Upstream: ggml-org#27351
A router preset with no model source returns success and launches a child llama-server that also enters router mode.
Environment: Linux aarch64, NVIDIA GB10, CUDA 13.0; current main f3f1a8f2760f28325a5ec20c05b171e5b7c83a29.
Reproduction:
version = 1
[broken-model]
ctx-size = 512
llama-server --models-preset broken.ini --models-max 1 --host 127.0.0.1 --port 18151
curl -X POST http://127.0.0.1:18151/models/load -H 'Content-Type: application/json' -d '{"model":"broken-model"}'Observed: HTTP 200 {"success":true}; a child starts without --model/--hf-repo and logs starting server in router mode.
Likely cause: server_models::load does not reject a preset lacking a model source before spawning the child. Upstream PR ggml-org#27541 proposed that guard but closed unmerged.
Upstream: ggml-org#27351
A router preset with no model source returns success and launches a child
llama-serverthat also enters router mode.Environment: Linux aarch64, NVIDIA GB10, CUDA 13.0; current main
f3f1a8f2760f28325a5ec20c05b171e5b7c83a29.Reproduction:
Observed: HTTP 200
{"success":true}; a child starts without--model/--hf-repoand logsstarting server in router mode.Likely cause:
server_models::loaddoes not reject a preset lacking a model source before spawning the child. Upstream PR ggml-org#27541 proposed that guard but closed unmerged.