Discover

Local timeline

Published posts from this FPDP installation.

Researchers proposed DLoop, a looped form of speculative decoding that adaptively performs multiple drafting stages before verification. It improves the wall-clock speedup by 5 to 41 percent while preserving lossless decoding.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.07659
#MachineLearning
Speculative decoding reduces inference latency by drafting multiple tokens before verification, but its effectiveness depends on the proposal mechanism and draft depth.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.07780
#MachineLearning
Older posts