Release v0.4.0: Google Gemini Provider Support

What's New in v0.4.0

🚀 Major Feature Addition: Google Gemini Provider Support

Added

• New Provider Support: Google Gemini integration with direct API access
• Support for Gemini models: gemini-1.5-pro, gemini-1.5-flash, gemini-1.0-pro
• Gemini-specific error handling for safety filter blocks
• Environment variable support for both GEMINI_API_KEY and GOOGLE_API_KEY
• Comprehensive Gemini documentation and examples in README
• Test script for validating Gemini provider functionality

Changed

• Updated provider documentation to include Gemini across all examples
• Enhanced supported models section with Gemini model descriptions
• Added google-generativeai dependency to pyproject.toml

Technical Details

• Async support for Gemini API calls
• Proper error handling for content safety blocks
• Token usage tracking for Gemini responses
• Integration with existing caching and retry mechanisms

Full Changelog: v0.3.3...v0.4.0

v0.2.0 - OpenRouter Provider Support

🚀 Major Feature Release: OpenRouter Provider Support

🎯 What's New

OpenRouter Integration

• 100+ AI Models: Access to OpenAI, Anthropic, Meta, Google, Mistral, and more through one unified API
• DeepSeek Support: Full support for deepseek/deepseek-v3.1-base and deepseek/deepseek-chat
• Cost Effective: Often cheaper than going directly to providers
• Unified API: Single integration for multiple model providers

Enhanced Configuration

• Added **kwargs support to LLMConfig for provider-specific parameters
• Maintains full backward compatibility with existing code
• Enhanced error handling and retry logic

📖 Quick Start with OpenRouter

# Install the latest version
pip install llm-batch-helper==0.2.0

# Set up your API key
export OPENROUTER_API_KEY="your-openrouter-api-key"

from llm_batch_helper import LLMConfig, process_prompts_batch

# Use any OpenRouter model
config = LLMConfig(
    model_name="deepseek/deepseek-v3.1-base",  # or openai/gpt-4o, anthropic/claude-3-5-sonnet, etc.
    temperature=0.7,
    max_completion_tokens=500
)

results = await process_prompts_batch(
    config=config,
    provider="openrouter",  # New provider!
    prompts=your_prompts
)

🌟 Supported Models (Examples)

DeepSeek Models

• deepseek/deepseek-v3.1-base - Latest base model
• deepseek/deepseek-chat - Chat-optimized version

OpenAI Models via OpenRouter

• openai/gpt-4o
• openai/gpt-4o-mini
• openai/gpt-4-turbo

Anthropic Models

• anthropic/claude-3-5-sonnet
• anthropic/claude-3-haiku

Meta Models

• meta-llama/llama-3.1-405b-instruct
• meta-llama/llama-3.1-70b-instruct

Google Models

• google/gemini-pro-1.5
• google/gemini-flash-1.5

🔧 Technical Improvements

• Full integration with existing caching and verification systems
• Proper retry logic with exponential backoff
• Token usage tracking and monitoring
• Credit-based rate limiting support
• Enhanced provider comparison documentation

📚 Documentation Updates

• Comprehensive OpenRouter setup guide
• Updated provider comparison table
• Model selection examples and best practices
• Enhanced API reference

🔄 Migration

No breaking changes! Existing code continues to work unchanged. Simply add OpenRouter as a new option when you need access to additional models.

🙏 Get Started

1. Get your OpenRouter API key: https://openrouter.ai/settings/keys
2. Install: pip install llm-batch-helper==0.2.0
3. Follow the documentation: Provider Guide

🚦 Supported Providers

Your package now supports:

• ✅ OpenAI (original)
• ✅ Together.ai (existing)
• ✅ OpenRouter (new!) - 100+ models including DeepSeek v3.1

📊 Performance

OpenRouter integration has been tested and verified with:

• Fast response times (~1.8s per request for DeepSeek)
• Reliable error handling and retry logic
• Proper token usage tracking
• Full compatibility with existing batch processing features

---

Full Changelog: https://github.com/TianyiPeng/LLM_batch_helper/blob/main/CHANGELOG.md

PyPI Package: https://pypi.org/project/llm-batch-helper/0.2.0/

v0.1.6 - Environment Variable Handling Improvements

🔧 Breaking Changes

⚠️ BREAKING CHANGE: Removed automatic load_dotenv() call from providers.py

✨ Improvements

Better Environment Variable Handling

• Removed Side Effects: No more automatic .env file loading on import
• User Control: Users now decide when/how to load environment variables
• Smaller Dependencies: python-dotenv is now optional (moved to dev dependencies)
• Library Best Practices: Follows Python packaging guidelines

Updated Documentation

• Clear instructions for both environment variable approaches
• Examples showing proper .env file usage
• Better code examples throughout

📦 Migration Guide

Before (v0.1.5)

# .env file was automatically loaded
from llm_batch_helper import LLMConfig, process_prompts_batch

After (v0.1.6)

# Option 1: Use system environment variables
export OPENAI_API_KEY="your-key"

# Option 2: Load .env file yourself (recommended for development)
from dotenv import load_dotenv
load_dotenv()  # You control when this happens

from llm_batch_helper import LLMConfig, process_prompts_batch

💡 Benefits

• No Import Side Effects: Cleaner library behavior
• Better Testing: Easier to test with controlled environments
• Reduced Dependencies: Smaller package footprint
• More Flexibility: Users control environment loading

📦 Installation

pip install llm_batch_helper==0.1.6

# Optional: Install dotenv support for development
pip install python-dotenv

Full Changelog: v0.1.5...v0.1.6

v0.1.5 - Together.ai Provider Support

🚀 New Features

• Together.ai Provider Support: Added support for Together.ai API as a new provider
• Support for various open-source models through Together.ai (Llama, Mixtral, etc.)
• Comprehensive documentation with Read the Docs integration

📝 Changes

• Added _get_together_response_direct() function for Together.ai API calls
• Updated provider selection logic to support multiple providers
• Enhanced error handling for Together.ai specific errors
• Updated docstrings and examples to reflect new provider options

📖 Documentation

• Complete Sphinx documentation structure
• Read the Docs configuration
• API reference, examples, and tutorials
• Provider comparison guide

🔧 Usage

# New Together.ai provider usage
config = LLMConfig(
    model_name="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
    temperature=0.7,
    max_completion_tokens=300
)

results = await process_prompts_batch(
    config=config,
    provider="together",  # New provider option
    prompts=your_prompts
)

📦 Installation

pip install llm_batch_helper==0.1.5

Full Changelog: v0.1.4...v0.1.5

Release v0.1.4

What's Changed

• Added backward compatibility for max_completion_tokens parameter
• Relaxed httpx dependency constraint to >=0.24.0,<2.0.0 for better compatibility
• Maintained support for legacy max_tokens parameter

Technical Changes

• Updated OpenAI API calls to use max_completion_tokens instead of deprecated max_tokens
• Added automatic fallback when only max_tokens is provided
• Expanded httpx version range to support more environments

Full Changelog: v0.1.3...v0.1.4