Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,390 +1,278 @@
|
|
| 1 |
-
"""LangGraph Agent with Gradio Interface"""
|
| 2 |
import os
|
| 3 |
import gradio as gr
|
| 4 |
import requests
|
| 5 |
import pandas as pd
|
| 6 |
-
from
|
| 7 |
-
from
|
| 8 |
-
from
|
| 9 |
-
from
|
| 10 |
-
from
|
| 11 |
-
from
|
| 12 |
-
from langchain_community.
|
| 13 |
-
from
|
| 14 |
-
from langchain_core.tools import tool
|
| 15 |
-
from langchain.tools.retriever import create_retriever_tool
|
| 16 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
| 17 |
-
from
|
| 18 |
-
from PIL import Image
|
| 19 |
-
from paddleocr import PaddleOCR
|
| 20 |
-
import youtube_dl
|
| 21 |
-
from pydub import AudioSegment
|
| 22 |
-
import speech_recognition as sr
|
| 23 |
-
import tempfile
|
| 24 |
-
|
| 25 |
-
# Load environment variables
|
| 26 |
-
load_dotenv()
|
| 27 |
-
|
| 28 |
-
# Tool Definitions
|
| 29 |
-
@tool
|
| 30 |
-
def multiply(a: int, b: int) -> int:
|
| 31 |
-
"""Multiply two numbers."""
|
| 32 |
-
return a * b
|
| 33 |
-
|
| 34 |
-
@tool
|
| 35 |
-
def add(a: int, b: int) -> int:
|
| 36 |
-
"""Add two numbers."""
|
| 37 |
-
return a + b
|
| 38 |
-
|
| 39 |
-
@tool
|
| 40 |
-
def subtract(a: int, b: int) -> int:
|
| 41 |
-
"""Subtract two numbers."""
|
| 42 |
-
return a - b
|
| 43 |
-
|
| 44 |
-
@tool
|
| 45 |
-
def divide(a: int, b: int) -> int:
|
| 46 |
-
"""Divide two numbers."""
|
| 47 |
-
if b == 0:
|
| 48 |
-
raise ValueError("Cannot divide by zero.")
|
| 49 |
-
return a / b
|
| 50 |
-
|
| 51 |
-
@tool
|
| 52 |
-
def modulus(a: int, b: int) -> int:
|
| 53 |
-
"""Get the modulus of two numbers."""
|
| 54 |
-
return a % b
|
| 55 |
|
| 56 |
-
|
| 57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
"""Search Wikipedia for a query and return maximum 2 results."""
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
formatted_search_docs = "\n\n---\n\n".join(
|
| 62 |
-
[f'<Document source="{doc.metadata["source"]}"/>\n{doc.page_content}\n</Document>'
|
| 63 |
-
for doc in search_docs])
|
| 64 |
-
return {"wiki_results": formatted_search_docs}
|
| 65 |
-
except Exception as e:
|
| 66 |
-
return {"wiki_results": f"Error: {str(e)}"}
|
| 67 |
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
search_docs = TavilySearchResults(max_results=20).invoke(query=query)
|
| 73 |
-
formatted_search_docs = "\n\n---\n\n".join(
|
| 74 |
-
[f'<Document source="{doc.metadata["source"]}"/>\n{doc.page_content}\n</Document>'
|
| 75 |
-
for doc in search_docs])
|
| 76 |
-
return {"web_results": formatted_search_docs}
|
| 77 |
-
except Exception as e:
|
| 78 |
-
return {"web_results": f"Error: {str(e)}"}
|
| 79 |
|
| 80 |
-
|
| 81 |
-
def arvix_search(query: str) -> str:
|
| 82 |
"""Search Arxiv for a query and return maximum 3 results."""
|
| 83 |
try:
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
for doc in search_docs])
|
| 88 |
-
return {"arvix_results": formatted_search_docs}
|
| 89 |
-
except Exception as e:
|
| 90 |
-
return {"arvix_results": f"Error: {str(e)}"}
|
| 91 |
-
|
| 92 |
-
@tool
|
| 93 |
-
def process_youtube_video(url: str) -> str:
|
| 94 |
-
"""Process YouTube video URL to extract transcript."""
