DFlash: Block Diffusion for Flash Speculative Decoding
Pulse
0
Velocity
Stars
4,320
Diffusion model accelerates speculative decoding for large language models.
Block diffusion model for flash speculative decoding to improve LLM inference speed.
Accelerates LLM inference up to 2.98x using block diffusion.
Why Trending
Novel block diffusion technique offers faster LLM inference.
Target Audience
LLM researchers and developers optimizing inference.
Similar Projects