Nodes/Kinburg-Nodes/Safetensors -> GGUF (llama.cpp)
ComfyUI Node

Safetensors -> GGUF (llama.cpp)

A ComfyUI node in Kinburg-Nodes/LLM/GGUF with 13 inputs and 3 outputs.

By Kinburg·Created 2 months ago·Updated 3 days ago· 1
Safetensors -> GGUF (llama.cpp)
    • gguf_path
    • log
    • help
    source
    output_dir
    output_name
    outtypef16
    quantizenone
    quantize_binary
    quantize_binary_path
    llama_cpp_dir
    auto_clonetrue
    hf_token
    hf_download_dir
    keep_intermediatefalse
    forcefalse
    CategoryKinburg-Nodes/LLM/GGUF

    Inputs (13)

    NameTypeDefaultDescription
    sourceSTRINGHF repo id ('owner/name'), a https://huggingface.co/owner/name URL, a local HF model folder, or a single .safetensors file. Windows 'Copy as path' quotes are stripped.
    output_dirSTRINGFolder to write the .gguf into. Created if missing. Blank -> ComfyUI/models/llm.
    output_nameSTRINGBase filename without extension. Blank -> derived from the model name. The precision/quant tag and .gguf are added automatically.
    outtypeCOMBOf16Precision written by convert_hf_to_gguf.py. f16 is the usual base to quantize from. When quantize != none this is the intermediate.
    quantizeCOMBOnoneOptional 2nd pass with llama-quantize to shrink the model (e.g. Q4_K_M). none = keep the outtype gguf. Needs a compiled llama-quantize binary.
    quantize_binaryCOMBOllama-quantize executable (auto-listed from ComfyUI/models/llm). Choose the placeholder to type a path in quantize_binary_path. Only used when quantize != none.
    quantize_binary_pathSTRINGFull path to llama-quantize(.exe), used when the dropdown is the placeholder.
    llama_cpp_diroptSTRINGPath to a llama.cpp checkout (holds convert_hf_to_gguf.py). Blank -> use/auto-clone one under ComfyUI/models/llm/llama.cpp.
    auto_cloneoptBOOLEANtrueIf no llama.cpp is found, git clone it automatically. Needs git on PATH. Clones source only (no compiled llama-quantize).
    hf_tokenoptSTRINGHuggingFace token for gated/private repos. Leave blank for public models.
    hf_download_diroptSTRINGWhere to download HF repos. Blank -> the default HuggingFace cache. Ignored for local sources.
    keep_intermediateoptBOOLEANfalseWhen quantizing, keep the intermediate outtype gguf as well as the quantized one. Off -> delete it after quantizing.
    forceoptBOOLEANfalseRe-run even if the target .gguf already exists. Off -> return the existing file without redoing the work.

    Outputs (3)

    NameTypeDescription
    gguf_pathSTRING
    logSTRING
    helpSTRING