Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,27 +2,19 @@
|
|
| 2 |
import os
|
| 3 |
import gradio as gr
|
| 4 |
import requests
|
| 5 |
-
|
| 6 |
-
from smolagents import Tool
|
| 7 |
-
|
| 8 |
-
from smolagents import Agent
|
| 9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
from audio_transcriber import AudioTranscriptionTool
|
| 12 |
from image_analyzer import ImageAnalysisTool
|
| 13 |
from wikipedia_searcher import WikipediaSearcher
|
| 14 |
|
| 15 |
|
| 16 |
-
#
|
| 17 |
-
HF_API_TOKEN = os.getenv("HF_API_TOKEN")
|
| 18 |
-
HF_CHAT_MODEL_URL = "https://api-inference.huggingface.com/models/HuggingFaceH4/zephyr-7b-beta"
|
| 19 |
-
|
| 20 |
-
HEADERS = {
|
| 21 |
-
"Authorization": f"Bearer {HF_API_TOKEN}",
|
| 22 |
-
"Content-Type": "application/json"
|
| 23 |
-
}
|
| 24 |
-
|
| 25 |
-
# Static system prompt
|
| 26 |
SYSTEM_PROMPT = """You are an agent solving the GAIA benchmark and you are required to provide exact answers.
|
| 27 |
Rules to follow:
|
| 28 |
1. Return only the exact requested answer: no explanation and no reasoning.
|
|
@@ -39,9 +31,11 @@ Examples of good responses:
|
|
| 39 |
Never include phrases like "the answer is..." or "Based on my research".
|
| 40 |
Only return the exact answer."""
|
| 41 |
|
| 42 |
-
|
|
|
|
| 43 |
audio_tool = AudioTranscriptionTool()
|
| 44 |
image_tool = ImageAnalysisTool()
|
|
|
|
| 45 |
wiki_tool = Tool.from_function(
|
| 46 |
name="wikipedia_search",
|
| 47 |
description="Search for facts using Wikipedia.",
|
|
@@ -52,68 +46,153 @@ wiki_tool = Tool.from_function(
|
|
| 52 |
|
| 53 |
tools = [audio_tool, image_tool, wiki_tool]
|
| 54 |
|
| 55 |
-
agent = Agent(
|
| 56 |
-
tools=tools,
|
| 57 |
-
system_prompt=SYSTEM_PROMPT
|
| 58 |
-
)
|
| 59 |
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
try:
|
| 62 |
-
response = requests.post(
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
"max_new_tokens": 256,
|
| 72 |
-
"return_full_text": False
|
| 73 |
-
}
|
| 74 |
-
}
|
| 75 |
)
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
except Exception as e:
|
| 81 |
-
return f"
|
| 82 |
-
|
| 83 |
-
def run_and_submit_all(question, file):
|
| 84 |
-
if file:
|
| 85 |
-
file_path = file.name
|
| 86 |
-
if file_path.endswith((".mp3", ".wav")):
|
| 87 |
-
transcript = audio_tool.forward(file_path)
|
| 88 |
-
question = f"{question}\n\nTranscription of audio: {transcript}"
|
| 89 |
-
elif file_path.endswith((".png", ".jpg", ".jpeg")):
|
| 90 |
-
image_answer = image_tool.forward(file_path, question)
|
| 91 |
-
return image_answer
|
| 92 |
-
elif file_path.endswith(".py"):
|
| 93 |
-
try:
|
| 94 |
-
with open(file_path, "r") as f:
|
| 95 |
-
code = f.read()
|
| 96 |
-
question = f"{question}\n\nPython code:\n{code}"
|
| 97 |
-
except Exception as e:
|
| 98 |
-
return f"Error reading code file: {e}"
|
| 99 |
-
else:
|
| 100 |
-
return "Unsupported file type."
