ComfyUI Node

SID LLM Local

Local Vision Models [No GPU] - No API needed

By slahiri·Created 9 months ago·Updated 7 months ago· 60
SID LLM Local
    • LLM_MODEL
    modelQwen3-VL-2B-Instruct
    quantizationAuto (Detect Best)
    deviceauto
    temperature0.3
    max_tokens_presetMedium (1024)
    custom_max_tokens1024
    keep_model_loadedtrue
    attention_modeauto
    repetition_penalty1.2
    top_p0.90
    use_torch_compilefalse
    CategorySID Photography Toolkit/LLM Providers

    Inputs (11)

    NameTypeDefaultDescription
    modelCOMBOQwen3-VL-2B-InstructSelect model: Name | Max Tokens | VRAM (4-bit)
    quantizationCOMBOAuto (Detect Best)Auto: detects best based on VRAM, 4-bit: lowest VRAM, FP16: best quality
    deviceCOMBOautoDevice to run on (auto recommended)
    temperatureFLOAT0.30–1Creativity (0=deterministic, 0.3=balanced, 1=creative)
    max_tokens_presetCOMBOMedium (1024)Output length: Low=512, Medium=1024, High=2048, Very High=Model Max
    custom_max_tokensINT1024128–4096Custom max tokens (only used when preset is 'Custom')
    keep_model_loadedBOOLEANtrueKeep model in VRAM between runs (faster repeat inference)
    attention_modeCOMBOautoAttention implementation: auto (recommended), flash_attention_2 (Ampere+), sdpa, eager
    repetition_penaltyFLOAT1.20.8–2Penalize repeated tokens (1.0=off, 1.2=recommended, 2.0=strong)
    top_pFLOAT0.900.1–1Nucleus sampling (0.9=recommended, lower=more focused)
    use_torch_compileBOOLEANfalseEnable torch.compile for faster inference (CUDA + Torch 2.1+ only, first run slower)

    Outputs (1)

    NameTypeDescription
    LLM_MODELLLM_MODELLLM configuration to connect to prompt generator