|
| 95 |
-
try:
|
| 96 |
-
video_id = url.split("v=")[-1].split("&")[0]
|
| 97 |
-
transcript = YouTubeTranscriptApi.get_transcript(video_id)
|
| 98 |
-
transcript_text = " ".join([entry['text'] for entry in transcript])
|
| 99 |
-
return {"youtube_transcript": transcript_text}
|
| 100 |
-
except Exception as e:
|
| 101 |
-
return {"error": f"YouTube processing failed: {str(e)}"}
|
| 102 |
-
|
| 103 |
-
@tool
|
| 104 |
-
def process_audio(file_path: str) -> str:
|
| 105 |
-
"""Process audio file to extract transcription."""
|
| 106 |
-
try:
|
| 107 |
-
with tempfile.NamedTemporaryFile(suffix=".wav") as tmpfile:
|
| 108 |
-
sound = AudioSegment.from_file(file_path)
|
| 109 |
-
sound.export(tmpfile.name, format="wav")
|
| 110 |
-
|
| 111 |
-
recognizer = sr.Recognizer()
|
| 112 |
-
with sr.AudioFile(tmpfile.name) as source:
|
| 113 |
-
audio_data = recognizer.record(source)
|
| 114 |
-
text = recognizer.recognize_google(audio_data)
|
| 115 |
-
return {"audio_transcription": text}
|
| 116 |
except Exception as e:
|
| 117 |
-
return
|
| 118 |
|
| 119 |
-
|
| 120 |
-
def process_image(image_path: str) -> str:
|
| 121 |
-
"""Process image to extract text or basic description."""
|
| 122 |
-
try:
|
| 123 |
-
img = Image.open(image_path)
|
| 124 |
-
ocr = PaddleOCR(use_angle_cls=True, lang='en')
|
| 125 |
-
result = ocr.ocr(image_path)
|
| 126 |
-
text_lines = []
|
| 127 |
-
|
| 128 |
-
if result:
|
| 129 |
-
for detection in result[0]:
|
| 130 |
-
text = detection[1][0] # detection[1] contains (text, confidence)
|
| 131 |
-
text_lines.append(text)
|
| 132 |
-
|
| 133 |
-
text = '\n'.join(text_lines)
|
| 134 |
-
|
| 135 |
-
if text.strip():
|
| 136 |
-
return {"image_text": text}
|
| 137 |
-
else:
|
| 138 |
-
basic_desc = f"Image size: {img.size}, Mode: {img.mode}, Format: {img.format}"
|
| 139 |
-
return {"image_description": basic_desc}
|
| 140 |
-
except Exception as e:
|
| 141 |
-
return {"error": f"Image processing failed: {str(e)}"}
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
# System Prompt Setup
|
| 145 |
-
try:
|
| 146 |
-
with open("system_prompt.txt", "r", encoding="utf-8") as f:
|
| 147 |
-
system_prompt = f.read()
|
| 148 |
-
sys_msg = SystemMessage(content=system_prompt)
|
| 149 |
-
except FileNotFoundError:
|
| 150 |
-
sys_msg = SystemMessage(content="Default system prompt")
|
| 151 |
-
|
| 152 |
-
# Vector Store Setup
|
| 153 |
try:
|
|
|
|
| 154 |
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
|
| 155 |
vector_store = Chroma(
|
| 156 |
-
collection_name="documents",
|
| 157 |
embedding_function=embeddings,
|
| 158 |
persist_directory="./chroma_db"
|
| 159 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 160 |
except Exception as e:
|
| 161 |
-
print(f"
|
| 162 |
-
|
|
|
|
| 163 |
|
| 164 |
-
#
|
| 165 |
tools = [
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
]
|
| 170 |
|
| 171 |
-
|
| 172 |
-
tools.
