Repository navigation
Misc. bug: blank log lines in router mode #29878
Description
Name and Version
$ ./build/bin/llama-server --version
version: 0.5.0-dev (build 11370, commit bed0a85)
built with GNU 13.3.0 for Linux x86_64
Operating systems
Linux
Which llama.cpp modules do you know to be affected?
llama-server
Command line
./build/bin/llama-server \
--models-preset models/custom/config.ini \
--models-max 1Problem description & steps to reproduce
As of #28747, when the UI triggers a model to be loaded in router mode, it prints a ton of blank log lines by default.
First Bad Commit
Relevant log output
Logs
0.00.000.796 W server tools or MCP servers are enabled, using localhost as default CORS origin (change via --cors-origins)
0.00.000.846 I srv llama_server: initializing ...
0.00.443.062 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
0.00.447.317 I srv operator(): Available models (4):
0.00.447.320 I srv operator(): [ preset] Gemma4-26B-A4B
0.00.447.320 I srv operator(): [ preset] Impish-Bloodmoon
0.00.447.321 I srv operator(): [ preset] Qwen3.8-27B
0.00.447.321 I srv operator(): [ preset] Qwen3.8-Flash-Next
0.00.447.413 W srv llama_server: security: router mode, MCP proxy (experimental), server tools (experimental) enabled - do not expose to untrusted environments
0.00.447.416 I srv llama_server: starting server in router mode. models will be automatically loaded on-demand
0.00.449.619 I srv llama_server: listening on http://0.0.0.0:9931
0.05.018.325 I srv proxy_reques: proxying POST request to http://localhost:8123/mcp
0.05.018.332 I srv proxy_reques: proxying POST request to http://localhost:8000/mcp
0.05.047.643 I srv proxy_reques: proxying POST request to http://localhost:8000/mcp
0.05.055.668 I srv proxy_reques: proxying POST request to http://localhost:8000/mcp
0.05.055.676 I srv proxy_reques: proxying GET request to http://localhost:8000/mcp
0.05.056.058 I srv proxy_reques: proxying POST request to http://localhost:8123/mcp
0.05.062.161 I srv proxy_reques: proxying GET request to http://localhost:8123/mcp
0.05.062.210 I srv proxy_reques: proxying POST request to http://localhost:8123/mcp
0.10.001.212 I srv load: spawning server instance with name=Qwen3.8-Flash-Next on port 60267
0.10.001.243 I srv load: spawning server instance with args:
0.10.001.243 I srv load: /home/eapache/src/llama.cpp/build/bin/llama-server
0.10.001.244 I srv load: --host
0.10.001.244 I srv load: 127.0.0.1
0.10.001.244 I srv load: --min-p
0.10.001.244 I srv load: 0.0
0.10.001.244 I srv load: --port
0.10.001.244 I srv load: 60267
0.10.001.245 I srv load: --presence-penalty
0.10.001.245 I srv load: 0.0
0.10.001.245 I srv load: --reasoning-preserve
0.10.001.245 I srv load: --repeat-penalty
0.10.001.245 I srv load: 1.0
0.10.001.245 I srv load: --temperature
0.10.001.245 I srv load: 1.0
0.10.001.246 I srv load: --tools
0.10.001.246 I srv load: get_info
0.10.001.246 I srv load: --top-k
0.10.001.246 I srv load: 20
0.10.001.246 I srv load: --top-p
0.10.001.247 I srv load: 0.95
0.10.001.247 I srv load: --webui-mcp-proxy
0.10.001.247 I srv load: --alias
0.10.001.247 I srv load: Qwen3.8-Flash-Next
0.10.001.247 I srv load: --ctx-size
0.10.001.247 I srv load: 200000
0.10.001.248 I srv load: --fit-target
0.10.001.248 I srv load: 256,512
0.10.001.248 I srv load: --load-mode
0.10.001.248 I srv load: none
0.10.001.248 I srv load: --lazy-mode
0.10.001.249 I srv load: off
0.10.001.249 I srv load: --model
0.10.001.249 I srv load: ./models/custom/Qwen3.8-Flash-Next/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf
0.10.001.249 I srv load: --mmproj
0.10.001.249 I srv load: ./models/custom/Qwen3.8-Flash-Next/mmproj-BF16.gguf
0.10.001.250 I srv load: --ubatch-size
0.10.001.250 I srv load: 2048
[60267] 0.00.000.903 W server tools or MCP servers are enabled, using localhost as default CORS origin (change via --cors-origins)
[60267] 0.00.000.959 I srv llama_server: initializing ...
[60267] 0.00.085.135 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
[60267] 0.00.085.494 W srv llama_server: security: MCP proxy (experimental), server tools (experimental) enabled - do not expose to untrusted environments
[60267]
[60267] 0.00.088.130 I srv load_model: loading model './models/custom/Qwen3.8-Flash-Next/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf'
0.20.058.113 E srv operator(): http client error: Connection handling canceled
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267]
[60267] 0.51.393.553 I cmn init: llama threadpool init, n_threads = 12
[60267]
[60267] 0.51.601.038 W load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
[60267] 0.51.601.040 W load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
[60267] 0.51.601.040 W load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842
[60267]
[60267]
[60267]
[60267]
[60267]
[60267] 0.52.018.473 I srv load_model: loaded multimodal model, './models/custom/Qwen3.8-Flash-Next/mmproj-BF16.gguf'
[60267] 0.52.151.692 I srv load_model: initializing, n_slots = 4, n_ctx_slot = 200192, kv_unified = 'true'
[60267] 0.52.156.384 I srv llama_server: model loaded
[60267] 0.52.156.390 I srv llama_server: listening on http://127.0.0.1:60267
Activity
eapache commentedon Oct 2, 2026
cc @ServeurpersoCom
eapache commentedon Oct 2, 2026
Presumably the leading newline added by that PR is getting logged on its own by the router for some reason.
ServeurpersoCom commentedon Oct 3, 2026
I'm going to fix it so that the color displays everywhere on Windows 10 and Windows 11, just like on Mac and Linux, without blank lines (and without ugly workarounds, such as a blank-line filter on the router)
ServeurpersoCom commentedon Oct 3, 2026
Done in #29895, could you give it a try in your setup? You should no longer get any stray empty lines, and once CI passes we should be good.