Misc. bug: failed to initialize router models: option 'reasoning-effort' not recognized

[Jackal830]

Name and Version

version: 10563 (bd9bd1b)
built with GNU 15.2.0 for Linux x86_64

Recently converted over from mainline llama.cpp and my config was not able to be 'forklifted' as is due to reasoning-effort not being supported in my models.ini section

Section in question:

[qwen-27b]
model = /docker/llama/ai_models/Qwen3.8-27B-UD-IQ4_XS.gguf
alias = qwen-27b
np = 4
reasoning-effort = low
spec-type = draft-mtp
spec-draft-n-max = 2
fit-target = 0
fit-ctx = 32768
c = 32768

Message received on startup:

llama_server: failed to initialize router models: option 'reasoning-effort' not recognized in preset 'qwen-27b'

Any chance support for reasoning-effort flag could be supported for Qwen?

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-cli

Command line

Runs as docker container, compose is as follows. The models.ini file has the qwen section stated above.

root@wd40:/docker/configs/llama# cat docker-compose.yaml
services:
  llama:
    image: local/vulkan-gfx1201-base:latest
    container_name: llama
    restart: unless-stopped
    privileged: true
    volumes:
      - /dev/dri:/dev/dri
      - /docker/llama:/docker/llama
    ports:
      - "8083:8083"
    entrypoint: /docker/llama/llama-cpp-turboquant/build/bin/llama-server
    command: >
      --models-preset /docker/llama/ai_models/models.ini
      --models-max 1
      --host 0.0.0.0
      --port 8083
      --timeout 3600
      --kv-unified

Problem description & steps to reproduce

Simply launch server with reasoning-effort flag, states it's not supported. It is supported in mainline and am hopeful that support can be merged here?

First Bad Commit

No response

Relevant log output

llama_server: failed to initialize router models: option 'reasoning-effort' not recognized in preset 'qwen-27b'

[Jackal830]

I should note that you can still specify a default via:

chat-template-kwargs = {"reasoning_effort":"low"}

It still would be nice to support the same flags as mainline though.