Error Handling

This document details the error handling capabilities in Ollama Toolkit v0.1.9, following Eidosian principles of self-awareness and structural robustness.

Exception Hierarchy

Ollama Toolkit provides a precise hierarchy of exception types for optimal error handling:

OllamaAPIError (base)
├── ConnectionError
├── TimeoutError
├── ModelNotFoundError
├── ServerError
├── InvalidRequestError
├── StreamingError
├── ParseError
├── AuthenticationError
├── EndpointNotFoundError
├── ModelCompatibilityError
└── StreamingTimeoutError

Each exception type serves a specific purpose, allowing for precise handling of different error scenarios.

Basic Error Handling Pattern

from ollama_toolkit import OllamaClient
from ollama_toolkit.exceptions import ModelNotFoundError, ConnectionError, TimeoutError, OllamaAPIError

client = OllamaClient()

try:
    response = client.generate(
        model="nonexistent-model",
        prompt="This won't work",
        stream=False
    )
except ModelNotFoundError as e:
    print(f"Model not found: {e}")
    # Handle missing model (e.g., suggest alternatives)
except ConnectionError as e:
    print(f"Connection error: {e}")
    # Handle connection issues (e.g., check if server is running)
except TimeoutError as e:
    print(f"Request timed out: {e}")
    # Handle timeout (e.g., suggest using a smaller model)
except OllamaAPIError as e:
    print(f"API error: {e}")
    # Generic error handling

Advanced Fallback Mechanisms

Version 0.1.9 introduces sophisticated fallback mechanisms that operate at multiple levels:

Model Fallback

from ollama_toolkit import OllamaClient
from ollama_toolkit.utils.model_constants import get_fallback_model

client = OllamaClient()

def generate_with_fallback(model, prompt):
    try:
        return client.generate(model=model, prompt=prompt)
    except ModelNotFoundError:
        fallback_model = get_fallback_model(model)
        print(f"Model '{model}' not found. Using fallback model: {fallback_model}")
        return client.generate(model=fallback_model, prompt=prompt)

# Example usage
response = generate_with_fallback("missing-model", "Hello, world!")
print(response.get("response", ""))

Comprehensive Fallback Strategy

from ollama_toolkit import OllamaClient
from ollama_toolkit.exceptions import *
from ollama_toolkit.utils.model_constants import get_fallback_model

client = OllamaClient()

def safe_generate(model, prompt, max_attempts=3):
    """Recursively refined error handling approach with multiple fallback levels"""
    attempts = 0
    current_model = model
    
    while attempts < max_attempts:
        try:
            return client.generate(model=current_model, prompt=prompt, stream=False)
        except ModelNotFoundError as e:
            print(f"Model not found: {e}")
            current_model = get_fallback_model(current_model)
            print(f"Trying fallback model: {current_model}")
        except ConnectionError as e:
            print(f"Connection error: {e}")
            print("Attempting to restart connection...")
            time.sleep(1)  # Brief pause before retry
        except TimeoutError as e:
            print(f"Request timed out: {e}")
            if "llama" in current_model or "deepseek" in current_model:
                current_model = "tinyllama"  # Try smaller model
                print(f"Trying lighter model: {current_model}")
            else:
                print("Reducing complexity and trying again...")
                prompt = prompt[:len(prompt)//2]  # Simplify prompt
        except OllamaAPIError as e:
            print(f"API error: {e}")
            return {"error": str(e), "response": "Error occurred during generation"}
        
        attempts += 1
    
    return {"error": "Maximum retry attempts reached", "response": "Failed to generate response"}

Exception Details

ModelNotFoundError

Raised when a requested model cannot be found in the Ollama server.

try:
    client.generate(model="nonexistent-model", prompt="Hello")
except ModelNotFoundError as e:
    print(f"Error: {e}")  # "Error: Model 'nonexistent-model' not found"
    print(f"Available models: {[m['name'] for m in client.list_models().get('models', [])]}")

ConnectionError

Raised when the client cannot connect to the Ollama server.

try:
    client = OllamaClient(base_url="http://incorrect-url:11434")
    client.get_version()
except ConnectionError as e:
    print(f"Connection failed: {e}")
    print("Please ensure the Ollama server is running with: 'ollama serve'")

TimeoutError

Raised when a request takes longer than the specified timeout.

try:
    client = OllamaClient(timeout=1)  # Very short timeout
    client.generate(model="llama2", prompt="Write a novel", stream=False)
except TimeoutError as e:
    print(f"Request timed out: {e}")
    print("Try increasing the timeout or using a smaller model")

StreamingError

Raised when there’s an error during a streaming response.

try:
    for chunk in client.generate(model="llama2", prompt="Hello", stream=True):
        print(chunk.get("response", ""), end="", flush=True)
except StreamingError as e:
    print(f"\nStreaming error: {e}")

Error Logging

Ollama Toolkit provides comprehensive logging for error diagnosis:

import logging
from ollama_toolkit import OllamaClient

# Configure logging
logging.basicConfig(
    level=logging.DEBUG,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
    handlers=[logging.FileHandler("ollama_debug.log"), logging.StreamHandler()]
)

client = OllamaClient()
try:
    client.generate(model="nonexistent-model", prompt="Test")
except Exception as e:
    logging.error(f"Generation failed: {e}", exc_info=True)

Eidosian Error Messages

Following the principle of “Humor as Cognitive Leverage,” error messages are designed to be informative and memorable:

  • When a model isn’t found: “Model ‘nonexistent-model’ not found. Like searching for unicorns—majestic but absent. Try ‘llama2’ instead.”

  • When a connection fails: “Connection refused—like knocking on a door with no one home. Is Ollama running with ‘ollama serve’?”

  • When a request times out: “Time waits for no one, and neither does your request. Consider a smaller model or a larger timeout.”

Best Practices

  1. Always handle specific exceptions before generic ones

  2. Implement fallback mechanisms for critical operations

  3. Use proper timeout values based on model size and task complexity

  4. Log errors with sufficient context for debugging

  5. Provide helpful feedback to users when errors occur

By following these patterns, your applications will achieve a level of robustness and resilience that embodies the Eidosian principle of “Self-Awareness as Foundation.”