Repository navigation

Add BF16 support for int8_vectorwise_quant / LLM.int8 activation quant - #1985

Open

kru2710shna wants to merge 6 commits into

bitsandbytes-foundation:mainfrom

kru2710shna:feature/int8-bf16-vectorwise-quant

Open

Add BF16 support for int8_vectorwise_quant / LLM.int8 activation quant#1985
kru2710shna wants to merge 6 commits into
bitsandbytes-foundation:mainfrom
kru2710shna:feature/int8-bf16-vectorwise-quant

Commits on Jul 7, 2026
Commits on Jul 10, 2026
- Add native bf16 output support to int8_mm_dequant
  kru2710shna
  committed
  786d3f3
Commits on Nov 2, 2026
- Merge branch 'main' into feature/int8-bf16-vectorwise-quant
  kru2710shna
  authored
  f904f72
- Fix CUDA<12.4/ROCm builds, restore fp32 support, rename dequant_mm_int32
  
  kru2710shna
  authored and
  kru2710shna
  committed
  42b08e1