|
| 101 |
-
|
| 102 |
-
full_prompt = f"{SYSTEM_PROMPT}\nQUESTION:\n{question}"
|
| 103 |
-
return query_hf_model(full_prompt)
|
| 104 |
-
|
| 105 |
-
with gr.Blocks(title="GAIA Agent with HF API") as demo:
|
| 106 |
-
gr.Markdown("### GAIA Evaluation Agent (Hugging Face-based)")
|
| 107 |
-
|
| 108 |
-
with gr.Row():
|
| 109 |
-
question_input = gr.Textbox(label="Question", placeholder="Enter your question here...", lines=3)
|
| 110 |
-
file_input = gr.File(label="Optional File (Audio, Image, or Python)", file_types=[".mp3", ".wav", ".jpg", ".jpeg", ".png", ".py"])
|
| 111 |
-
|
| 112 |
-
submit_button = gr.Button("Run Agent")
|
| 113 |
-
output_box = gr.Textbox(label="Answer")
|
| 114 |
-
|
| 115 |
-
submit_button.click(fn=run_and_submit_all, inputs=[question_input, file_input], outputs=output_box)
|
| 116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
if __name__ == "__main__":
|
| 118 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
|
|
|
|
| 2 |
import os
|
| 3 |
import gradio as gr
|
| 4 |
import requests
|
| 5 |
+
import pandas as pd
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
+
#from smolagents.agent import CodeAgent
|
| 8 |
+
#from smolagents.models import HfApiModel
|
| 9 |
+
from smolagents import Tool
|
| 10 |
+
from smolagents import CodeAgent, HfApiModel
|
| 11 |
|
| 12 |
from audio_transcriber import AudioTranscriptionTool
|
| 13 |
from image_analyzer import ImageAnalysisTool
|
| 14 |
from wikipedia_searcher import WikipediaSearcher
|
| 15 |
|
| 16 |
|
| 17 |
+
# System prompt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
SYSTEM_PROMPT = """You are an agent solving the GAIA benchmark and you are required to provide exact answers.
|
| 19 |
Rules to follow:
|
| 20 |
1. Return only the exact requested answer: no explanation and no reasoning.
|
|
|
|
| 31 |
Never include phrases like "the answer is..." or "Based on my research".
|
| 32 |
Only return the exact answer."""
|
| 33 |
|
| 34 |
+
|
| 35 |
+
# Tool definitions
|
| 36 |
audio_tool = AudioTranscriptionTool()
|
| 37 |
image_tool = ImageAnalysisTool()
|
| 38 |
+
|
| 39 |
wiki_tool = Tool.from_function(
|
| 40 |
name="wikipedia_search",
|
| 41 |
description="Search for facts using Wikipedia.",
|
|
|
|
| 46 |
|
| 47 |
tools = [audio_tool, image_tool, wiki_tool]
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
+
# Agent factory
|
| 51 |
+
def MyAgent():
|
| 52 |
+
return CodeAgent(
|
| 53 |
+
tools=tools,
|
| 54 |
+
system_prompt=SYSTEM_PROMPT,
|
| 55 |
+
model=HfApiModel(
|
| 56 |
+
api_url="https://api-inference.huggingface.com/models/HuggingFaceH4/zephyr-7b-beta",
|
| 57 |
+
api_key=os.getenv("HF_API_TOKEN")
|
| 58 |
+
)
|
| 59 |
+
)
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
# Main run and submission logic
|
| 63 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 64 |
+
space_id = os.getenv("SPACE_ID")
|
| 65 |
+
|
| 66 |
+
if profile:
|
| 67 |
+
username = profile.username
|
| 68 |
+
print(f"User logged in: {username}")
|
| 69 |
+
else:
|
| 70 |
+
print("User not logged in.")
|
| 71 |
+
return "Please Login to Hugging Face with the button.", None
|
| 72 |
+
|
| 73 |
+
api_url = os.getenv("GAIA_API_URL", "https://gaia-benchmark.com/api")
|
| 74 |
+
questions_url = f"{api_url}/questions"
|
| 75 |
+
submit_url = f"{api_url}/submit"
|
| 76 |
+
|
| 77 |
+
try:
|
| 78 |
+
agent = MyAgent()
|
| 79 |
+
except Exception as e:
|
| 80 |
+
print(f"Error initializing agent: {e}")
|
| 81 |
+
return f"Error initializing agent: {e}", None
|
| 82 |
+
|
| 83 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 84 |
+
print(f"Agent code URL: {agent_code}")
|
| 85 |
+
|
| 86 |
+
print(f"Fetching questions from: {questions_url}")
|
| 87 |
+
try:
|
| 88 |
+
response = requests.get(questions_url, timeout=15)
|
| 89 |
+
response.raise_for_status()
|
| 90 |
+
questions_data = response.json()
|
| 91 |
+
if not questions_data:
|
| 92 |
+
return "Fetched questions list is empty or invalid format.", None
|
| 93 |
+
print(f"Fetched {len(questions_data)} questions.")
|
| 94 |
+
except Exception as e:
|
| 95 |
+
return f"Error fetching questions: {e}", None
|
| 96 |
+
|
| 97 |
+
results_log = []
|
| 98 |
+
answers_payload = []
|
| 99 |
+
print(f"Running agent on {len(questions_data)} questions...")
