Major Bug Fixes and Notebook Improvements

[allanps]

Major Bug Fixes and Notebook Improvements

Overview

This PR resolves critical issues in the LLMomics notebook and core functionality, making the library fully functional for end-users.

Issues Fixed

1. OpenAI API Authentication Issues

• Problem: Error code: 401 - Incorrect API key provided due to quoted API keys
• Solution: Fixed environment variable handling and API key validation
• Impact: Users can now authenticate successfully with OpenAI API

2. JSON Mode Compatibility Issues

• Problem: Error code: 400 - 'response_format' of type 'json_object' is not supported with this model
• Solution: Added model compatibility detection and fallback JSON parsing
• Impact: Works with all OpenAI models (GPT-4, GPT-3.5-turbo, etc.)

3. Pipeline Template Formatting Errors

• Problem: IndexError: Replacement index 0 out of range for positional args tuple
• Solution: Fixed empty {} placeholders in Snakemake templates by proper escaping
• Impact: Pipeline generation now works correctly

4. End-to-End Function Parameter Mismatch

• Problem: TypeError: generate() got an unexpected keyword argument 'dataset_ids'
• Solution: Corrected parameter mapping between ParsedQuery and PipelineGenerator
• Impact: Complete workflow now functions end-to-end

Files Changed

Core Functionality

• src/llmomics/core/llm_provider.py - Enhanced JSON mode compatibility and error handling
• src/llmomics/core/config.py - Fixed environment variable handling
• src/llmomics/pipeline/templates.py - Fixed template placeholder escaping

Documentation & Examples

• examples/getting_started.ipynb - Updated with corrected functions and working examples

Testing

All notebook sessions now work correctly:

• Basic LLM interaction - API authentication and JSON parsing
• Query parsing - Natural language processing
• Pipeline generation - Template rendering and file creation
• Complete end-to-end - Full workflow from query to pipeline

Impact

Before

• Notebook sessions failed with authentication, formatting, and parameter errors
• Users couldn't complete basic workflows
• Multiple TypeError and IndexError exceptions

After

• All notebook sessions work seamlessly
• Complete end-to-end pipeline generation functional
• Robust error handling and fallback mechanisms
• Compatible with all supported OpenAI models

🔧 Technical Details

API Key Handling

# Before: API key wrapped in quotes causing 401 errors
OPENAI_API_KEY = "sk-proj-..."

# After: Proper environment variable handling
api_key = os.getenv("OPENAI_API_KEY", "").strip().strip('"')

JSON Mode Compatibility

# Added model detection and fallback
json_mode_models = ["gpt-4-turbo", "gpt-3.5-turbo", ...]
use_json_mode = self.model in json_mode_models

if use_json_mode:
    api_params["response_format"] = {"type": "json_object"}

Template Escaping

# Before: Empty {} caused IndexError
THREADS = config.get("threads", {})

# After: Proper escaping for Snakemake
THREADS = config.get("threads", {{}})

Ready for Production

This PR makes LLMomics fully functional for:

• Bioinformatics researchers using the notebook interface
• Developers integrating LLM-powered pipeline generation
• CI/CD pipelines requiring stable API interactions

All critical user-facing issues have been resolved with comprehensive testing.