Repository navigation

[1/2] Reload versioned quantizer recipes in vLLM fake-quant serving - #2888

Merged

kevalmorabia97 merged 5 commits into

mainfrom

kmorabia/vllm-fq-recipe-v1

Oct 22, 2026

Merged

[1/2] Reload versioned quantizer recipes in vLLM fake-quant serving#2888
kevalmorabia97 merged 5 commits into
mainfrom
kmorabia/vllm-fq-recipe-v1

Commits on Oct 21, 2026
- Reload versioned quantizer recipes in vLLM fake-quant serving
  
  kevalmorabia97
  and
  claude
  committed
  d297af2
- Keep reloaded quantizer ranges on modules vLLM registers twice
  
  kevalmorabia97
  and
  claude
  committed
  e8e648a
- Load an aliased quantizer's range under whichever name the checkpoint uses
  
  kevalmorabia97
  and
  claude
  committed
  01f9355
- Write the versioned quantizer recipe from the Megatron-LM fake-quant export too
  
  kevalmorabia97
  and
  claude
  committed
  5b0c000
Commits on Oct 22, 2026
- Write no sidecars from the Megatron-LM fake-quant export when nothing is quantized
  
  kevalmorabia97
  and
  claude
  committed
  7334f1b