Complete GAIA agent with LlamaIndex - fixed all issues

Dockerfile

@@ -1,17 +1,71 @@
-# Use a base Python image
+# Use a base Python image with better compatibility
 FROM python:3.10-slim
 
+# Set environment variables to fix permission issues
+ENV PYTHONUNBUFFERED=1
+ENV NLTK_DATA=/app/nltk_data
+ENV MPLCONFIGDIR=/app/matplotlib_cache
+ENV HF_HOME=/app/huggingface_cache
+ENV TORCH_HOME=/app/torch_cache
+ENV TRANSFORMERS_CACHE=/app/huggingface_cache
+ENV GRADIO_SERVER_NAME=0.0.0.0
+ENV GRADIO_SERVER_PORT=7860
+
+# Create app user and group for better security
+RUN groupadd -r appuser && useradd -r -g appuser appuser
+
 # Set working directory
 WORKDIR /app
 
-# Copy code
+# Create cache directories with proper permissions
+RUN mkdir -p /app/nltk_data \
+    /app/matplotlib_cache \
+    /app/huggingface_cache \
+    /app/torch_cache \
+    /app/temp && \
+    chown -R appuser:appuser /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    gcc \
+    g++ \
+    git \
+    curl \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements first for better Docker layer caching
+COPY requirements.txt .
+
+# Install Python dependencies
+RUN pip install --upgrade pip && \
+    pip install --no-cache-dir -r requirements.txt
+
+# Copy application code
 COPY . .
 
-# Install dependencies
-RUN pip install --upgrade pip && pip install --no-cache-dir -r requirements.txt
+# Change ownership of all app files to appuser
+RUN chown -R appuser:appuser /app
+
+# Switch to non-root user
+USER appuser
+
+# Create a startup script to handle initialization
+RUN echo '#!/bin/bash\n\
+echo "Starting GAIA Agent..."\n\
+echo "Environment check:"\n\
+echo "NLTK_DATA: $NLTK_DATA"\n\
+echo "HF_HOME: $HF_HOME"\n\
+echo "MPLCONFIGDIR: $MPLCONFIGDIR"\n\
+echo "Working directory: $(pwd)"\n\
+echo "User: $(whoami)"\n\
+python app.py' > /app/start.sh && chmod +x /app/start.sh
+
+# Expose the port
+EXPOSE 7860
 
-# Set gradio port
-ENV PORT 7860
+# Health check
+HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
+    CMD curl -f http://localhost:7860/ || exit 1
 
-# Run app
-CMD ["python", "app.py"]
+# Run the application
+CMD ["/app/start.sh"]

app.py

@@ -3,14 +3,41 @@ import sys
 import json
 import traceback
 from typing import List, Dict
+import warnings
+
+# Suppress warnings for cleaner output
+warnings.filterwarnings("ignore", category=FutureWarning)
+warnings.filterwarnings("ignore", category=UserWarning)
 
 import gradio as gr
 
 # --- Environment variable setup to fix permission issues ---
-os.environ["NLTK_DATA"] = "/tmp/nltk_data"
-os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
-os.environ["HF_HOME"] = "/tmp/huggingface_cache"  # Updated from TRANSFORMERS_CACHE
-os.environ["TORCH_HOME"] = "/tmp/torch_cache"
+def setup_environment():
+    """Setup environment variables and create necessary directories"""
+    env_vars = {
+        "NLTK_DATA": "/app/nltk_data",
+        "MPLCONFIGDIR": "/app/matplotlib_cache", 
+        "HF_HOME": "/app/huggingface_cache",
+        "TORCH_HOME": "/app/torch_cache",
+        "TRANSFORMERS_CACHE": "/app/huggingface_cache"
+    }
+    
+    for var, path in env_vars.items():
+        os.environ[var] = path
+        # Create directory if it doesn't exist
+        try:
+            os.makedirs(path, exist_ok=True)
+            print(f"✅ Created/verified directory: {path}")
+        except PermissionError:
+            print(f"⚠️ Permission denied for {path}, using /tmp fallback")
+            fallback_path = f"/tmp/{var.lower()}"
+            os.environ[var] = fallback_path
+            os.makedirs(fallback_path, exist_ok=True)
+        except Exception as e:
+            print(f"❌ Error setting up {var}: {e}")
+
+# Setup environment first
+setup_environment()
 