|
| 100 |
+
|
| 101 |
+
for item in questions_data:
|
| 102 |
+
task_id = item.get("task_id")
|
| 103 |
+
if not task_id:
|
| 104 |
+
continue
|
| 105 |
+
try:
|
| 106 |
+
submitted_answer = agent(item)
|
| 107 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 108 |
+
results_log.append({
|
| 109 |
+
"Task ID": task_id,
|
| 110 |
+
"Question": item.get("question", ""),
|
| 111 |
+
"Submitted Answer": submitted_answer
|
| 112 |
+
})
|
| 113 |
+
except Exception as e:
|
| 114 |
+
error_msg = f"AGENT ERROR: {e}"
|
| 115 |
+
results_log.append({
|
| 116 |
+
"Task ID": task_id,
|
| 117 |
+
"Question": item.get("question", ""),
|
| 118 |
+
"Submitted Answer": error_msg
|
| 119 |
+
})
|
| 120 |
+
|
| 121 |
+
if not answers_payload:
|
| 122 |
+
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 123 |
+
|
| 124 |
+
submission_data = {
|
| 125 |
+
"username": username.strip(),
|
| 126 |
+
"agent_code": agent_code,
|
| 127 |
+
"answers": answers_payload
|
| 128 |
+
}
|
| 129 |
+
|
| 130 |
+
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 131 |
try:
|
| 132 |
+
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 133 |
+
response.raise_for_status()
|
| 134 |
+
result_data = response.json()
|
| 135 |
+
final_status = (
|
| 136 |
+
f"Submission Successful!\n"
|
| 137 |
+
f"User: {result_data.get('username')}\n"
|
| 138 |
+
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
| 139 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 140 |
+
f"Message: {result_data.get('message', 'No message received.')}"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
)
|
| 142 |
+
results_df = pd.DataFrame(results_log)
|
| 143 |
+
return final_status, results_df
|
| 144 |
+
except requests.exceptions.HTTPError as e:
|
| 145 |
+
try:
|
| 146 |
+
detail = e.response.json().get("detail", e.response.text)
|
| 147 |
+
except Exception:
|
| 148 |
+
detail = e.response.text[:500]
|
| 149 |
+
return f"Submission Failed: {detail}", pd.DataFrame(results_log)
|
| 150 |
+
except requests.exceptions.Timeout:
|
| 151 |
+
return "Submission Failed: The request timed out.", pd.DataFrame(results_log)
|
| 152 |
except Exception as e:
|
| 153 |
+
return f"An unexpected error occurred during submission: {e}", pd.DataFrame(results_log)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 154 |
|
| 155 |
+
|
| 156 |
+
# Gradio UI setup
|
| 157 |
+
with gr.Blocks() as demo:
|
| 158 |
+
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 159 |
+
gr.Markdown("""
|
| 160 |
+
**Instructions:**
|
| 161 |
+
1. Clone this space, modify code to define your agent's logic, tools, and packages.
|
| 162 |
+
2. Log in to your Hugging Face account using the button below.
|
| 163 |
+
3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see your score.
|
| 164 |
+
**Note:** Submitting can take some time.
|
| 165 |
+
""")
|
| 166 |
+
|
| 167 |
+
gr.LoginButton()
|
| 168 |
+
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 169 |
+
|
| 170 |
+
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 171 |
+
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 172 |
+
|
| 173 |
+
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
| 174 |
+
|
| 175 |
+
# App startup logs
|
| 176 |
if __name__ == "__main__":
|
| 177 |
+
print("\n" + "-" * 30 + " App Starting " + "-" * 30)
|
| 178 |
+
space_host = os.getenv("SPACE_HOST")
|
| 179 |
+
space_id = os.getenv("SPACE_ID")
|
| 180 |
+
|
| 181 |
+
if space_host:
|
| 182 |
+
print(f"✅ SPACE_HOST found: {space_host}")
|
| 183 |
+
print(f" Runtime URL should be: https://{space_host}.hf.space")
|
| 184 |
+
else:
|
| 185 |
+
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
| 186 |
+
|
| 187 |
+
if space_id:
|
| 188 |
+
print(f"✅ SPACE_ID found: {space_id}")
|
| 189 |
+
print(f" Repo URL: https://huggingface.co/spaces/{space_id}")
|
| 190 |
+
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id}/tree/main")
|
| 191 |
+
else:
|
| 192 |
+
print("ℹ️ SPACE_ID environment variable not found (running locally?).")
|
| 193 |
+
|
| 194 |
+
print("-" * (60 + len(" App Starting ")) + "\n")
|
| 195 |
+
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
| 196 |
+
demo.launch(debug=True, share=False)
|
| 197 |
+
|
| 198 |
|