Back to feed

b7538

Dec 25, 2025
Meta/llama.cppCLIvb7538

ggml-cuda: fix blackwell native builds (#18361)

  • ggml-cuda: fix blackwell native builds

Replace 12x in native architectures by 12xa

  • replace for GGML_NATIVE=OFF too

  • only replace for native

  • remove 120f-virtual for default compilation


Co-authored-by: Aman Gupta

macOS/iOS:

Linux:

Windows:

openEuler: