Misc. bug: Reasoning missing from output on Llama server webui

[webitube]

Name and Version

v.0.3.1: Windows x64, CUDA 13.1: v0.3.1

Operating systems

Windows

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

llama-server

Command line

beellama-v0.3.1-bin-win-cuda-13.1-x64\llama-server.exe -m "F:\LlamaGGUFModels\unsloth\Qwen3.6-27B-Q4_K_M\Qwen3.6-27B-Q4_K_M.gguf" --spec-draft-model "F:\LlamaGGUFModels\dflash\qwen3.6-27B-dflash-Q4_K_M.gguf" --spec-type dflash --spec-dflash-cross-ctx 1024 --port 8080 -np 1 --kv-unified -ngl all --spec-draft-ngl all -b 2048 -ub 512 --ctx-size 102400 --cache-type-k q5_0 --cache-type-v q4_1 --flash-attn on --jinja --no-mmap --mlock --no-host --reasoning on --chat-template-kwargs "{\"preserve_thinking\":true}" --temp 0.6 --top-k 20 --top-p 1.0 --min-p 0.0 --verbose

Problem description & steps to reproduce

prompt: "What is the capital of California?"

In the output, there is no thinking even though "--reasoning on" is in the command line and the llama-server webui developer "enable thinking" is checked.

Note that the server request shows the following in the verbose log (i.e., "enable_thinking": false):

"chat_template_kwargs":{"enable_thinking":false},"reasoning_control":true,"backend_sampling":false,"timings_per_token":true}

First Bad Commit

Update: I've tested master branch builds on Llama.cpp and it looks like the last build where reasoning worked for me was:

b9473: ggml-org#23981: ggerganov: f7a0777: kv-cache : SWA checkpoints store only non-masked cells (ggml-org#23981)

So I guess maybe b9474?

ggml-org#23434: allozaur; f8e67fc: ui: Add Thinking mode toggle with reasoning effort levels + improvements for Chat Form Add Action UI (ggml-org#23434)

I have no idea though.

Relevant log output

Logs

�[34m0.12.827.122�[0m �[33mD srv  log_server_r: request:  {"messages":[{"role":"user","content":"What is the capital of California?"}],"stream":true,"return_progress":true,"reasoning_format":"auto","chat_template_kwargs":{"enable_thinking":false},"reasoning_control":true,"backend_sampling":false,"timings_per_token":true}

verbose.log

model_info
model_info.txt

[Anbeeld]

There's a thinking toggle that was added to webui, a bulb icon at the bottom of the prompt text box.

[webitube]

Ah, I see it now. It's so small that I mistook it for a microphone icon.
Anyway, it's working! Thank you!