Repository navigation

ggml-cuda: Q8_0 weights take dequant-to-BF16 + cuBLAS by default; k-quants stay on MMQ - #39

Open

danielhanchen wants to merge 4 commits into

cuda-q8-bf16-cublas

Open

ggml-cuda: Q8_0 weights take dequant-to-BF16 + cuBLAS by default; k-quants stay on MMQ#39
danielhanchen wants to merge 4 commits into
masterfrom
cuda-q8-bf16-cublas

Commits on Oct 22, 2026
- Merge upstream master (a77d362) into the fork
  danielhanchen
  committed
  f3b773a
- Merge upstream master (fd21ecf) into the fork
  danielhanchen
  committed
  344a3b6
Commits on Oct 23, 2026
- ggml-patches/0016: BF16 cuBLAS for large Q8_0 matmuls by default on A100, Hopper and datacenter Blackwell
  danielhanchen
  committed
  ce2bbc3
Commits on Nov 7, 2026
- Merge master into cuda-q8-bf16-cublas
  danielhanchen
  committed
  ef55f7d