Repository navigation

[Feature][Quantization] Support compressed-tensors FP8 Marlin cache restarts - #64738

Draft

liusy58 wants to merge 11 commits into

vllm-project:mainfrom

liusy58:fast-start-compressed-marlin-fp8

Draft

[Feature][Quantization] Support compressed-tensors FP8 Marlin cache restarts#64738
liusy58 wants to merge 11 commits into
vllm-project:mainfrom
liusy58:fast-start-compressed-marlin-fp8

Commits on Oct 25, 2026
- Support online FP8 and CUTLASS NVFP4 weight-cache restarts
  
  liusy58
  and
  codex
  committed
  6183e75
Commits on Oct 26, 2026
- Merge main and preserve shared daemon test logging
  
  liusy58
  and
  codex
  committed
  084cab1
- Support compressed-tensors CUTLASS FP8 weight-cache restarts
  
  liusy58
  and
  codex
  committed
  f3f4f35
- Merge main with daemon lifetime and kernel cache-key fixes
  
  liusy58
  and
  codex
  committed
  d771fd8
- Merge validated online quantization branch and latest main
  
  liusy58
  and
  codex
  committed
  ae2eac2
Commits on Oct 29, 2026
- Merge upstream main for quantization parameter retention
  
  liusy58
  and
  codex
  committed
  aa14188
- Merge upstream parameter retention into compressed FP8 cache support
  
  liusy58
  and
  codex
  committed
  86c1741
- Check retained parameter objects and shared storage separately
  
  liusy58
  and
  codex
  committed
  6fb5a58
- Merge upstream quantization fixes
  
  liusy58
  and
  codex
  committed
  3571116
- Merge updated prerequisite for #63672
  
  liusy58
  and
  codex
  committed
  03b39f9
Commits on Oct 30, 2026
- [Feature][Quantization] Support compressed-tensors FP8 Marlin cache restarts
  
  liusy58
  and
  codex
  committed
  babf3e8