 # Import nltk AFTER setting environment variables
 try:
@@ -18,60 +45,93 @@ try:
     # Download required NLTK data upfront
     nltk.download('punkt', download_dir=os.environ["NLTK_DATA"], quiet=True)
     nltk.download('stopwords', download_dir=os.environ["NLTK_DATA"], quiet=True)
+    print("✅ NLTK data downloaded successfully")
 except Exception as e:
-    print(f"NLTK setup warning: {e}")
+    print(f"⚠️ NLTK setup warning: {e}")
 
 # Add current directory to path for local imports
 sys.path.append(os.path.dirname(os.path.abspath(__file__)))
 
-# Import GaiaAPI early (this fixes the undefined error)
-from utils.gaia_api import GaiaAPI
+# Import dependencies with better error handling
+try:
+    from utils.gaia_api import GaiaAPI
+    print("✅ GaiaAPI imported successfully")
+except ImportError as e:
+    print(f"⚠️ Failed to import GaiaAPI: {e}")
+    # Create a fallback GaiaAPI
+    class GaiaAPI:
+        @classmethod
+        def get_questions(cls):
+            return [{"task_id": "fallback", "question": "What is 2+2?"}]
+        @classmethod 
+        def get_random_question(cls):
+            return {"task_id": "fallback", "question": "What is 2+2?"}
+        @classmethod
+        def submit_answers(cls, username, code_url, answers):
+            return {"error": "GaiaAPI not available", "score": 0}
 
 # Initialize global agent state
 AGENT_READY = False
 agent = None
 initialization_error = None
+agent_info = {}
 
-# Import agent-related modules and initialize
-try:
-    print("Importing modules...")
-    from agent.local_llm import LocalLLM
-    from agent.tools import gaia_tools
-    from llama_index.core.agent import ReActAgent
-    from llama_index.core.memory import ChatMemoryBuffer
-
-    print("All imports successful!")
-
-    print("Initializing Local LLM...")
-    local_llm = LocalLLM()
-    llm = local_llm.get_llm()
-
-    print("Creating ReAct Agent...")
-    memory = ChatMemoryBuffer.from_defaults(token_limit=2000)
+def initialize_agent():
+    """Initialize the LlamaIndex agent with comprehensive error handling"""
+    global agent, AGENT_READY, initialization_error, agent_info
     
-    # Only create ReAct agent if we have a proper LLM
-    if hasattr(llm, 'chat') and not llm.__class__.__name__ == 'MockLLM':
-        agent = ReActAgent.from_tools(
-            tools=gaia_tools,
-            llm=llm,
-            memory=memory,
-            verbose=True,
-            max_iterations=3
-        )
-        print("Agent initialized successfully!")
-        AGENT_READY = True
-    else:
-        print("Using mock mode - agent partially ready")
-        agent = llm  # Use the mock LLM directly
-        AGENT_READY = True
+    try:
+        print("🔄 Starting agent initialization...")
+        
+        # Import agent-related modules
+        print("📦 Importing modules...")
+        from agent.local_llm import LocalLLM
+        from agent.tools import gaia_tools
+        from llama_index.core.agent import ReActAgent
+        from llama_index.core.memory import ChatMemoryBuffer
+        
+        agent_info["modules_imported"] = True
+        print("✅ All modules imported successfully!")
 
-except Exception as e:
-    print(f"Failed to initialize agent: {str(e)}")
-    traceback.print_exc()
-    AGENT_READY = False
-    agent = None
-    initialization_error = str(e)
+        print("🤖 Initializing Local LLM...")
+        local_llm = LocalLLM()
+        llm = local_llm.get_llm()
+        agent_info["llm_type"] = llm.__class__.__name__
+        
+        print("🧠 Creating ReAct Agent...")
+        memory = ChatMemoryBuffer.from_defaults(token_limit=2000)
+        
+        # Check if we have a proper LLM or mock
+        if hasattr(llm, 'chat') and llm.__class__.__name__ != 'MockLLM':
+            agent = ReActAgent.from_tools(
+                tools=gaia_tools,
+                llm=llm,
+                memory=memory,
+                verbose=True,
+                max_iterations=3
+            )
+            agent_info["agent_type"] = "ReActAgent"
+            print("✅ ReAct Agent initialized successfully!")
+        else:
+            agent = llm  # Use the mock LLM directly
+            agent_info["agent_type"] = "MockLLM"
+            print("⚠️ Using mock mode - agent partially ready")
+        
+        agent_info["tools_count"] = len(gaia_tools) if 'gaia_tools' in locals() else 0
+        AGENT_READY = True
+        print("🎉 Agent initialization complete!")
+        
+    except Exception as e:
+        error_msg = f"Failed to initialize agent: {str(e)}"
+        print(f"❌ {error_msg}")
+        traceback.print_exc()
+        AGENT_READY = False
+        agent = None
+        initialization_error = error_msg
+        agent_info["error"] = error_msg
 
