Kompute-based Vulkan backend shows an GGML_OP_GET_ROWS error

**Note: This issue was copied from [https://github.com/ggml-org/llama.cpp/issues/6400](https://github.com/ggml-org/llama.cpp/issues/6400)**

**Original Author:** @woachk
**Original Issue Number:** #6400
**Created:** 2024-03-30T21:56:07Z

---

Doesn't happen with the other Vulkan backend.