Repository navigation

[2/3] Speed up layerwise GPTQ on MoE models - #2864

Merged

cjluo-nv merged 5 commits into

mainfrom

chenjiel/gptq-speedup

Oct 23, 2026

Merged

[2/3] Speed up layerwise GPTQ on MoE models#2864
cjluo-nv merged 5 commits into
mainfrom
chenjiel/gptq-speedup

Commits on Oct 23, 2026
- Speed up GPTQ calibration on MoE models
  
  cjluo-nv
  and
  claude
  authored
  0101143
- Address review: Hessian GPU capacity check and layerwise batch tests
  
  cjluo-nv
  and
  claude
  authored
  a1ff4c7
- Skip per-expert GPTQ error logs
  
  cjluo-nv
  and
  claude
  authored
  bf99bc3
- Reject a non-positive --calib_seq in hf_ptq
  
  cjluo-nv
  and
  claude
  authored
  6cd5d46
- Test hf_ptq's layerwise calibration batch at the dataloader
  
  cjluo-nv
  and
  claude
  authored
  5487343