biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Add GPTQ Marlin 2:4 sparse structured support (#4790)

Browse Source 
Co-authored-by: Robert Shaw <rshaw@neuralmagic.com>
This commit is contained in:
Alexander Matveev
2024-05-16 12:56:15 -04:00
committed by GitHub
parent 9216b9cc38
commit 6979ade384
18 changed files with 2130 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
csrc/pybind.cpp
View File

@@ -66,7 +66,8 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
ops.def("aqlm_gemm", &aqlm_gemm, "Quantized GEMM for AQLM");
ops.def("aqlm_dequant", &aqlm_dequant, "Decompression method for AQLM");
ops.def("awq_gemm", &awq_gemm, "Quantized GEMM for AWQ");
ops.def("marlin_gemm", &marlin_gemm, "Marlin Optimized Quantized GEMM for GPTQ");
ops.def("marlin_gemm", &marlin_gemm, "Marlin (Dense) Optimized Quantized GEMM for GPTQ");
ops.def("gptq_marlin_24_gemm", &gptq_marlin_24_gemm, "Marlin_24 (Sparse) Optimized Quantized GEMM for GPTQ");
ops.def("gptq_marlin_gemm", &gptq_marlin_gemm, "gptq_marlin Optimized Quantized GEMM for GPTQ");
ops.def("gptq_marlin_repack", &gptq_marlin_repack, "gptq_marlin repack from GPTQ");
ops.def("awq_dequantize", &awq_dequantize, "Dequantization for AWQ");