This adds nvfp4 support for get_rows, dequant, and mul_mat(_id). For mul_mat, it does not add support for the dp4/q8_1 path, it's all via fp16/fp32. |
||
|---|---|---|
| .. | ||
| cmake | ||
| vulkan-shaders | ||
| CMakeLists.txt | ||
| ggml-vulkan.cpp | ||
This adds nvfp4 support for get_rows, dequant, and mul_mat(_id). For mul_mat, it does not add support for the dp4/q8_1 path, it's all via fp16/fp32. |
||
|---|---|---|
| .. | ||
| cmake | ||
| vulkan-shaders | ||
| CMakeLists.txt | ||
| ggml-vulkan.cpp | ||