Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
import os
|
| 2 |
-
import
|
|
|
|
| 3 |
import requests
|
| 4 |
-
import inspect
|
| 5 |
import pandas as pd
|
| 6 |
-
import re
|
| 7 |
from pathlib import Path
|
| 8 |
from typing import Optional
|
| 9 |
from dotenv import load_dotenv
|
|
|
|
|
|
|
| 10 |
|
| 11 |
from langgraph.prebuilt import create_react_agent
|
| 12 |
from langchain_core.messages import HumanMessage
|
|
@@ -15,13 +16,13 @@ from langchain_openai import ChatOpenAI
|
|
| 15 |
|
| 16 |
load_dotenv()
|
| 17 |
|
| 18 |
-
# ------------------
|
| 19 |
-
|
| 20 |
class OpenRouterLLM(ChatOpenAI):
|
| 21 |
"""Custom OpenRouter LLM wrapper for LangGraph"""
|
| 22 |
-
|
| 23 |
def __init__(self, model: str = "deepseek/deepseek-v3.1-terminus", **kwargs):
|
| 24 |
-
api_key = os.getenv("OPENROUTER_API_KEY")
|
|
|
|
|
|
|
| 25 |
super().__init__(
|
| 26 |
model=model,
|
| 27 |
openai_api_key=api_key,
|
|
@@ -31,43 +32,54 @@ class OpenRouterLLM(ChatOpenAI):
|
|
| 31 |
|
| 32 |
# ------------------ TOOLS ------------------
|
| 33 |
|
|
|
|
|
|
|
| 34 |
@tool
|
| 35 |
def search_web(query: str) -> str:
|
| 36 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
try:
|
| 38 |
-
|
| 39 |
-
response
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
results.append(f"Related: {topic['Text']}")
|
| 49 |
-
return "\n".join(results) if results else f"No results for '{query}'."
|
| 50 |
-
return f"Search failed with status code {response.status_code}"
|
| 51 |
except Exception as e:
|
| 52 |
-
return f"
|
| 53 |
|
| 54 |
@tool
|
| 55 |
def search_wikipedia(query: str) -> str:
|
| 56 |
-
"""
|
| 57 |
try:
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
except Exception as e:
|
| 66 |
return f"Wikipedia search error: {str(e)}"
|
| 67 |
|
| 68 |
@tool
|
| 69 |
def execute_python(code: str) -> str:
|
| 70 |
-
"""Execute Python code safely and return
|
| 71 |
try:
|
| 72 |
safe_globals = {
|
| 73 |
'__builtins__': {
|
|
@@ -95,7 +107,7 @@ def execute_python(code: str) -> str:
|
|
| 95 |
|
| 96 |
@tool
|
| 97 |
def read_excel_file(file_path: str, sheet_name: Optional[str] = None) -> str:
|
| 98 |
-
"""Read an Excel file and return
|
| 99 |
try:
|
| 100 |
file_path_obj = Path(file_path)
|
| 101 |
if not file_path_obj.exists():
|
|
@@ -119,7 +131,7 @@ def read_excel_file(file_path: str, sheet_name: Optional[str] = None) -> str:
|
|
| 119 |
|
| 120 |
@tool
|
| 121 |
def read_text_file(file_path: str) -> str:
|
| 122 |
-
"""Read a text file and return
|
| 123 |
try:
|
| 124 |
file_path_obj = Path(file_path)
|
| 125 |
if not file_path_obj.exists():
|
|
@@ -128,7 +140,7 @@ def read_text_file(file_path: str) -> str:
|
|
| 128 |
for encoding in encodings:
|
| 129 |
try:
|
| 130 |
with open(file_path_obj, 'r', encoding=encoding) as f:
|
| 131 |
-
return f
|
| 132 |
except UnicodeDecodeError:
|
| 133 |
continue
|
| 134 |
return "Error: Could not decode file with any standard encoding"
|
|
@@ -136,12 +148,10 @@ def read_text_file(file_path: str) -> str:
|
|
| 136 |
return f"Error reading file: {str(e)}"
|
| 137 |
|
| 138 |
# ------------------ GAIA AGENT ------------------
|
| 139 |
-
|
| 140 |
class GaiaAgent:
|
| 141 |
-
"""LangGraph-based agent
|
| 142 |
-
|
| 143 |
def __init__(self):
|
| 144 |
-
print("Initializing GaiaAgent...")
