CUDA: only support F32/F16 for GGML_OP_REPEAT (#24533)

2026-06-15 09:11:00 +03:00
parent 6e9007ae61
commit 3686e9d643
2 changed files with 3 additions and 1 deletions
@@ -5337,8 +5337,9 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
            } break;
        case GGML_OP_REPEAT:
            {
+                // the CUDA REPEAT path only implements F32/F16; other types assert at runtime
                ggml_type src0_type = op->src[0]->type;
-                return src0_type != GGML_TYPE_I32 && src0_type != GGML_TYPE_I16;
+                return src0_type == GGML_TYPE_F32 || src0_type == GGML_TYPE_F16;
            } break;
        case GGML_OP_REPEAT_BACK:
                return op->type == GGML_TYPE_F32 && (op->src[0]->ne[2]*op->src[0]->ne[3]) <= (1 << 15);