+# Initialize agent
+initialize_agent()
 
 def process_single_question(question_text: str) -> str:
     """Process a single GAIA question through the agent"""
@@ -91,6 +151,8 @@ Answer the following question directly and concisely. Do not include "FINAL ANSW
 Question: {question_text}
 """
         
+        print(f"🤔 Processing question: {question_text[:50]}...")
+        
         # Handle both ReAct agent and mock LLM
         if hasattr(agent, 'query'):
             response = agent.query(enhanced_prompt)
@@ -107,11 +169,13 @@ Question: {question_text}
             if answer.startswith(prefix):
                 answer = answer[len(prefix):].strip()
 
+        print(f"✅ Generated answer: {answer[:50]}...")
         return answer
 
     except Exception as e:
-        return f"❌ Error: {str(e)}"
-
+        error_msg = f"❌ Error processing question: {str(e)}"
+        print(error_msg)
+        return error_msg
 
 def process_all_questions() -> str:
     """Process all GAIA questions and prepare answers for submission"""
@@ -119,9 +183,11 @@ def process_all_questions() -> str:
         return "❌ Agent not ready."
 
     try:
+        print("📥 Fetching all GAIA questions...")
         questions = GaiaAPI.get_questions()
         processed_answers = []
 
+        print(f"🔄 Processing {len(questions)} questions...")
         for i, question in enumerate(questions):
             print(f"Processing question {i + 1}/{len(questions)}: {question['task_id']}")
             answer = process_single_question(question['question'])
@@ -130,20 +196,25 @@ def process_all_questions() -> str:
                 "submitted_answer": answer
             })
 
-        with open("gaia_answers.json", "w") as f:
+        # Save answers to file
+        output_file = "/app/gaia_answers.json"
+        with open(output_file, "w") as f:
             json.dump(processed_answers, f, indent=2)
 
         summary = f"✅ Processed {len(processed_answers)} questions.\n"
-        summary += "Answers saved to gaia_answers.json\n"
-        summary += "First 3 answers:\n"
+        summary += f"💾 Answers saved to {output_file}\n"
+        summary += "📋 First 3 answers:\n"
         for ans in processed_answers[:3]:
             summary += f"- {ans['task_id']}: {ans['submitted_answer'][:50]}...\n"
 
+        print(summary)
         return summary
 
     except Exception as e:
-        return f"❌ Error: {str(e)}"
-
+        error_msg = f"❌ Error processing questions: {str(e)}"
+        print(error_msg)
+        traceback.print_exc()
+        return error_msg
 
 def submit_to_gaia(username: str, code_url: str) -> str:
     """Submit answers to GAIA benchmark"""
@@ -154,8 +225,10 @@ def submit_to_gaia(username: str, code_url: str) -> str:
         return "❌ Please provide both username and code URL."
 
     try:
-        with open("gaia_answers.json", "r") as f:
+        answers_file = "/app/gaia_answers.json"
+        with open(answers_file, "r") as f:
             answers = json.load(f)
+        print(f"📤 Submitting {len(answers)} answers...")
     except FileNotFoundError:
         return "❌ No processed answers found. Please process them first."
 
@@ -164,10 +237,13 @@ def submit_to_gaia(username: str, code_url: str) -> str:
         if "error" in result:
             return f"❌ Submission failed: {result['error']}"
         score = result.get("score", "Unknown")
-        return f"✅ Submission successful!\n📊 Score: {score}"
+        success_msg = f"✅ Submission successful!\n📊 Score: {score}"
+        print(success_msg)
+        return success_msg
     except Exception as e:
-        return f"❌ Submission error: {str(e)}"
-
+        error_msg = f"❌ Submission error: {str(e)}"
+        print(error_msg)
+        return error_msg
 
 def get_sample_question() -> str:
     """Load a sample question for testing"""
@@ -177,6 +253,37 @@ def get_sample_question() -> str:
     except Exception as e:
         return f"Error loading sample question: {str(e)}"
 
