llama.cpp b11222+ on Windows: llama-cli crashes right after "Loading model" (event log 0xc0000094)
质量检查: WARNING · 2 项提醒
- WARNING C: placeholders used but not declared in variables
- WARNING H: no rollback (acceptable for read-only fixes, but state it)
静态检查通过不代表修复有效；服务端从不执行提交的命令。
💡 When a crash starts at a specific build, bisect by release and check the tracker before debugging your setup.
发生场景
Source(s): https://github.com/ggml-org/llama.cpp/issues/29551 , https://github.com/ggml-org/llama.cpp/pull/29632 Status: source reports fixed; not reproduced by us.
问题
From build 11222 on, llama-cli exits during model load on Windows with any model. Build 11221 works. Oddly, adding -dev <device> or -lv 4 (or higher) lets it run.
报错原文（已脱敏/规范化）
llama-cli exits right after "Loading model"; Windows event log exception code 0xc0000094
适用环境
llama.cpp builds b11222-b11243 on Windows; official CUDA, Vulkan and ROCm builds and self-built ones; any model.
如何确认是这个问题
Crash appears on b11222+ but not on b11221, and goes away with -dev <device> or -lv 4.
不是这个问题时 / 边界
The crash also happens on b11221 or earlier - a different bug.
根因
A regression in b11222, fixed by PR #29632.
占位变量
<model>: path to a GGUF model. <your device>: device name from llama.cpp's device list.
修复步骤
- Upgrade to release b11249 or newer (the reporter confirmed "Fixed!" on that build).
- Until you can upgrade, stopgaps from the thread: pass
-dev <your device>or-lv 4, or stay on b11221.
如何验证
llama-cli -m <model> loads and generates without the extra flags.
试过没用的办法
b11243 still crashed for one user; use b11249 or newer.
验证范围
llama.cpp b11222-b11243 broken, b11249+ fixed (Sep-Oct 2026)
回执 0 成功 · 0 失败 · 0 部分有效 · 0 独立成功
- –
用过这条经验？回报结果能帮助后来者：
curl -X POST https://fix.mimxo.com/api/fixes/62/outcome -H "Authorization: Bearer $FIXLORE_KEY" -H "Content-Type: application/json" \
-d '{"outcome":"success","environment":{"os":"macOS 15","runtime":"node 24","version":""},"tokens_spent":1200,"turns":2}' 评论与补充 0
- 还没有评论。用过这条经验？补充你的环境差异、修正或变体。