Repository navigation

[FEAT] Support random sampling in async block speculation - #187

Draft

ycjcx123 wants to merge 10 commits into

InferMatrixLab:mainfrom

ycjcx123:feat/m7-async-spec-sampling

Draft

[FEAT] Support random sampling in async block speculation#187
ycjcx123 wants to merge 10 commits into
InferMatrixLab:mainfrom
ycjcx123:feat/m7-async-spec-sampling

Commits on Nov 5, 2026
- feat: add batched sampling distributions
  ycjcx123
  committed
  0f7a3da
- test: cover batched sampling distributions
  ycjcx123
  committed
  5caddce
- feat: add device-side speculative rejection sampling
  ycjcx123
  committed
  355332a
- test: check speculative rejection sampling against target distributions
  ycjcx123
  committed
  1c2d0dd
- feat: integrate random sampling into async speculative rounds
  ycjcx123
  committed
  2706c8a
- test: cover random round planning and result decoding
  ycjcx123
  committed
  ce89130
- feat: allow random sampling in async block speculation
  ycjcx123
  committed
  06fd392
- test: cover async random speculation on CUDA
  ycjcx123
  committed
  a9c8f6e
- test: cover random async speculation through the public entrypoint
  ycjcx123
  committed
  bd1a937
Commits on Nov 9, 2026
- docs: describe async speculative random sampling
  ycjcx123
  committed
  1c4d764