spec-stream 0.1.0

High-performance, model-agnostic speculative decoding and paged KV-cache engine for accelerated LLM inference
Documentation