[Kernel] Add more dtype support for GGUF dequantization (#15879)

Signed-off-by: lukas.bluebaum <lukas.bluebaum@aleph-alpha.com>
This commit is contained in:
LukasBluebaum
2025-04-02 01:58:48 -07:00
committed by GitHub
parent 101f1481f9
commit 90969fb39a
9 changed files with 80 additions and 50 deletions
+3 -1
View File
@@ -295,7 +295,9 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
#endif
// Dequantization for GGML.
ops.def("ggml_dequantize(Tensor W, int type, SymInt m, SymInt n) -> Tensor");
ops.def(
"ggml_dequantize(Tensor W, int type, SymInt m, SymInt n, ScalarType? "
"dtype) -> Tensor");
ops.impl("ggml_dequantize", torch::kCUDA, &ggml_dequantize);
// mmvq kernel for GGML.