+def get_system_status() -> str:
+    """Get detailed system status for debugging"""
+    status = "🔍 System Status:\n\n"
+    
+    # Agent status
+    status += f"🤖 Agent Ready: {'✅ Yes' if AGENT_READY else '❌ No'}\n"
+    if initialization_error:
+        status += f"❌ Error: {initialization_error}\n"
+    
+    # Agent info
+    status += f"🧠 LLM Type: {agent_info.get('llm_type', 'Unknown')}\n"
+    status += f"🔧 Agent Type: {agent_info.get('agent_type', 'Unknown')}\n"
+    status += f"🛠️ Tools Count: {agent_info.get('tools_count', 0)}\n"
+    
+    # Environment
+    status += "\n📁 Environment Variables:\n"
+    for var in ["NLTK_DATA", "HF_HOME", "MPLCONFIGDIR", "TORCH_HOME"]:
+        path = os.environ.get(var, 'Not set')
+        exists = "✅" if os.path.exists(path) else "❌"
+        status += f"  {var}: {path} {exists}\n"
+    
+    # Directory permissions
+    status += "\n📂 Directory Status:\n"
+    for path in ["/app", "/tmp"]:
+        try:
+            writable = os.access(path, os.W_OK)
+            status += f"  {path}: {'✅ Writable' if writable else '❌ Not writable'}\n"
+        except:
+            status += f"  {path}: ❌ Error checking\n"
+    
+    return status
 
 # ---------- Gradio UI ----------
 with gr.Blocks(title="🦙 GAIA LlamaIndex Agent", theme=gr.themes.Soft()) as demo:
@@ -241,31 +348,21 @@ Submit your processed answers to the GAIA benchmark for official scoring.
 
         submit_btn.click(submit_to_gaia, inputs=[username_input, code_url_input], outputs=submission_output)
 
-    with gr.Tab("ℹ️ Info & Debug"):
-        gr.Markdown(f"""
-## About This Agent
-
-This agent uses:
-- **LlamaIndex** (ReAct Agent + Tools)
-- **Local LLM** (DialoGPT-small with fallback to GPT2 or mock)
-- **GAIA Tools** (question fetch, file reader, math, etc.)
-
-## Current Status
-- Agent Ready: {"✅ Yes" if AGENT_READY else "❌ No"}
-- Tools Loaded: {len(gaia_tools) if 'gaia_tools' in globals() else 0}
-- Initialization Error: {initialization_error or "None"}
-
-## Environment Variables Set
-- NLTK_DATA: {os.environ.get('NLTK_DATA', 'Not set')}
-- HF_HOME: {os.environ.get('HF_HOME', 'Not set')}
-- MPLCONFIGDIR: {os.environ.get('MPLCONFIGDIR', 'Not set')}
-
-## Usage Tips
-1. Start with the "Test Single Question" tab
-2. Try the sample question first
-3. If agent works, proceed to "Full Evaluation"
-4. Submit to GAIA when ready
-""")
+    with gr.Tab("ℹ️ System Status"):
+        gr.Markdown("## System Information and Debugging")
+        
+        refresh_btn = gr.Button("🔄 Refresh Status")
+        status_output = gr.Textbox(label="System Status", lines=20, interactive=False)
+        
+        # Load initial status
+        demo.load(get_system_status, outputs=status_output)
+        refresh_btn.click(get_system_status, outputs=status_output)
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
+    print("🚀 Starting Gradio interface...")
+    demo.launch(
+        server_name="0.0.0.0", 
+        server_port=7860, 
+        show_error=True,
+        show_tips=True
+    )

requirements.txt

@@ -1,11 +1,27 @@
-llama-index==0.10.0
-llama-index-llms-huggingface==0.2.0
-transformers==4.39.3
-torch==2.1.2
-gradio==3.41.0
-requests==2.31.0
-accelerate==0.23.0
-sentence-transformers==2.2.2
-python-dotenv==1.0.0
-nltk==3.8.1
-huggingface-hub==0.23.0  
+# Core dependencies
+gradio>=4.0.0
+requests>=2.31.0
+
+# LlamaIndex core
+llama-index>=0.10.0
+llama-index-llms-huggingface>=0.2.0
+
+# Transformers and ML libraries
+transformers>=4.35.0
+torch>=2.0.0
+tokenizers>=0.15.0
+
+# NLP dependencies
+nltk>=3.8.1
+
+# Data processing
+numpy>=1.24.0
+pandas>=2.0.0
+
+# Optional: for better model performance
+accelerate>=0.24.0
+bitsandbytes>=0.41.0
+
+# Utilities
+typing-extensions>=4.5.0
+pydantic>=2.0.0