← Lcoalhost

vLLM Speculative Decoding: 1.8x Faster Alone, Slower at 32 Users

Source : DEV · #llm

See it live in context on Lcoalhost →