|
| 145 |
self.llm = OpenRouterLLM(
|
| 146 |
model="deepseek/deepseek-v3.1-terminus",
|
| 147 |
temperature=0.1,
|
|
@@ -149,6 +159,8 @@ class GaiaAgent:
|
|
| 149 |
)
|
| 150 |
self.tools = [search_web, search_wikipedia, execute_python, read_excel_file, read_text_file]
|
| 151 |
prompt_modifier = self._get_system_prompt()
|
|
|
|
|
|
|
| 152 |
sig = inspect.signature(create_react_agent)
|
| 153 |
accepted = sig.parameters.keys()
|
| 154 |
kwargs = {}
|
|
@@ -158,18 +170,24 @@ class GaiaAgent:
|
|
| 158 |
kwargs["state_modifier"] = prompt_modifier
|
| 159 |
elif "prompt" in accepted:
|
| 160 |
kwargs["prompt"] = prompt_modifier
|
|
|
|
| 161 |
self.agent = create_react_agent(self.llm, self.tools, **kwargs)
|
| 162 |
print("GaiaAgent initialized successfully!")
|
| 163 |
|
| 164 |
def _get_system_prompt(self) -> str:
|
| 165 |
-
return """You are an advanced AI agent designed to answer complex questions.
|
| 166 |
-
Keep answers concise and factual where possible."""
|
| 167 |
|
| 168 |
-
def __call__(self, question: str) -> str:
|
| 169 |
try:
|
| 170 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 171 |
result = self.agent.invoke({"messages": messages})
|
| 172 |
-
|
|
|
|
| 173 |
return self._clean_answer(answer)
|
| 174 |
except Exception as e:
|
| 175 |
return f"Agent error: {e}"
|
|
@@ -180,94 +198,25 @@ Keep answers concise and factual where possible."""
|
|
| 180 |
parts = re.split(r'final answer:', answer, flags=re.IGNORECASE)
|
| 181 |
if len(parts) > 1:
|
| 182 |
answer = parts[-1].strip()
|
| 183 |
-
prefixes = ["The answer is", "Answer:", "Result:", "Solution:"
|
|
|
|
| 184 |
for prefix in prefixes:
|
| 185 |
if answer.lower().startswith(prefix.lower()):
|
| 186 |
answer = answer[len(prefix):].strip()
|
| 187 |
if answer.startswith(':'):
|
| 188 |
answer = answer[1:].strip()
|
| 189 |
break
|
|
|
|
|
|
|
| 190 |
return answer
|
| 191 |
|
| 192 |
-
# ------------------ RUN AND SUBMIT ------------------
|
| 193 |
-
|
| 194 |
-
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 195 |
-
|
| 196 |
-
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 197 |
-
if profile:
|
| 198 |
-
username = profile.username
|
| 199 |
-
else:
|
| 200 |
-
return "Please login to Hugging Face first.", None
|
| 201 |
-
|
| 202 |
-
space_id = os.getenv("SPACE_ID") or "your_space_username/your_space_name"
|
| 203 |
-
api_url = DEFAULT_API_URL
|
| 204 |
-
questions_url = f"{api_url}/questions"
|
| 205 |
-
submit_url = f"{api_url}/submit"
|
| 206 |
-
|
| 207 |
-
# Instantiate GaiaAgent
|
| 208 |
-
try:
|
| 209 |
-
agent = GaiaAgent()
|
| 210 |
-
except Exception as e:
|
| 211 |
-
return f"Error initializing GaiaAgent: {e}", None
|
| 212 |
-
|
| 213 |
-
# Fetch questions
|
| 214 |
-
try:
|
| 215 |
-
response = requests.get(questions_url, timeout=15)
|
| 216 |
-
response.raise_for_status()
|
| 217 |
-
questions_data = response.json()
|
| 218 |
-
except Exception as e:
|
| 219 |
-
return f"Error fetching questions: {e}", None
|
| 220 |
-
|
| 221 |
-
# Run agent on questions
|
| 222 |
-
answers_payload = []
|
| 223 |
-
results_log = []
|
| 224 |
-
for item in questions_data:
|
| 225 |
-
task_id = item.get("task_id")
