[Kernel][Core] Add AWQ support to the Marlin kernel (#6612)

This commit is contained in:
Alexander Matveev
2024-07-21 19:41:42 -04:00
committed by GitHub
parent 25e778aa16
commit 396d92d5e0
21 changed files with 1594 additions and 276 deletions
+4
View File
@@ -141,6 +141,10 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
ops.def("gptq_marlin_repack", &gptq_marlin_repack);
ops.impl("gptq_marlin_repack", torch::kCUDA, &gptq_marlin_repack);
// awq_marlin repack from AWQ.
ops.def("awq_marlin_repack", &awq_marlin_repack);
ops.impl("awq_marlin_repack", torch::kCUDA, &awq_marlin_repack);
// fp8_marlin Optimized Quantized GEMM for FP8 weight-only.
ops.def("fp8_marlin_gemm", &fp8_marlin_gemm);
ops.impl("fp8_marlin_gemm", torch::kCUDA, &fp8_marlin_gemm);