File tree Expand file tree Collapse file tree
onnxruntime/contrib_ops/cpu/quantization Expand file tree Collapse file tree Original file line number Diff line number Diff line change @@ -104,7 +104,8 @@ class MatMulNBits final : public OpKernel {
104104 nbits_{narrow<size_t >(info.GetAttr <int64_t >(" bits" ))},
105105 has_g_idx_{info.GetInputCount () > InputIndex::g_idx && info.node ().InputDefs ()[InputIndex::g_idx]->Exists ()},
106106 has_bias_{info.GetInputCount () > InputIndex::bias && info.node ().InputDefs ()[InputIndex::bias]->Exists ()},
107- prefer_lut_gemm_{info.GetConfigOptions ().GetConfigEntry (kOrtSessionOptionsMlasLutGemm ) == " 1" &&
107+ prefer_lut_gemm_{std::is_same_v<T1, float > &&
108+ info.GetConfigOptions ().GetConfigEntry (kOrtSessionOptionsMlasLutGemm ) == " 1" &&
108109 MlasIsLutGemmAvailable (narrow<size_t >(info.GetAttr <int64_t >(" N" )),
109110 narrow<size_t >(info.GetAttr <int64_t >(" K" )),
110111 narrow<size_t >(info.GetAttr <int64_t >(" bits" )),
You can’t perform that action at this time.
0 commit comments