29 |
-
return "No summary found."
|
| 30 |
-
else:
|
| 31 |
-
return "[Unknown Tool]"
|
| 32 |
-
except Exception as e:
|
| 33 |
-
return f"[Tool Error: {e}]"
|
| 34 |
|
| 35 |
def __call__(self, question: str) -> str:
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
You are an advanced AI assistant with access to tools like calculator, date lookup, and Wikipedia.
|
| 39 |
-
Follow this format:
|
| 40 |
-
Step 1: Think step-by-step.
|
| 41 |
-
Step 2: Use tool if needed.
|
| 42 |
-
Step 3: Final answer.
|
| 43 |
-
|
| 44 |
-
Question: {question}
|
| 45 |
-
Answer:
|
| 46 |
-
"""
|
| 47 |
-
messages = [
|
| 48 |
-
{"role": "system", "content": "You are a helpful assistant that uses tools and thinks step-by-step."},
|
| 49 |
-
{"role": "user", "content": prompt}
|
| 50 |
-
]
|
| 51 |
-
response = self.client.chat.completions.create(
|
| 52 |
-
model="gpt-4",
|
| 53 |
-
messages=messages,
|
| 54 |
-
temperature=0.3,
|
| 55 |
-
max_tokens=700
|
| 56 |
-
)
|
| 57 |
-
return response.choices[0].message.content.strip()
|
| 58 |
-
except Exception as e:
|
| 59 |
-
return f"[Agent Error: {e}]"
|
| 60 |
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
space_id = os.getenv("SPACE_ID")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 65 |
-
|
| 66 |
-
api_url = DEFAULT_API_URL
|
| 67 |
-
questions_url = f"{api_url}/questions"
|
| 68 |
-
submit_url = f"{api_url}/submit"
|
| 69 |
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
except Exception as e:
|
| 73 |
-
return f"Failed to initialize agent: {e}", pd.DataFrame()
|
| 74 |
|
|
|
|
| 75 |
try:
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
|
|
|
| 79 |
except Exception as e:
|
| 80 |
-
return f"
|
| 81 |
|
|
|
|
| 82 |
answers_payload = []
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
for q in questions_data:
|
| 86 |
task_id = q.get("task_id")
|
| 87 |
-
|
| 88 |
-
if not task_id or not
|
| 89 |
continue
|
| 90 |
try:
|
| 91 |
-
answer = agent(
|
| 92 |
except Exception as e:
|
| 93 |
-
answer = f"
|
| 94 |
-
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
| 95 |
-
log.append({"Task ID": task_id, "Question": question, "Submitted Answer": answer})
|
| 96 |
|
| 97 |
-
|
| 98 |
-
|
|
|
|
| 99 |
|
|
|
|
| 100 |
submission = {
|
| 101 |
-
"username":
|
| 102 |
"agent_code": agent_code,
|
| 103 |
-
"answers": answers_payload
|
| 104 |
}
|
| 105 |
|
| 106 |
try:
|
| 107 |
-
res = requests.post(submit_url, json=submission, timeout=
|
| 108 |
res.raise_for_status()
|
| 109 |
result = res.json()
|
| 110 |
-
|
| 111 |
-
f"
|
| 112 |
-
f"User: {result.get('username')}\n"
|
| 113 |
-
f"Score: {result.get('score')}%\n"
|
| 114 |
-
f"Correct: {result.get('correct_count')}/{result.get('total_attempted')}\n"
|
| 115 |
-
f"Message: {result.get('message')}"
|
| 116 |
)
|
| 117 |
-
return msg, pd.DataFrame(log)
|
| 118 |
except Exception as e:
|
| 119 |
-
|
| 120 |
|
| 121 |
-
|
| 122 |
-
gr.Markdown("# π GAIA Agent Evaluator")
|
| 123 |
-
gr.Markdown("Log in, then click the button to evaluate your agent.")
|
| 124 |
-
run_button = gr.Button("βΆοΈ Run Evaluation & Submit All Answers")
|
| 125 |
-
status = gr.Textbox(label="Status", lines=4)
|
| 126 |
-
table = gr.DataFrame(label="Agent Logs")
|
| 127 |
|
| 128 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
if __name__ == "__main__":
|
| 131 |
-
print("
|
| 132 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
|
|
|
|
|
|
| 5 |
|
| 6 |
+
# --- Constants ---
|
| 7 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 8 |
|
| 9 |
+
# --- Agent Definition (replace with your own logic) ---
|
| 10 |
+
class BasicAgent:
|
| 11 |
def __init__(self):
|
| 12 |
+
print("β
BasicAgent initialized.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
def __call__(self, question: str) -> str:
|
| 15 |
+
print(f"π₯ Question received: {question[:60]}...")
