Repository navigation

Add the IQ2_XXS weight-only quantization format - #2511

Merged

cjluo-nv merged 4 commits into

mainfrom

chenjiel/iq2-xxs-format

Sep 23, 2026

Merged

Add the IQ2_XXS weight-only quantization format#2511
cjluo-nv merged 4 commits into
mainfrom
chenjiel/iq2-xxs-format

Commits on Sep 22, 2026
- Add the IQ2_XXS weight-only quantization format
  
  cjluo-nv
  and
  claude
  committed
  a3052f9
- Carry IQ block metadata through HF config conversion for every format
  
  cjluo-nv
  and
  claude
  committed
  f6f658c
- Cover the mixed IQ export path and correct the changelog
  
  cjluo-nv
  and
  claude
  committed
  37ae39c
Commits on Sep 23, 2026
- Run the Megatron IQ export tests for every format
  
  cjluo-nv
  and
  claude
  committed
  cecb5d6