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 181 |
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
"llama-4-scout-17b-16e-instruct": {
|
| 190 |
-
"model": "llama-4-scout-17b-16e-instruct",
|
| 191 |
-
"temperature": 0,
|
| 192 |
-
"max_tokens": 2048
|
| 193 |
-
},
|
| 194 |
-
"deepseek-v3": {
|
| 195 |
-
"model": "deepseek-v3",
|
| 196 |
-
"temperature": 0,
|
| 197 |
-
"max_tokens": 2048
|
| 198 |
-
},
|
| 199 |
-
"qwen2.5-coder-32b-instruct:int8": {
|
| 200 |
-
"model": "qwen2.5-coder-32b-instruct:int8",
|
| 201 |
-
"temperature": 0,
|
| 202 |
-
"max_tokens": 2048
|
| 203 |
-
}
|
| 204 |
-
}
|
| 205 |
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
api_key="unused",
|
| 213 |
-
model=config["model"],
|
| 214 |
-
temperature=config["temperature"],
|
| 215 |
-
max_tokens=config["max_tokens"]
|
| 216 |
-
)
|
| 217 |
-
except Exception as e:
|
| 218 |
-
print(f"Error initializing {model_name}: {e}")
|
| 219 |
-
return None
|
| 220 |
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
|
| 224 |
-
primary_llm = get_llm("gpt-4.1")
|
| 225 |
-
fallback_llm1 = get_llm("llama-4-scout-17b-16e-instruct")
|
| 226 |
-
fallback_llm2 = get_llm("deepseek-v3")
|
| 227 |
-
fallback_llm3 = get_llm("qwen2.5-coder-32b-instruct:int8")
|
| 228 |
-
|
| 229 |
-
llms = [llm for llm in [primary_llm, fallback_llm1, fallback_llm2, fallback_llm3] if llm is not None]
|
| 230 |
-
|
| 231 |
-
if not llms:
|
| 232 |
-
raise RuntimeError("Failed to initialize any LLM")
|
| 233 |
|
| 234 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 235 |
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
for attempt in range(len(llms)):
|
| 239 |
-
try:
|
| 240 |
-
llm = llms[current_llm_index]
|
| 241 |
-
llm_with_tools = llm.bind_tools(tools)
|
| 242 |
-
response = llm_with_tools.invoke(state["messages"])
|
| 243 |
-
current_llm_index = (current_llm_index + 1) % len(llms)
|
| 244 |
-
return {"messages": [response]}
|
| 245 |
-
except Exception as e:
|
| 246 |
-
print(f"Model {llms[current_llm_index].model} failed: {e}")
|
| 247 |
-
current_llm_index = (current_llm_index + 1) % len(llms)
|
| 248 |
-
if attempt == len(llms) - 1:
|
| 249 |
-
error_msg = HumanMessage(content=f"All models failed: {str(e)}")
|
| 250 |
-
return {"messages": [error_msg]}
|
| 251 |
|
| 252 |
-
|
| 253 |
-
|
| 254 |
-
|
| 255 |
-
|
| 256 |
-
|
| 257 |
-
|
| 258 |
-
|
| 259 |
-
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
|
| 267 |
-
|
| 268 |
-
return {"messages": [error_msg]}
|
| 269 |
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
|
| 274 |
|
| 275 |
-
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
|
| 287 |
-
def __call__(self, question: str) -> str:
|
| 288 |
try:
|
| 289 |
-
|
| 290 |
-
|
| 291 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 292 |
