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

[CPU] Fix torch version in x86 CPU backend (#19258)

Browse Source 
Signed-off-by: jiang1.li <jiang1.li@intel.com>
This commit is contained in:
Li, Jiang
2025-06-26 18:34:47 +08:00
committed by GitHub
parent d188913d99
commit 0567c8249f
8 changed files with 52 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
csrc/cpu/torch_bindings.cpp
View File

@ -131,16 +131,19 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
// Quantization
#ifdef __AVX512F__
at::Tag stride_tag = at::Tag::needs_fixed_stride_order;
// Compute int8 quantized tensor for given scaling factor.
ops.def(
"static_scaled_int8_quant(Tensor! out, Tensor input, Tensor scale,"
"Tensor? azp) -> ()");
"Tensor? azp) -> ()",
{stride_tag});
ops.impl("static_scaled_int8_quant", torch::kCPU, &static_scaled_int8_quant);
// Compute int8 quantized tensor and scaling factor
ops.def(
"dynamic_scaled_int8_quant(Tensor! out, Tensor input, Tensor! scale, "
"Tensor!? azp) -> ()");
"Tensor!? azp) -> ()",
{stride_tag});
ops.impl("dynamic_scaled_int8_quant", torch::kCPU,
&dynamic_scaled_int8_quant);
// W8A8 GEMM, supporting symmetric per-tensor or per-row/column
@ -148,7 +151,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
ops.def(
"cutlass_scaled_mm(Tensor! out, Tensor a,"
" Tensor b, Tensor a_scales,"
" Tensor b_scales, Tensor? bias) -> ()");
" Tensor b_scales, Tensor? bias) -> ()",
{stride_tag});
ops.impl("cutlass_scaled_mm", torch::kCPU, &int8_scaled_mm);
// w8a8 GEMM, supporting asymmetric per-tensor or per-row/column
// quantization.
@ -156,7 +160,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
"cutlass_scaled_mm_azp(Tensor! out, Tensor a,"
" Tensor b, Tensor a_scales,"
" Tensor b_scales, Tensor azp_adj,"
" Tensor? azp, Tensor? bias) -> ()");
" Tensor? azp, Tensor? bias) -> ()",
{stride_tag});
ops.impl("cutlass_scaled_mm_azp", torch::kCPU, &int8_scaled_mm_azp);
#elif defined(__powerpc64__)
// Compute int8 quantized tensor for given scaling factor.