Medusa

Medusa decoding module for faster LLM inference.

Status

This module is currently under development. Check back for updates.

Planned Features

• Medusa head implementation
• Speculative decoding
• Multi-token prediction