diff --git a/ggml/src/ggml-cuda/cpy.cu b/ggml/src/ggml-cuda/cpy.cu
index fd7ffc0bc..7a9984585 100644
--- a/ggml/src/ggml-cuda/cpy.cu
+++ b/ggml/src/ggml-cuda/cpy.cu
@@ -589,6 +589,14 @@ void ggml_cuda_cpy(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, gg
             ggml_cpy_scalar_cuda<int32_t, int32_t>
                 (src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
         }
+    } else if (src0->type == GGML_TYPE_I16 && src1->type == GGML_TYPE_I16) {
+        if (can_be_transposed) {
+            ggml_cpy_scalar_cuda<int16_t, int16_t, true>
+                (src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+        } else {
+            ggml_cpy_scalar_cuda<int16_t, int16_t>
+                (src0_ddc, src1_ddc, ne, ne00, ne01, ne02, nb00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb13, main_stream);
+        }
     } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_I32) {
         if (contiguous_srcs) {
             ggml_cpy_scalar_contiguous_cuda<float, int32_t>
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 790553888..43003245c 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -5298,10 +5298,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
                 return false;
             } break;
         case GGML_OP_DUP:
-            {
-                ggml_type src0_type = op->src[0]->type;
-                return src0_type != GGML_TYPE_I32 && src0_type != GGML_TYPE_I16;
-            } break;
+                return true;
         case GGML_OP_ARGMAX:
         case GGML_OP_COUNT_EQUAL:
             {