project

XPress

A speculative decoding component that appends a small causal refiner, around 80M parameters, after a parallel block drafter to correct sequences that are locally probable but jointly inconsistent.

Current clusters

build1 publisher

Block diffusion drafting collapses EAGLE-3's eleven serial forward passes into one

A September 3 playbook traces speculative decoding's draft architectures from EAGLE-3 to DFlash, and grounds the case in a 70B model that decodes at 15 to 20 tokens a second on eight H100s.

Publishers:dev.to

Reality

Evidence24
Adoption31
Hype gap+38
Incentives38
Confidence33