1

Signed-off-by: Xiaodong Ye <[email protected]>

Author: yeahdongcn

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -3030,8 +3030,8 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
                 const int cc = ggml_cuda_info().devices[dev_ctx->device].cc;
                 if (GGML_CUDA_CC_IS_MTHREADS(cc) && b->ne[2]*b->ne[3] > 1 &&
                     !ggml_is_transposed(a) && !ggml_is_transposed(b)) {
-                    if (GGML_CUDA_CC_IS_QY1(cc) && op->op == GGML_OP_MUL_MAT
-                        && b->type == GGML_TYPE_F16) {
+                    if (GGML_CUDA_CC_IS_QY1(cc) && op->op == GGML_OP_MUL_MAT &&
+                        a->type == GGML_TYPE_F16 && b->type == GGML_TYPE_F16) {
                         return false;
                     }
                     if (GGML_CUDA_CC_IS_QY2(cc) && op->op == GGML_OP_MUL_MAT_ID &&