Misc. bug: [ROCm] Significantly lower token generation performance vs Vulkan on RX 7900 XTX (gfx1100)
![Misc. bug: [ROCm] Significantly lower token generation performance vs Vulkan on RX 7900 XTX (gfx1100)](https://www.chat-gpts.plus/wp-content/uploads/2026/06/20934-97b8a216-768x403.jpg)
在 Linux 系统(Ubuntu 24.04.3/24.04.4)上使用 llama.cpp(版本 2999,2026年3月多個近期 commit)运行 GGUF 格式模型(LLaMA 7B Q4_0、Qwen2 7B Q4_K_M 等)进行 prompt 处理和 token 生成基准测试时,发现






![[Feature Request]: Restart Failed Subtasks Instead of Restarting Entire Process](https://www.chat-gpts.plus/wp-content/uploads/2026/06/8682-f9e725d4-768x403.jpg)
![[Bug]: /team/member_update does not support setting budget_duration on individual member budgets](https://www.chat-gpts.plus/wp-content/uploads/2026/06/25509-cb8f8d8a-768x403.jpg)