|
| 226 |
-
question_text = item.get("question")
|
| 227 |
-
if not task_id or question_text is None:
|
| 228 |
-
continue
|
| 229 |
-
try:
|
| 230 |
-
submitted_answer = agent(question_text)
|
| 231 |
-
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 232 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 233 |
-
except Exception as e:
|
| 234 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
| 235 |
-
|
| 236 |
-
# Prepare submission
|
| 237 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 238 |
-
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
| 239 |
-
|
| 240 |
-
# Submit answers
|
| 241 |
-
try:
|
| 242 |
-
response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 243 |
-
response.raise_for_status()
|
| 244 |
-
result_data = response.json()
|
| 245 |
-
final_status = (
|
| 246 |
-
f"Submission Successful!\n"
|
| 247 |
-
f"User: {result_data.get('username')}\n"
|
| 248 |
-
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
| 249 |
-
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
| 250 |
-
f"Message: {result_data.get('message', 'No message received.')}"
|
| 251 |
-
)
|
| 252 |
-
results_df = pd.DataFrame(results_log)
|
| 253 |
-
return final_status, results_df
|
| 254 |
-
except Exception as e:
|
| 255 |
-
results_df = pd.DataFrame(results_log)
|
| 256 |
-
return f"Submission failed: {e}", results_df
|
| 257 |
-
|
| 258 |
# ------------------ GRADIO INTERFACE ------------------
|
|
|
|
| 259 |
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
gr.LoginButton()
|
| 263 |
-
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 264 |
-
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 265 |
-
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 266 |
|
| 267 |
-
|
| 268 |
-
fn=run_and_submit_all,
|
| 269 |
-
outputs=[status_output, results_table]
|
| 270 |
-
)
|
| 271 |
|
| 272 |
if __name__ == "__main__":
|
| 273 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 1 |
import os
|
| 2 |
+
import re
|
| 3 |
+
import json
|
| 4 |
import requests
|
|
|
|
| 5 |
import pandas as pd
|
|
|
|
| 6 |
from pathlib import Path
|
| 7 |
from typing import Optional
|
| 8 |
from dotenv import load_dotenv
|
| 9 |
+
import inspect
|
| 10 |
+
import gradio as gr
|
| 11 |
|
| 12 |
from langgraph.prebuilt import create_react_agent
|
| 13 |
from langchain_core.messages import HumanMessage
|
|
|
|
| 16 |
|
| 17 |
load_dotenv()
|
| 18 |
|
| 19 |
+
# ------------------ LLM ------------------
|
|
|
|
| 20 |
class OpenRouterLLM(ChatOpenAI):
|
| 21 |
"""Custom OpenRouter LLM wrapper for LangGraph"""
|
|
|
|
| 22 |
def __init__(self, model: str = "deepseek/deepseek-v3.1-terminus", **kwargs):
|
| 23 |
+
api_key = os.getenv("OPENROUTER_API_KEY")
|
| 24 |
+
if not api_key:
|
| 25 |
+
raise ValueError("OPENROUTER_API_KEY not set in environment variables.")
|
| 26 |
super().__init__(
|
| 27 |
model=model,
|
| 28 |
openai_api_key=api_key,
|
|
|
|
| 32 |
|
| 33 |
# ------------------ TOOLS ------------------
|
| 34 |
|
| 35 |
+
SERPAPI_KEY = os.getenv("SERPAPI_KEY")
|
| 36 |
+
|
| 37 |
@tool
|
| 38 |
def search_web(query: str) -> str:
|
| 39 |
+
"""Perform a reliable web search using SerpAPI."""
