ComfyUI Node
[llama.cpp] Decide
Scores letter-token choices with a Native Session prefill or a Runtime or Connect Session's constrained llama-server completion. Optional media is added to the context; question and answers remain text-only.
[llama.cpp] Decide
- session
- question
- images
- audio
- video
- model_profile
- selected
- probabilities_json
- metrics
- media_diagnostics
- session
◄system►
◄context►
◄video_with_audiofalse►
◄seed-1►
◄session_unloadfalse►
Categoryllama_cpp/decision
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| session | OLLAMA_IMAGE_LIST_LLAMA_CPP_SESSION | — | |
| system | STRING | — | |
| context | STRING | — | |
| question | OLLAMA_IMAGE_LIST_LLAMA_CPP_QUESTION | — | |
| video_with_audio | BOOLEAN | false | — |
| seed | INT | -1-1–4294967295 | Runtime completion seed; Native prefill scoring is deterministic. |
| session_unload | BOOLEAN | false | Unload the session after the decision completes. |
| imagesopt | IMAGE | — | |
| audioopt | AUDIO | — | |
| videoopt | VIDEO | — | |
| model_profileopt | OLLAMA_IMAGE_LIST_LLAMA_CPP_MODEL_PROFILE | Optional override; server sessions apply sampling and explicit on/off reasoning settings per request. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| selected | STRING | — |
| probabilities_json | STRING | — |
| metrics | STRING | — |
| media_diagnostics | OLLAMA_IMAGE_LIST_LLAMA_CPP_MEDIA_DIAGNOSTICS | — |
| session | OLLAMA_IMAGE_LIST_LLAMA_CPP_SESSION | — |