llama.cpp
ggml-cuda: fallback to F32 on device without BF16 hardware acceleration
#28846
Merged

Commits
  • ggml-cuda: fallback to F32 on device without BF16 hardware acceleration: (Nvidia >= AMPERE, AMD >= RDNA3 or = CDNA)
    thelittlefireman committed 66 days ago
  • apply logic to NVIDIA as well
    JohannesGaessler committed 65 days ago
Loading