|
| 40 |
+
if not SERPAPI_KEY:
|
| 41 |
+
return "Error: SERPAPI_KEY not set."
|
| 42 |
+
search_url = "https://serpapi.com/search.json"
|
| 43 |
+
params = {"q": query, "api_key": SERPAPI_KEY, "num": 3}
|
| 44 |
try:
|
| 45 |
+
response = requests.get(search_url, params=params, timeout=10)
|
| 46 |
+
response.raise_for_status()
|
| 47 |
+
data = response.json()
|
| 48 |
+
results = []
|
| 49 |
+
for item in data.get("organic_results", []):
|
| 50 |
+
title = item.get("title", "")
|
| 51 |
+
snippet = item.get("snippet", "")
|
| 52 |
+
link = item.get("link", "")
|
| 53 |
+
results.append(f"{title}\n{snippet}\n{link}")
|
| 54 |
+
return "\n\n".join(results) if results else f"No results for '{query}'."
|
|
|
|
|
|
|
|
|
|
| 55 |
except Exception as e:
|
| 56 |
+
return f"Web search error: {str(e)}"
|
| 57 |
|
| 58 |
@tool
|
| 59 |
def search_wikipedia(query: str) -> str:
|
| 60 |
+
"""Retrieve full Wikipedia article text."""
|
| 61 |
try:
|
| 62 |
+
url = f"https://en.wikipedia.org/w/api.php"
|
| 63 |
+
params = {
|
| 64 |
+
"action": "query",
|
| 65 |
+
"format": "json",
|
| 66 |
+
"prop": "extracts",
|
| 67 |
+
"explaintext": True,
|
| 68 |
+
"titles": query
|
| 69 |
+
}
|
| 70 |
+
response = requests.get(url, params=params, timeout=10)
|
| 71 |
+
response.raise_for_status()
|
| 72 |
+
pages = response.json()["query"]["pages"]
|
| 73 |
+
text = next(iter(pages.values())).get("extract", "")
|
| 74 |
+
if not text:
|
| 75 |
+
return f"No Wikipedia content found for '{query}'."
|
| 76 |
+
return text[:2000] # truncate if too long
|
| 77 |
except Exception as e:
|
| 78 |
return f"Wikipedia search error: {str(e)}"
|
| 79 |
|
| 80 |
@tool
|
| 81 |
def execute_python(code: str) -> str:
|
| 82 |
+
"""Execute Python code safely and return output."""
|
| 83 |
try:
|
| 84 |
safe_globals = {
|
| 85 |
'__builtins__': {
|
|
|
|
| 107 |
|
| 108 |
@tool
|
| 109 |
def read_excel_file(file_path: str, sheet_name: Optional[str] = None) -> str:
|
| 110 |
+
"""Read an Excel file and return contents."""
|
| 111 |
try:
|
| 112 |
file_path_obj = Path(file_path)
|
| 113 |
if not file_path_obj.exists():
|
|
|
|
| 131 |
|
| 132 |
@tool
|
| 133 |
def read_text_file(file_path: str) -> str:
|
| 134 |
+
"""Read a text file and return contents."""
|
| 135 |
try:
|
| 136 |
file_path_obj = Path(file_path)
|
| 137 |
if not file_path_obj.exists():
|
|
|
|
| 140 |
for encoding in encodings:
|
| 141 |
try:
|
| 142 |
with open(file_path_obj, 'r', encoding=encoding) as f:
|
| 143 |
+
return f.read()
|
| 144 |
except UnicodeDecodeError:
|
| 145 |
continue
|
| 146 |
return "Error: Could not decode file with any standard encoding"
|
|
|
|
| 148 |
return f"Error reading file: {str(e)}"
|
| 149 |
|
| 150 |
# ------------------ GAIA AGENT ------------------
|
|
|
|
| 151 |
class GaiaAgent:
|
| 152 |
+
"""LangGraph-based agent with DeepSeek and enhanced tools."""