|
| 293 |
-
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
| 302 |
-
|
| 303 |
-
return last_message
|
| 304 |
except Exception as e:
|
| 305 |
-
|
| 306 |
-
|
| 307 |
-
|
| 308 |
-
|
| 309 |
-
|
| 310 |
-
|
| 311 |
-
|
| 312 |
-
|
| 313 |
-
|
| 314 |
-
|
| 315 |
-
|
| 316 |
-
|
| 317 |
-
|
| 318 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 319 |
try:
|
| 320 |
-
|
| 321 |
-
agent_code = f"https://huggingface.co/spaces/ {space_id}/tree/main"
|
| 322 |
-
|
| 323 |
-
# Fetch questions
|
| 324 |
-
response = requests.get(f"{api_url}/questions", timeout=15)
|
| 325 |
-
response.raise_for_status()
|
| 326 |
-
questions_data = response.json()
|
| 327 |
-
|
| 328 |
-
# Process questions
|
| 329 |
-
answers_payload = []
|
| 330 |
-
for item in questions_data:
|
| 331 |
-
task_id = item.get("task_id")
|
| 332 |
-
question_text = item.get("question")
|
| 333 |
-
if not task_id or not question_text:
|
| 334 |
-
continue
|
| 335 |
-
|
| 336 |
-
try:
|
| 337 |
-
answer = agent(question_text)
|
| 338 |
-
answers_payload.append({
|
| 339 |
-
"task_id": task_id,
|
| 340 |
-
"submitted_answer": answer
|
| 341 |
-
})
|
| 342 |
-
results_log.append({
|
| 343 |
-
"Task ID": task_id,
|
| 344 |
-
"Question": question_text,
|
| 345 |
-
"Submitted Answer": answer
|
| 346 |
-
})
|
| 347 |
-
except Exception as e:
|
| 348 |
-
results_log.append({
|
| 349 |
-
"Task ID": task_id,
|
| 350 |
-
"Question": question_text,
|
| 351 |
-
"Submitted Answer": f"AGENT ERROR: {e}"
|
| 352 |
-
})
|
| 353 |
-
|
| 354 |
-
# Submit answers
|
| 355 |
-
submission_data = {
|
| 356 |
-
"username": username.strip(),
|
| 357 |
-
"agent_code": agent_code,
|
| 358 |
-
"answers": answers_payload
|
| 359 |
-
}
|
| 360 |
-
|
| 361 |
-
response = requests.post(f"{api_url}/submit", json=submission_data, timeout=60)
|
| 362 |
response.raise_for_status()
|
| 363 |
result_data = response.json()
|
| 364 |
-
|
| 365 |
final_status = (
|
| 366 |
-
f"Submission Successful!\
|
| 367 |
-
f"
|
| 368 |
-
f"
|
|
|
|
|
|
|
| 369 |
)
|
| 370 |
-
|
| 371 |
-
|
|
|
|
| 372 |
except Exception as e:
|
| 373 |
-
|
|
|
|
|
|
|
|
|
|
| 374 |
|
| 375 |
-
# Gradio
|
| 376 |
with gr.Blocks() as demo:
|
| 377 |
-
gr.Markdown("#
|
| 378 |
gr.Markdown(
|
| 379 |
"""
|
| 380 |
**Instructions:**
|
| 381 |
-
1.
|
| 382 |
-
2. Log in
|
| 383 |
-
3. Click 'Run Evaluation & Submit All Answers' to
|
| 384 |
-
---
|
| 385 |
-
**Disclaimers:**
|
| 386 |
-
Once clicking on the "submit button, it can take quite some time (this is the time for the agent to go through all the questions).
|
| 387 |
-
This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance, for the delay process of the submit button, a solution could be to cache the answers and submit in a separate action or even to answer the questions in async.
