llama.cpp
CUDA: Improve NVFP4 W4A4 activation quantization
#25730

Merged

Commits

Squash history before conflict-resolution during rebase on master

ORippler committed 48 days ago
compiler massaging to avoid unnecessary LDCs

ORippler committed 47 days ago
kvalues_mxfp4 -> kvalues_nvfp4 in quantize_mmq_nvfp4

ORippler committed 42 days ago
Always pass in src1_scale.ptr

ORippler committed 42 days ago
Extract ggml_cuda_is_aligned helper

ORippler committed 42 days ago