|
| 16 |
+
return "Paris" if "capital of France" in question else "42"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
# --- Evaluation Function ---
|
| 19 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 20 |
+
space_id = os.getenv("SPACE_ID") or "your-username/your-space" # fallback
|
| 21 |
+
|
| 22 |
+
if not profile:
|
| 23 |
+
return "β οΈ Please log in to Hugging Face to submit.", None
|
| 24 |
+
|
| 25 |
+
username = profile.username
|
| 26 |
+
print(f"π Logged in as: {username}")
|
| 27 |
+
|
| 28 |
+
questions_url = f"{DEFAULT_API_URL}/questions"
|
| 29 |
+
submit_url = f"{DEFAULT_API_URL}/submit"
|
| 30 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
+
# Instantiate your agent
|
| 33 |
+
agent = BasicAgent()
|
|
|
|
|
|
|
| 34 |
|
| 35 |
+
# Fetch questions
|
| 36 |
try:
|
| 37 |
+
response = requests.get(questions_url, timeout=15)
|
| 38 |
+
response.raise_for_status()
|
| 39 |
+
questions = response.json()
|
| 40 |
+
print(f"π¦ {len(questions)} questions fetched.")
|
| 41 |
except Exception as e:
|
| 42 |
+
return f"β Failed to fetch questions: {e}", None
|
| 43 |
|
| 44 |
+
# Process answers
|
| 45 |
answers_payload = []
|
| 46 |
+
logs = []
|
| 47 |
+
for q in questions:
|
|
|
|
| 48 |
task_id = q.get("task_id")
|
| 49 |
+
question_text = q.get("question")
|
| 50 |
+
if not task_id or not question_text:
|
| 51 |
continue
|
| 52 |
try:
|
| 53 |
+
answer = agent(question_text)
|
| 54 |
except Exception as e:
|
| 55 |
+
answer = f"AGENT ERROR: {e}"
|
|
|
|
|
|
|
| 56 |
|
| 57 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
| 58 |
+
logs.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": answer})
|
| 59 |
+
print(f"β
Task ID: {task_id} | Answer: {answer}")
|
| 60 |
|
| 61 |
+
# Submit answers
|
| 62 |
submission = {
|
| 63 |
+
"username": username,
|
| 64 |
"agent_code": agent_code,
|
| 65 |
+
"answers": answers_payload,
|
| 66 |
}
|
| 67 |
|
| 68 |
try:
|
| 69 |
+
res = requests.post(submit_url, json=submission, timeout=60)
|
| 70 |
res.raise_for_status()
|
| 71 |
result = res.json()
|
| 72 |
+
status = (
|
| 73 |
+
f"π Submission Successful!\n"
|
| 74 |
+
f"User: {result.get('username', 'N/A')}\n"
|
| 75 |
+
f"Score: {result.get('score', 0)}%\n"
|
| 76 |
+
f"Correct: {result.get('correct_count', 0)}/{result.get('total_attempted', 0)}\n"
|
| 77 |
+
f"Message: {result.get('message', '')}"
|
| 78 |
)
|
|
|
|
| 79 |
except Exception as e:
|
| 80 |
+
status = f"β Submission failed: {e}"
|
| 81 |
|
| 82 |
+
return status, pd.DataFrame(logs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
|
| 84 |
+
# --- Gradio UI ---
|
| 85 |
+
with gr.Blocks() as demo:
|
| 86 |
+
gr.Markdown("# π€ GAIA Evaluation Agent")
|
| 87 |
+
gr.Markdown("""
|
| 88 |
+
1. Log in with your Hugging Face account below.
|
| 89 |
+
2. Click the button to evaluate and submit answers.
|
| 90 |
+
3. Your score and submission details will appear below.
|
| 91 |
+
""")
|
| 92 |
+
|
| 93 |
+
profile_input = gr.OAuthProfile()
|
| 94 |
+
run_button = gr.Button("π Run Evaluation & Submit All Answers")
|
| 95 |
+
status_output = gr.Textbox(label="Submission Result", lines=6, interactive=False)
|
| 96 |
+
results_table = gr.DataFrame(label="Answer Log")
|
| 97 |
+
|
| 98 |
+
run_button.click(
|
| 99 |
+
fn=run_and_submit_all,
|
| 100 |
+
inputs=[profile_input],
|
| 101 |
+
outputs=[status_output, results_table]
|
| 102 |
+
)
|
| 103 |
|
| 104 |
if __name__ == "__main__":
|
| 105 |
+
print("\n================== GAIA Agent App Starting ==================")
|
| 106 |
+
if os.getenv("SPACE_ID"):
|
| 107 |
+
print(f"π¦ SPACE_ID = {os.getenv('SPACE_ID')}")
|
| 108 |
+
else:
|
| 109 |
+
print("βΉοΈ No SPACE_ID found. Using fallback link.")
|
| 110 |
+
|
| 111 |
+
demo.launch(debug=True)
|