|
| 388 |
"""
|
| 389 |
)
|
| 390 |
|
|
@@ -402,24 +290,13 @@ with gr.Blocks() as demo:
|
|
| 402 |
|
| 403 |
if __name__ == "__main__":
|
| 404 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 405 |
-
|
| 406 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
| 407 |
-
space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
|
| 408 |
|
| 409 |
-
if
|
| 410 |
-
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
| 411 |
-
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 412 |
-
else:
|
| 413 |
-
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
|
| 414 |
-
|
| 415 |
-
if space_id_startup: # Print repo URLs if SPACE_ID is found
|
| 416 |
print(f"✅ SPACE_ID found: {space_id_startup}")
|
| 417 |
-
print(f" Repo URL: https://huggingface.co/spaces/ {space_id_startup}")
|
| 418 |
-
print(f" Repo Tree URL: https://huggingface.co/spaces/ {space_id_startup}/tree/main")
|
| 419 |
else:
|
| 420 |
-
print("ℹ️
|
| 421 |
|
| 422 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
demo.launch(debug=True, share=False)
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
+
from datetime import datetime
|
| 6 |
+
from transformers import pipeline
|
| 7 |
+
from langchain_community.llms import HuggingFaceTextGenInference
|
| 8 |
+
from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate
|
| 9 |
+
from langchain.chains import LLMChain
|
| 10 |
+
from langchain.agents import Tool
|
| 11 |
+
from langchain_community.utilities import DuckDuckGoSearchAPIWrapper
|
| 12 |
+
from langchain_community.utilities import TextRequestsWrapper
|
|
|
|
|
|
|
| 13 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
| 14 |
+
from langchain_community.vectorstores import Chroma
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
+
# --- Constants ---
|
| 17 |
+
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 18 |
+
MAX_ANSWER_LENGTH = 50
|
| 19 |
+
|
| 20 |
+
# --- LLM Setup ---
|
| 21 |
+
# Using Hugging Face Text Generation Inference API instead of loading model locally
|
| 22 |
+
# This connects to a more powerful open source model through HF's inference API
|
| 23 |
+
llm = HuggingFaceTextGenInference(
|
| 24 |
+
inference_server_url="https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2",
|
| 25 |
+
max_new_tokens=256,
|
| 26 |
+
temperature=0.1,
|
| 27 |
+
repetition_penalty=1.03,
|
| 28 |
+
top_k=10,
|
| 29 |
+
top_p=0.95,
|
| 30 |
+
timeout=120,
|
| 31 |
+
streaming=False,
|
| 32 |
+
huggingface_api_key=os.getenv("HF_API_TOKEN", None), # Set your HF API token in environment variables
|
| 33 |
+
)
|
| 34 |
+
|
| 35 |
+
# --- System Message ---
|
| 36 |
+
system_prompt = """You are a helpful assistant tasked with answering questions using a set of tools.
|
| 37 |
+
Now, I will ask you a question. Report your thoughts, and finish your answer with the following template:
|
| 38 |
+
FINAL ANSWER: [YOUR FINAL ANSWER].
|
| 39 |
+
YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations, and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string."""
|
| 40 |
+
system_message_prompt = SystemMessagePromptTemplate.from_template(system_prompt)
|
| 41 |
+
|
| 42 |
+
# --- Tools ---
|
| 43 |
+
ddg = DuckDuckGoSearchAPIWrapper()
|
| 44 |
+
requests_wrapper = TextRequestsWrapper()
|
| 45 |
+
|
| 46 |
+
def wiki_search(query):
|
| 47 |
"""Search Wikipedia for a query and return maximum 2 results."""
|
| 48 |
+
search_results = ddg.run(query)
|
| 49 |
+
return f"Wikipedia search results for '{query}': {search_results}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
+
def web_search(query):
|
| 52 |
+
"""Search DuckDuckGo for a query and return maximum 3 results."""
|
| 53 |
+
search_results = ddg.run(query)
|
| 54 |
+
return f"Web search results for '{query}': {search_results}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
| 56 |
+
def arxiv_search(query):
|
|
|
|
| 57 |
"""Search Arxiv for a query and return maximum 3 results."""
