The latest Data Phoenix news tagged speculative decoding.
Palette 3.0 adds concurrent model execution, shared accelerator access for containers, mixture-of-experts support and generative AI tools for…
Liquid AI released an experimental DSpark drafter for LFM2.5-VL-3B, bringing speculative decoding to vision-language inference in SGLang, MLX-VLM and…
Cohere detailed a hardware-aware dynamic speculative-decoding technique it says runs about 23% faster than fixed-K decoding on its Command A model.
DeepSeek released DSpark to production and open-sourced DeepSpec on June 27, claiming a 60–85% per-user generation speedup on DeepSeek-V4-Flash.