ComfyUI Node
BSAI NPU LLM Decision Text
A ComfyUI node in BSAI/NPU with 8 inputs and 2 outputs.
BSAI NPU LLM Decision Text
- text
- meta_json
◄promptSummarize the scene in no more than 30 words and suggest the next processing step.►
◄max_new_tokens128►
◄temperature0.70►
◄top_p0.90►
◄enable_thinkingfalse►
◄backendNPU (local 4B INT4)►
◄ninfer_model(Auto — first model found)►
◄nemotron_model(Auto — first model found)►
CategoryBSAI/NPU
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Summarize the scene in no more than 30 words and suggest the next processing step. | — |
| max_new_tokens | INT | 1288–1024 | — |
| temperature | FLOAT | 0.700.1–2 | — |
| top_p | FLOAT | 0.900.1–1 | — |
| enable_thinking | BOOLEAN | false | — |
| backend | COMBO | NPU (local 4B INT4) | NPU runs the 4B Qwen locally; Strata/Nemotron/NInfer call the local HTTP engines (dynamic free port for NInfer) |
| ninfer_model | COMBO | (Auto — first model found) | NInfer backend only: which *.ninfer model to load. Changing it restarts the NInfer engine. |
| nemotron_model | COMBO | (Auto — first model found) | Nemotron backend only: which GGUF model to load (models/LLM/Nemotron, excludes mmproj). Changing it restarts llama.cpp. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| text | STRING | — |
| meta_json | STRING | — |