llama.cpp
Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support
#16900
Merged
Overview
Commits
15
Changes
View On GitHub
Commits
  • vulkan: split mul_mmq_funcs for mul_mat_vecq use
    0cc4m committed 118 days ago
  • add mxfp4 mmvq
    0cc4m committed 118 days ago
  • add q2_k mmvq
    0cc4m committed 118 days ago
  • add q3_k mmvq
    0cc4m committed 118 days ago
  • add q4_k and q5_k mmvq
    0cc4m committed 118 days ago
  • add q6_k mmvq
    0cc4m committed 118 days ago
  • handle 4x4 quants per mmvq thread
    0cc4m committed 118 days ago
  • enable MUL_MAT_ID mmvq support
    0cc4m committed 118 days ago
  • enable subgroup optimizations for mul_mat_vec_id shaders
    0cc4m committed 118 days ago
  • device tuning
    0cc4m committed 118 days ago
  • request prealloc_y sync after quantization
    0cc4m committed 118 days ago
  • fix indentation
    0cc4m committed 118 days ago
  • fix llvmpipe test failures
    0cc4m committed 118 days ago
  • fix mul_mat_id mmvq condition
    0cc4m committed 118 days ago
  • fix unused variable warning
    0cc4m committed 115 days ago
FAQ Terms Privacy Refunds Impressum
Loading