apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[XPU] support moe_expert_ffn TGEMM selection (#5375)

Browse Source
This commit is contained in:
Lucas
2025-12-05 17:49:40 +08:00
committed by GitHub
parent 3aed8d257d
commit 8f2b85362d
1 changed files with 13 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
custom_ops/xpu_ops/src/ops/moe_expert_ffn.cc
View File

@@ -393,7 +393,19 @@ std::vector<paddle::Tensor> MoeExpertFFNKernel(
smooth_data, \
hadamard_blocksize)
if (quant_method == "weight_only_int8") {
FFN_IMPL(XPU_TX1, XPU_TX2, int8_t, float);
static const char* xft_moe_fc_wint8_tgemm =
std::getenv("XFT_MOE_FC_WINT8_TGEMM");
if (xft_moe_fc_wint8_tgemm != nullptr) {
if (std::string(xft_moe_fc_wint8_tgemm) == "INT8") {
FFN_IMPL(XPU_TX1, XPU_TX2, int8_t, int8_wo_t);
} else if (std::string(xft_moe_fc_wint8_tgemm) == "FLOAT16") {
FFN_IMPL(XPU_TX1, XPU_TX2, int8_t, float16);
} else {
FFN_IMPL(XPU_TX1, XPU_TX2, int8_t, float);
}
} else {
FFN_IMPL(XPU_TX1, XPU_TX2, int8_t, float);
}
} else if (quant_method == "weight_only_int4") {
FFN_IMPL(XPU_TX1, XPU_TX2, int4_t, int4_wo_int15);
} else if (quant_method == "w4a8") {