Eval bug: Vulkan on Intel ARC – Gemma 4 models (especially with mtp) produce unusable, garbled output

用户在 Docker 中运行 llama.cpp server (vulkan 后端) 加载 Gemma 4 系列模型(Gemma4-12B qat+mtp 和 Gemma4-26B-A4B),使用 Intel ARC B580 显卡。输出包含乱码(符号、不同语种混杂)。同配置下 CPU 推理正常,
![[Bug]: Image Recognition Regression in Chat Assistant after Authorization Fix](https://www.chat-gpts.plus/wp-content/uploads/2026/06/6822-f34e0198-768x403.jpg)
![[Bug] /v1/internal/model/load silently ignores `args` (loader flags like ctx-size, cache-type) — endpoint returns OK but model loads with UI](https://www.chat-gpts.plus/wp-content/uploads/2026/06/7577-f376545a-768x403.jpg)
![[SYCL][DOCKER] Increased Gemma 4 26b performance with updated docker dependencies](https://www.chat-gpts.plus/wp-content/uploads/2026/06/24045-65cf0f68-768x403.jpg)




![[Question]: Failed to build knowledge graph](https://www.chat-gpts.plus/wp-content/uploads/2026/06/6886-32ff5dfd-768x403.jpg)
