Init Release: v0.0.1

Release Notes

v0.0.1 (2024-01-26)

Initial release of llm-rewards.

Features

• Modular reward function architecture
• Framework-agnostic design
• Built-in reward functions:
  • Pretrained reward models
  • Length-based rewards
  • XML/format validation
  • Text similarity metrics
  • Prompt relevance scoring
• Easy reward stacking and composition

Installation

pip install llm-rewards

Coming Soon

• Multi-GPU support
• More example reward functions
• Explicit RL framework integrations
• Async reward computation
• Performance optimizations

For detailed documentation, visit our GitHub repository.