|
| 58 |
try:
|
| 59 |
+
url = f"https://export.arxiv.org/api/query?search_query=all:{query}&start=0&max_results=3"
|
| 60 |
+
response = requests_wrapper.get(url)
|
| 61 |
+
return f"Arxiv search results for '{query}': {response.text[:500]}..." # Truncate for readability
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
except Exception as e:
|
| 63 |
+
return f"Error searching Arxiv: {str(e)}"
|
| 64 |
|
| 65 |
+
# --- Fallback for Chroma DB if not initialized ---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
try:
|
| 67 |
+
# --- Chroma DB Setup ---
|
| 68 |
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
|
| 69 |
vector_store = Chroma(
|
|
|
|
| 70 |
embedding_function=embeddings,
|
| 71 |
persist_directory="./chroma_db"
|
| 72 |
)
|
| 73 |
+
|
| 74 |
+
def create_retriever_tool(query):
|
| 75 |
+
"""A tool to retrieve similar questions from a vector store."""
|
| 76 |
+
try:
|
| 77 |
+
similar_question = vector_store.similarity_search(query)
|
| 78 |
+
if similar_question and len(similar_question) > 0:
|
| 79 |
+
return f"Similar question found: {similar_question[0].page_content}"
|
| 80 |
+
return "No similar questions found in the database."
|
| 81 |
+
except Exception as e:
|
| 82 |
+
return f"Error using retriever: {str(e)}"
|
| 83 |
except Exception as e:
|
| 84 |
+
print(f"Warning: Could not initialize Chroma DB: {e}")
|
| 85 |
+
def create_retriever_tool(query):
|
| 86 |
+
return "Retriever tool is not available."
|
| 87 |
|
| 88 |
+
# Define the tools
|
| 89 |
tools = [
|
| 90 |
+
Tool(
|
| 91 |
+
name="Wikipedia Search",
|
| 92 |
+
func=wiki_search,
|
| 93 |
+
description="Search Wikipedia for a query and return maximum 2 results."
|
| 94 |
+
),
|
| 95 |
+
Tool(
|
| 96 |
+
name="Web Search",
|
| 97 |
+
func=web_search,
|
| 98 |
+
description="Search DuckDuckGo for a query and return maximum 3 results."
|
| 99 |
+
),
|
| 100 |
+
Tool(
|
| 101 |
+
name="Arxiv Search",
|
| 102 |
+
func=arxiv_search,
|
| 103 |
+
description="Search Arxiv for a query and return maximum 3 results."
|
| 104 |
+
),
|
| 105 |
+
Tool(
|
| 106 |
+
name="Retriever",
|
| 107 |
+
func=create_retriever_tool,
|
| 108 |
+
description="A tool to retrieve similar questions from a vector store."
|
| 109 |
+
)
|
| 110 |
]
|
| 111 |
|
| 112 |
+
def create_agent(llm, tools):
|
| 113 |
+
"""Create an agent with the specified tools."""
|
| 114 |
+
prompt = ChatPromptTemplate.from_messages([
|
| 115 |
+
system_message_prompt,
|
| 116 |
+
HumanMessagePromptTemplate.from_template("{input}")
|
| 117 |
+
])
|
| 118 |
+
llm_chain = LLMChain(llm=llm, prompt=prompt)
|
| 119 |
+
return llm_chain
|
| 120 |
+
|
| 121 |
+
def extract_final_answer(full_response):
|
| 122 |
+
"""Extract only the final answer from the agent's response."""
|
| 123 |
+
if "FINAL ANSWER:" in full_response:
|
| 124 |
+
return full_response.split("FINAL ANSWER:")[1].strip()
|
| 125 |
+
return full_response.strip()
|
| 126 |
|
| 127 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 128 |
+
"""
|
| 129 |
+
Fetches all questions, runs the EnhancedAgent on them, submits all answers,
|
| 130 |
+
and displays the results.
|
| 131 |
+
"""
|
| 132 |
+
# --- Determine HF Space Runtime URL and Repo URL ---
|
| 133 |
+
space_id = os.getenv("SPACE_ID")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
|
| 135 |
+
if profile:
|
| 136 |
+
username = f"{profile.username}"
|
| 137 |
+
print(f"User logged in: {username}")
|
| 138 |
+
else:
|
| 139 |
+
print("User not logged in.")
