final / enhanced_agent.py
yoshizen's picture
Upload 4 files
162ee47 verified
"""
Enhanced GAIA-Ready AI Agent with integrated memory and reasoning systems
This is the main integration file that combines the agent, memory system,
and reasoning system into a complete solution for the Hugging Face Agents Course.
"""
import os
import sys
import json
import traceback
from typing import List, Dict, Any, Optional, Union
from datetime import datetime
# Import the memory and reasoning systems
try:
from memory_system import EnhancedMemoryManager
from reasoning_system import ReasoningSystem
except ImportError:
print("Error: Could not import memory_system or reasoning_system modules.")
print("Make sure memory_system.py and reasoning_system.py are in the same directory.")
sys.exit(1)
# Import smolagents
try:
from smolagents import Agent, InferenceClientModel, Tool, LiteLLMModel
except ImportError:
import subprocess
subprocess.check_call(["pip", "install", "smolagents"])
from smolagents import Agent, InferenceClientModel, Tool
try:
from smolagents import LiteLLMModel
except ImportError:
print("Warning: LiteLLMModel not available, will use InferenceClientModel only.")
# Import tool implementations
from agent import (
web_search_function,
web_page_content_function,
calculator_function,
python_executor_function,
image_analyzer_function,
text_processor_function,
file_manager_function
)
class EnhancedGAIAAgent:
"""
Enhanced AI Agent designed to perform well on the GAIA benchmark
Integrates memory and reasoning systems with the Think-Act-Observe workflow
"""
def __init__(self, api_key=None, use_local_model=False, use_semantic_memory=True):
"""
Initialize the enhanced GAIA agent
Args:
api_key: API key for Hugging Face Inference API
use_local_model: Whether to use a local model via Ollama
use_semantic_memory: Whether to use semantic search for memory retrieval
"""
# Initialize the memory system
self.memory_manager = EnhancedMemoryManager(use_semantic_search=use_semantic_memory)
# Initialize the LLM model
if use_local_model:
# Use Ollama for local model
try:
self.model = LiteLLMModel(
model_id="ollama_chat/qwen2:7b",
api_base="http://127.0.0.1:11434",
num_ctx=8192,
)
print("Using local Ollama model: qwen2:7b")
except Exception as e:
print(f"Error initializing local model: {str(e)}")
print("Falling back to Hugging Face Inference API")
self.model = InferenceClientModel(
model_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
api_key=api_key or os.environ.get("HF_API_KEY", "")
)
print("Using Hugging Face Inference API model: Mixtral-8x7B")
else:
# Use Hugging Face Inference API
self.model = InferenceClientModel(
model_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
api_key=api_key or os.environ.get("HF_API_KEY", "")
)
print("Using Hugging Face Inference API model: Mixtral-8x7B")
# Define tools
self.tools = [
Tool(
name="web_search",
description="Search the web for information",
function=web_search_function
),
Tool(
name="web_page_content",
description="Fetch and extract content from a web page",
function=web_page_content_function
),
Tool(
name="calculator",
description="Perform mathematical calculations",
function=calculator_function
),
Tool(
name="image_analyzer",
description="Analyze image content",
function=image_analyzer_function
),
Tool(
name="python_executor",
description="Execute Python code",
function=python_executor_function
),
Tool(
name="text_processor",
description="Process and analyze text",
function=text_processor_function
),
Tool(
name="file_manager",
description="Save and load data from files",
function=file_manager_function
)
]
# Enhanced system prompt for GAIA benchmark
self.system_prompt = """
You are an advanced AI assistant designed to solve complex tasks from the GAIA benchmark.
You have access to various tools that can help you solve these tasks.
Always follow the Think-Act-Observe workflow:
1. Think: Carefully analyze the task and plan your approach
- Break down complex tasks into smaller steps
- Consider what information you need and how to get it
- Plan your approach before taking action
2. Act: Use appropriate tools to gather information or perform actions
- web_search: Search the web for information
- web_page_content: Extract content from specific web pages
- calculator: Perform mathematical calculations
- image_analyzer: Analyze image content
- python_executor: Run Python code for complex operations
- text_processor: Process and analyze text (summarize, analyze_sentiment, extract_keywords)
- file_manager: Save and load data from files (save, load)
3. Observe: Analyze the results of your actions and adjust your approach
- Verify if the information answers the original question
- Identify any gaps or inconsistencies
- Determine if additional actions are needed
For complex tasks:
- Break them down into smaller, manageable steps
- Keep track of your progress and intermediate results
- Verify each step before moving to the next
- Always double-check your final answer
When reasoning:
- Be thorough and methodical
- Consider multiple perspectives
- Explain your thought process clearly
- Cite sources when providing factual information
Remember that the GAIA benchmark tests your ability to:
- Reason effectively about complex problems
- Understand and process multimodal information
- Navigate the web to find information
- Use tools appropriately to solve tasks
Always verify your answers before submitting them.