|
|
|
|
| 153 |
def __init__(self):
|
| 154 |
+
print("Initializing GaiaAgent with LangGraph and OpenRouter DeepSeek...")
|
| 155 |
self.llm = OpenRouterLLM(
|
| 156 |
model="deepseek/deepseek-v3.1-terminus",
|
| 157 |
temperature=0.1,
|
|
|
|
| 159 |
)
|
| 160 |
self.tools = [search_web, search_wikipedia, execute_python, read_excel_file, read_text_file]
|
| 161 |
prompt_modifier = self._get_system_prompt()
|
| 162 |
+
|
| 163 |
+
# Detect correct kwarg for your LangGraph version
|
| 164 |
sig = inspect.signature(create_react_agent)
|
| 165 |
accepted = sig.parameters.keys()
|
| 166 |
kwargs = {}
|
|
|
|
| 170 |
kwargs["state_modifier"] = prompt_modifier
|
| 171 |
elif "prompt" in accepted:
|
| 172 |
kwargs["prompt"] = prompt_modifier
|
| 173 |
+
|
| 174 |
self.agent = create_react_agent(self.llm, self.tools, **kwargs)
|
| 175 |
print("GaiaAgent initialized successfully!")
|
| 176 |
|
| 177 |
def _get_system_prompt(self) -> str:
|
| 178 |
+
return """You are an advanced AI agent designed to answer complex questions using all available tools, including web search, Wikipedia, Python execution, Excel and text file reading."""
|
|
|
|
| 179 |
|
| 180 |
+
def __call__(self, task_id: str, question: str) -> str:
|
| 181 |
try:
|
| 182 |
+
print(f"Processing task {task_id}: {question[:100]}...")
|
| 183 |
+
# Combine context from tools for better answers
|
| 184 |
+
wiki_text = search_wikipedia(question)
|
| 185 |
+
web_text = search_web(question)
|
| 186 |
+
combined_input = f"{wiki_text}\n\n{web_text}\n\nQuestion: {question}"
|
| 187 |
+
messages = [HumanMessage(content=combined_input)]
|
| 188 |
result = self.agent.invoke({"messages": messages})
|
| 189 |
+
final_message = result["messages"][-1]
|
| 190 |
+
answer = final_message.content
|
| 191 |
return self._clean_answer(answer)
|
| 192 |
except Exception as e:
|
| 193 |
return f"Agent error: {e}"
|
|
|
|
| 198 |
parts = re.split(r'final answer:', answer, flags=re.IGNORECASE)
|
| 199 |
if len(parts) > 1:
|
| 200 |
answer = parts[-1].strip()
|
| 201 |
+
prefixes = ["The answer is", "Answer:", "Result:", "Solution:",
|
| 202 |
+
"Based on", "Therefore", "In conclusion", "So the answer is"]
|
| 203 |
for prefix in prefixes:
|
| 204 |
if answer.lower().startswith(prefix.lower()):
|
| 205 |
answer = answer[len(prefix):].strip()
|
| 206 |
if answer.startswith(':'):
|
| 207 |
answer = answer[1:].strip()
|
| 208 |
break
|
| 209 |
+
if len(answer.split()) <= 3:
|
| 210 |
+
answer = answer.strip('"\'.')
|
| 211 |
return answer
|
| 212 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 213 |
# ------------------ GRADIO INTERFACE ------------------
|
| 214 |
+
agent = GaiaAgent()
|
| 215 |
|
| 216 |
+
def run_agent(prompt: str) -> str:
|
| 217 |
+
return agent("gaia_task", prompt)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
|
| 219 |
+
demo = gr.Interface(fn=run_agent, inputs="text", outputs="text", title="GAIA Agent")
|
|
|
|
|
|
|
|
|
|
| 220 |
|
| 221 |
if __name__ == "__main__":
|
| 222 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|