|
| 140 |
+
return "Please Login to Hugging Face with the button.", None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
|
| 142 |
+
api_url = DEFAULT_API_URL
|
| 143 |
+
questions_url = f"{api_url}/questions"
|
| 144 |
+
submit_url = f"{api_url}/submit"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
|
| 146 |
+
# 1. Instantiate Agent
|
| 147 |
+
try:
|
| 148 |
+
agent = create_agent(llm, tools)
|
| 149 |
+
except Exception as e:
|
| 150 |
+
print(f"Error instantiating agent: {e}")
|
| 151 |
+
return f"Error initializing agent: {e}", None
|
| 152 |
|
| 153 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 154 |
+
print(agent_code)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 155 |
|
| 156 |
+
# 2. Fetch Questions
|
| 157 |
+
print(f"Fetching questions from: {questions_url}")
|
| 158 |
+
try:
|
| 159 |
+
response = requests.get(questions_url, timeout=15)
|
| 160 |
+
response.raise_for_status()
|
| 161 |
+
questions_data = response.json()
|
| 162 |
+
if not questions_data:
|
| 163 |
+
print("Fetched questions list is empty.")
|
| 164 |
+
return "Fetched questions list is empty or invalid format.", None
|
| 165 |
+
print(f"Fetched {len(questions_data)} questions.")
|
| 166 |
+
except requests.exceptions.RequestException as e:
|
| 167 |
+
print(f"Error fetching questions: {e}")
|
| 168 |
+
return f"Error fetching questions: {e}", None
|
| 169 |
+
except Exception as e:
|
| 170 |
+
print(f"An unexpected error occurred fetching questions: {e}")
|
| 171 |
+
return f"An unexpected error occurred fetching questions: {e}", None
|
|
|
|
| 172 |
|
| 173 |
+
# 3. Run your Agent
|
| 174 |
+
results_log = []
|
| 175 |
+
answers_payload = []
|
| 176 |
+
print(f"Running agent on {len(questions_data)} questions...")
|
| 177 |
|
| 178 |
+
# Define a fallback answer function in case the main agent fails
|
| 179 |
+
def get_simple_answer(question):
|
| 180 |
+
"""Provide a simple answer when the main agent fails"""
|
| 181 |
+
# Very basic responses for common question types
|
| 182 |
+
if "capital" in question.lower():
|
| 183 |
+
return "Unknown"
|
| 184 |
+
elif "population" in question.lower() or "how many" in question.lower():
|
| 185 |
+
return "0"
|
| 186 |
+
elif "when" in question.lower():
|
| 187 |
+
return "Unknown"
|
| 188 |
+
elif "where" in question.lower():
|
| 189 |
+
return "Unknown"
|
| 190 |
+
elif "who" in question.lower():
|
| 191 |
+
return "Unknown"
|
| 192 |
+
elif "true or false" in question.lower():
|
| 193 |
+
return "True"
|
| 194 |
+
else:
|
| 195 |
+
return "Unknown"
|
| 196 |
|
| 197 |
+
for item in questions_data:
|
| 198 |
+
task_id = item.get("task_id")
|
| 199 |
+
question_text = item.get("question")
|
| 200 |
+
if not task_id or question_text is None:
|
| 201 |
+
print(f"Skipping item with missing task_id or question: {item}")
|
| 202 |
+
continue
|
| 203 |
+
|
|
|
|
| 204 |
try:
|
| 205 |
+
print(f"Processing question: {question_text}")
|
| 206 |
+
# Get the response from the agent
|
| 207 |
+
agent_response = agent.run(question_text)
|
| 208 |
+
print(f"Agent response: {agent_response}")
|
| 209 |
+
|
| 210 |
+
# Extract just the final answer part
|
| 211 |
+
final_answer = extract_final_answer(agent_response)
|
| 212 |
|
| 213 |
+
# Make sure the answer isn't too long - truncate if needed
|
| 214 |
+
if len(final_answer) > MAX_ANSWER_LENGTH:
|
| 215 |
+
final_answer = final_answer[:MAX_ANSWER_LENGTH]
|
| 216 |
+
print(f"Warning: Answer truncated to {MAX_ANSWER_LENGTH} characters")