"""
# Initialize the base agent
self.base_agent = Agent(
model=self.model,
tools=self.tools,
system_prompt=self.system_prompt
)
# Initialize the reasoning system
self.reasoning_system = ReasoningSystem(self.base_agent, self.memory_manager)
# Error handling and recovery settings
self.max_retries = 3
self.error_log = []
def solve(self, query: str, max_iterations: int = 5, verbose: bool = True) -> Dict[str, Any]:
"""
Solve a task using the enhanced Think-Act-Observe workflow
Args:
query: The user's query or task
max_iterations: Maximum number of iterations
verbose: Whether to print detailed progress
Returns:
Dictionary containing the final answer and metadata
"""
start_time = datetime.now()
if verbose:
print(f"\n{'='*50}")
print(f"Starting to solve: {query}")
print(f"{'='*50}\n")
try:
# Execute the reasoning cycle
final_answer = self.reasoning_system.execute_reasoning_cycle(query, max_iterations)
# Record execution time
execution_time = (datetime.now() - start_time).total_seconds()
if verbose:
print(f"\n{'='*50}")
print(f"Task completed in {execution_time:.2f} seconds")
print(f"{'='*50}\n")
# Get memory summary for debugging
memory_summary = self.memory_manager.get_memory_summary()
return {
"query": query,
"answer": final_answer,
"execution_time": execution_time,
"iterations": max_iterations,
"memory_summary": memory_summary,
"success": True,
"error": None
}
except Exception as e:
error_msg = f"Error solving task: {str(e)}\n{traceback.format_exc()}"
print(error_msg)
# Record the error
self.error_log.append({
"timestamp": datetime.now().isoformat(),
"query": query,
"error": str(e),
"traceback": traceback.format_exc()
})
# Try to recover and provide a partial answer
try:
recovery_prompt = f"""
I encountered an error while trying to solve this task: {query}
The error was: {str(e)}
Based on what I know so far, please provide the best possible answer or explanation.
If you can't provide a complete answer, explain what you do know and what information is missing.
"""
recovery_answer = self.base_agent.chat(recovery_prompt)
execution_time = (datetime.now() - start_time).total_seconds()
if verbose:
print(f"\n{'='*50}")
print(f"Task completed with recovery in {execution_time:.2f} seconds")
print(f"{'='*50}\n")
return {
"query": query,
"answer": recovery_answer,
"execution_time": execution_time,
"iterations": 0,
"success": False,
"error": str(e),
"recovery": True
}
except Exception as recovery_error:
# If recovery fails, return a basic error message
return {
"query": query,
"answer": f"I'm sorry, I encountered an error while solving this task and couldn't recover: {str(e)}",
"execution_time": (datetime.now() - start_time).total_seconds(),
"iterations": 0,
"success": False,
"error": str(e),
"recovery_error": str(recovery_error),
"recovery": False
}
def batch_solve(self, queries: List[str], max_iterations: int = 5, verbose: bool = True) -> List[Dict[str, Any]]:
"""
Solve multiple tasks in batch
Args:
queries: List of user queries or tasks
max_iterations: Maximum number of iterations per query
verbose: Whether to print detailed progress
Returns:
List of results for each query
"""
results = []
for i, query in enumerate(queries):
if verbose:
print(f"\n{'='*50}")
print(f"Processing task {i+1}/{len(queries)}: {query}")
print(f"{'='*50}\n")
result = self.solve(query, max_iterations, verbose)
results.append(result)
# Clear working memory between tasks
self.memory_manager.clear_working_memory()
return results
def save_results(self, results: Union[Dict[str, Any], List[Dict[str, Any]]], filename: str = "gaia_results.json") -> None:
"""
Save results to a file
Args:
results: Results from solve() or batch_solve()
filename: Name of the file to save results to
"""
try:
with open(filename, 'w') as f:
json.dump(results, f, indent=2)
print(f"Results saved to {filename}")
except Exception as e:
print(f"Error saving results: {str(e)}")
def load_results(self, filename: str = "gaia_results.json") -> Union[Dict[str, Any], List[Dict[str, Any]]]:
"""
Load results from a file
Args:
filename: Name of the file to load results from
Returns:
Loaded results
"""
try:
with open(filename, 'r') as f:
results = json.load(f)
print(f"Results loaded from {filename}")
return results
except Exception as e:
print(f"Error loading results: {str(e)}")
return []
def evaluate_performance(self, results: List[Dict[str, Any]]) -> Dict[str, Any]:
"""
Evaluate performance metrics from batch results
Args:
results: Results from batch_solve()
Returns:
Dictionary of performance metrics
"""
if not results:
return {"error": "No results to evaluate"}
total_queries = len(results)
successful_queries = sum(1 for r in results if r.get("success", False))
recovery_queries = sum(1 for r in results if not r.get("success", False) and r.get("recovery", False))
failed_queries = total_queries - successful_queries - recovery_queries
avg_execution_time = sum(r.get("execution_time", 0) for r in results) / total_queries
return {
"total_queries": total_queries,
"successful_queries": successful_queries,
"recovery_queries": recovery_queries,
"failed_queries": failed_queries,
"success_rate": successful_queries / total_queries if total_queries > 0 else 0,
"recovery_rate": recovery_queries / total_queries if total_queries > 0 else 0,
"failure_rate": failed_queries / total_queries if total_queries > 0 else 0,
"avg_execution_time": avg_execution_time
}
# Example usage
if __name__ == "__main__":
# Initialize the agent
agent = EnhancedGAIAAgent(use_local_model=False, use_semantic_memory=True)
# Example GAIA-style queries
sample_queries = [
"What is the capital of France and what is its population? Also, calculate 15% of this population.",
"Who was the first person to walk on the moon? What year did this happen?",
"Explain the concept of photosynthesis in simple terms."
]
# Solve a single query
print("\nSolving single query...")
result = agent.solve(sample_queries[0])
print("\nFinal Answer:")
print(result["answer"])
# Uncomment to solve batch queries
# print("\nSolving batch queries...")
# batch_results = agent.batch_solve(sample_queries)
#
# # Save results
# agent.save_results(batch_results)
#
# # Evaluate performance
# performance = agent.evaluate_performance(batch_results)
# print("\nPerformance Metrics:")
# for key, value in performance.items():
# print(f"{key}: {value}")