ComfyUI Node
[llama.cpp] Thinking / Reasoning Profile
Controls thinking/reasoning mode, effort, and token budget for supported model templates. A zero token limit applies no separate reasoning budget.
[llama.cpp] Thinking / Reasoning Profile
- reasoning
◄reasoning_modeauto►
◄reasoning_effortauto►
◄max_reasoning_tokens0►
◄preserve_thinkingfalse►
Categoryllama_cpp/profile
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| reasoning_mode | COMBO | auto | auto leaves template controls untouched; off and on explicitly disable or enable reasoning. |
| reasoning_effort | COMBO | auto | Used only when reasoning_mode is on. |
| max_reasoning_tokens | INT | 00–65536 | 0 applies no separate reasoning limit. Reasoning and final output still share Generate's max_tokens. |
| preserve_thinking | BOOLEAN | false | Preserve thinking/reasoning content in the chat history passed to the model.Currently, Qwen3.5+ only and may not work. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| reasoning | OLLAMA_IMAGE_LIST_LLAMA_CPP_REASONING_CONFIG | — |