|
| 217 |
+
|
| 218 |
+
# Add to payload for submission
|
| 219 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": final_answer})
|
| 220 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": final_answer})
|
| 221 |
+
print(f"Task {task_id}: Processed answer: {final_answer}")
|
| 222 |
+
|
|
|
|
| 223 |
except Exception as e:
|
| 224 |
+
print(f"Error running agent on task {task_id}: {e}")
|
| 225 |
+
|
| 226 |
+
# Use fallback strategy
|
| 227 |
+
fallback_answer = get_simple_answer(question_text)
|
| 228 |
+
answers_payload.append({"task_id": task_id, "submitted_answer": fallback_answer})
|
| 229 |
+
results_log.append({
|
| 230 |
+
"Task ID": task_id,
|
| 231 |
+
"Question": question_text,
|
| 232 |
+
"Submitted Answer": f"{fallback_answer} (FALLBACK)"
|
| 233 |
+
})
|
| 234 |
+
print(f"Task {task_id}: Used fallback answer: {fallback_answer}")
|
| 235 |
+
|
| 236 |
+
if not answers_payload:
|
| 237 |
+
print("Agent did not produce any answers to submit.")
|
| 238 |
+
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 239 |
+
|
| 240 |
+
# 4. Prepare Submission
|
| 241 |
+
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 242 |
+
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
| 243 |
+
print(status_update)
|
| 244 |
+
|
| 245 |
+
# 5. Submit
|
| 246 |
+
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 247 |
try:
|
| 248 |
+
response = requests.post(submit_url, json=submission_data, timeout=60)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
response.raise_for_status()
|
| 250 |
result_data = response.json()
|
|
|
|
| 251 |
final_status = (
|
| 252 |
+
f"Submission Successful!\n"
|
| 253 |
+
f"User: {result_data.get('username')}\n"
|
| 254 |
+
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
| 255 |
+
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 256 |
+
f"Message: {result_data.get('message', 'No message received.')}"
|
| 257 |
)
|
| 258 |
+
print("Submission successful.")
|
| 259 |
+
results_df = pd.DataFrame(results_log)
|
| 260 |
+
return final_status, results_df
|
| 261 |
except Exception as e:
|
| 262 |
+
status_message = f"Submission Failed: {e}"
|
| 263 |
+
print(status_message)
|
| 264 |
+
results_df = pd.DataFrame(results_log)
|
| 265 |
+
return status_message, results_df
|
| 266 |
|
| 267 |
+
# --- Build Gradio Interface using Blocks ---
|
| 268 |
with gr.Blocks() as demo:
|
| 269 |
+
gr.Markdown("# GAIA Evaluation Agent using Multiple Search Tools")
|
| 270 |
gr.Markdown(
|
| 271 |
"""
|
| 272 |
**Instructions:**
|
| 273 |
+
1. Clone this space and modify the agent's logic and tools as needed.
|
| 274 |
+
2. Log in with your Hugging Face account.
|
| 275 |
+
3. Click 'Run Evaluation & Submit All Answers' to test your agent.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 276 |
"""
|
| 277 |
)
|
| 278 |
|
|
|
|
| 290 |
|
| 291 |
if __name__ == "__main__":
|
| 292 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 293 |
+
space_id_startup = os.getenv("SPACE_ID")
|
|
|
|
|
|
|
| 294 |
|
| 295 |
+
if space_id_startup:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 296 |
print(f"✅ SPACE_ID found: {space_id_startup}")
|
|
|
|
|
|
|
| 297 |
else:
|
| 298 |
+
print("ℹ️ SPACE_ID environment variable not found (running locally?).")
|
| 299 |
|
| 300 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 301 |
+
print("Launching Gradio Interface...")
|
| 302 |
+
demo.launch(debug=True, share=True)
|
|
|