ComfyUI Node
Llama.cpp Generate
Runs one multimodal llama.cpp completion using separate Compact model and hardware profiles plus optional reasoning and speculative configs.
Llama.cpp Generate
- model_profile
- hardware_profile
- reasoning
- speculative
- images
- audio
- video
- response
- thinking
- raw JSON
- metrics
- media diagnostics
◄model_path[no GGUF models found]►
◄mmproj_path[none]►
◄system►
◄prompt►
◄n_ctx8192►
◄max_tokens512►
◄image_max_tokens0►
◄seed-1►
◄stop►
◄verbosefalse►
◄image_min_tokens0►
CategoryOllama/llama_cpp/compact
Inputs (18)
| Name | Type | Default | Description |
|---|---|---|---|
| model_path | COMBO | [no GGUF models found] | 1 options: [no GGUF models found] |
| mmproj_path | COMBO | [none] | 1 options: [none] |
| model_profile | OLLAMA_IMAGE_LIST_LLAMA_CPP_MODEL_PROFILE | Required output from Llama.cpp Model Profile. | |
| system | STRING | — | |
| prompt | STRING | — | |
| n_ctx | INT | 8192512–1048576 | — |
| max_tokens | INT | 5121–131072 | — |
| image_max_tokens | INT | 00–65536 | 0 uses the mmproj/handler default. A positive value overrides the per-image or per-video-frame token ceiling. |
| seed | INT | -1-1–4294967295 | — |
| stop | STRING | — | |
| verbose | BOOLEAN | false | — |
| image_min_tokens | INT | 00–65536 | 0 keeps the projector default. Qwen-VL grounding tasks may require 1024. |
| hardware_profileopt | OLLAMA_IMAGE_LIST_LLAMA_CPP_HARDWARE_RUNTIME_PROFILE | Optional output from Llama.cpp Hardware Runtime Profile. Disconnected uses GPU Full Offload. | |
| reasoningopt | OLLAMA_IMAGE_LIST_LLAMA_CPP_REASONING_CONFIG | Optional output from Llama.cpp Thinking / Reasoning Config. Disconnected uses model-default reasoning behavior. | |
| speculativeopt | OLLAMA_IMAGE_LIST_LLAMA_CPP_SPECULATIVE_CONFIG | Optional shared output from a Compact N-gram or Native Speculative Config node. | |
| imagesopt | IMAGE | — | |
| audioopt | AUDIO | — | |
| videoopt | VIDEO | — |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| response | STRING | — |
| thinking | STRING | — |
| raw JSON | STRING | — |
| metrics | STRING | — |
| media diagnostics | OLLAMA_IMAGE_LIST_LLAMA_CPP_MEDIA_DIAGNOSTICS | — |