llama.cpp
ggml-cuda: Add generic NVFP4 MMQ kernel
#21074
Merged

Commits
  • Introduced NVFP4 generic MMQ kernel
    michaelw9999 committed 121 days ago
  • Added extra FP8 guard, hope to solve ci HIP failure
    michaelw9999 committed 121 days ago
  • Rename tiles and use HIP_FP8_AVAILABLE
    michaelw9999 committed 121 days ago
  • Removed remaning FP8 straggler and added const int
    michaelw9999 committed 121 days ago
  • Const
    michaelw9999 committed 121 days ago
  • Removed DECL_MMQ_CASE artifact
    michaelw9999 committed 121 days ago
  • Removed newline
    michaelw9999 committed 121 days ago
  • Removed space after else
    michaelw9999 committed 121 days ago
  • Changed HIP FP8 NVFP4 conversion gate
    michaelw9999 committed 120 days ago
  • Added new line to bottom of mmq.cu 270
    michaelw9999 committed 120 days ago
  • Removed extra spaces
    michaelw9999 committed 120 days ago
  • Removed single space in front of else on line 814
    michaelw9999 committed 120 days ago
  • Added NVFP4 to generate cu script so HIP can see it, further tightened logic
    michaelw9999 committed 120 days ago
  • Include generated mmq-instance-nvfp4.cu
    michaelw9999 committed 120 days ago
  • Added NVFP4 mmq to HIP Check ignore list
    michaelw9999 committed 120 days ago
  • Update ggml/src/ggml-cuda/mmq.cuh
    michaelw9999 committed 119 days ago
  • Update ggml/src/ggml-cuda/mmq.cuh
    michaelw9999 committed 119 days ago
  • Update ggml/src/ggml-cuda/mmq.cuh
    michaelw9999 committed 119 days ago
  • Added function names to closing endif
    michaelw9999 committed 119 days ago
Loading