GGUF (GPT-Generated Unified Format) is a successor of GGML (GPT-Generated Model Language), it was released on August 21, 2023; by the way, GPT stands for Generative Pre-trained Transformer.

This package is a simple graphical user interface (GUI) application that uses the ctransformers or llama.cpp to interact with a chat model for generating responses.

Install the connector via pip (once only):

pip install gguf-connector

Update the connector (if previous version installed) by:

pip install gguf-connector --upgrade

With this version, you can interact straight with the GGUF file(s) available in the same directory by a simple command.

Select model(s) with ctransformers (optional: need ctransformers to work; pip install ctransformers):

ggc c

Select model(s) with llama.cpp connector (optional: need llama-cpp-python to work; get it here or nightly here):

ggc cpp

Select model(s) with ctransformers:

ggc g

Select model(s) with llama.cpp connector:

ggc gpp

Select model(s) with vision connector:

ggc v

Opt a clip handler then opt a vision model; prompt your picture link to process (see example here)

Select model(s) with metadata reader:

ggc r

Select model(s) with metadata fast reader:

ggc r2

Select model(s) with tensor reader (optional: need torch to work; pip install torch):

ggc r3

Load PDF(s) into a model with ctransformers:

ggc cp

Load PDF(s) into a model with llama.cpp connector:

ggc pp

optional: need pypdf; pip install pypdf

Prompt WAV(s) into a model with ctransformers:

ggc cs

Prompt WAV(s) into a model with llama.cpp connector:

ggc ps

optional: need speechrecognition, pocketsphinx; pip install speechrecognition, pocketsphinx

Prompt WAV(s) into a model with ctransformers:

ggc cg

Prompt WAV(s) into a model with llama.cpp connector:

ggc pg

Launch to page/container:

ggc w

Divide gguf into different part(s) with a cutoff point (size):

ggc d2

Merge all gguf into one:

ggc m2

Merge all safetensors into one (optional: need torch to work; pip install torch):

ggc ma

Split checkpoint into components (optional: need torch to work; pip install torch):

ggc s

Quantize safetensors to fp8 (downscale; optional: need torch to work; pip install torch):

ggc q

Quantize safetensors to fp32 (upscale; optional: need torch to work; pip install torch):

ggc q2

Convert safetensors to gguf (auto; optional: need torch to work; pip install torch):

ggc t

Convert safetensors to gguf (meta; optional: need torch to work; pip install torch):

ggc t1

Convert safetensors to gguf (unlimited; optional: need torch to work; pip install torch):

ggc t2

Convert gguf to safetensors (reversible; optional: need torch to work; pip install torch):

ggc t3

Rename lora tensor (base/unet swappable; optional: need torch to work; pip install torch):

ggc la

Tensor remover:

ggc rm

Tensor renamer:

ggc rn

Tensor extractor:

ggc ex

Get cutter for bf/f16 to q2-q8 quantization (see user guide here) by:

ggc u

Download comfy pack (see user guide here) by:

ggc y

Clone node (see user/setup guide here) by:

ggc n

Take pack (see user guide here) by:

ggc p

Take packpack (see user guide here) by:

ggc p2

Take framepack (portable packpack) by:

ggc p1

Run framepack - ggc edition by (optional: need framepack to work; pip install framepack):

ggc f2

Activate backend and frontend by (optional: need transformers to work; pip install transformers):

ggc g1

Activate backend and frontend by (optional: need torch, diffusers to work; pip install torch, diffusers):

ggc v1

Activate backend and frontend by (optional: need torch, diffusers to work; pip install torch, diffusers):

ggc v2

Activate backend and frontend by (optional: need torch, diffusers to work; pip install torch, diffusers):

ggc i2

Activate backend and frontend by (optional: need torch, diffusers to work; pip install torch, diffusers):

ggc k2

With lora selection:

ggc k1

With memory economy mode (low/no vram or w/o gpu option):

ggc k3

Activate backend and frontend by (optional: need torch, diffusers to work; pip install torch, diffusers):

ggc k4

Activate backend and frontend by (optional: need transformers to work; pip install transformers):

ggc n2

Activate backend and frontend by (optional: need torch to work; pip install torch):

ggc f5

Realtime live captioning:

ggc f7

Connector mode, opt a gguf to interact with (see example here):

ggc f6

Activate accurate/precise mode by (optional: need vtoo to work; pip install vtoo):

ggc h3

Opt a model file to interact with (see example here)

Activate backend and frontend by (optional: need diao to work; pip install diao):

ggc s2

Activate backend and frontend by (optional: need higgs to work; pip install higgs):

ggc h2

Multilingual supported, i.e., Spanish, German, Korean, etc.

Activate backend and frontend by (optional: need bagel2 to work; pip install bagel2):

ggc b2

Opt a vae then opt a model file (see example here)

Activate backend and frontend by (optional: need chichat to work; pip install chichat):

ggc c2

Opt a vae, a clip (t3-cfg) and a model file (see example here)

Multilingual (optional: need chichat to work; pip install chichat):

ggc c3

Opt a vae, a clip (t3-23lang) and a model file (see example here)

Activate backend and frontend by (optional: need fishaudio to work; pip install fishaudio):

ggc o2

Opt a codec then opt a model file (see example here)

Activate backend and frontend by (optional: need dequantor to work; pip install dequantor):

ggc k7

Opt a model file in the current directory (see example here)

Activate backend and frontend by (optional: need dequantor to work; pip install dequantor):

ggc k8

Opt a model file in the current directory (see example here)

Select flux image model(s) with k connector:

ggc k

Select qwen image model(s) with q5 connector:

ggc q5

Opt a model file to interact with (see example here)

Select image edit model(s) with q6 connector:

ggc q6

Opt a model file to interact with (see example here)

Select image edit plus model(s) with q7 connector:

ggc q7

Select image edit plus model(s) with q8 connector:

ggc q8

Opt a model file to interact with (see example here)

Select image edit plus/lite model(s) with q9 connector (need nunchaku to work; get it here):

ggc q9

Opt a scaled 4-bit safetensors model file to interact with

Select image edit lite model(s) with q0 connector:

ggc q0

Select image edit lite2 model(s) with p0 connector:

ggc p0

Select image edit lite3 model(s) with p9 connector:

ggc p9

Select lumina image model(s) with l2 connector:

ggc l2

Select wan video model(s) with w2 connector:

ggc w2

Select ltxv model(s) with x2 connector:

ggc x2

Select mochi model(s) with m1 connector:

ggc m1

Select kontext model(s) with k0 connector:

ggc k0

Opt a model file to interact with (see example here)

Select sd3.5 model(s) with s3 connector:

ggc s3

Opt a model file to interact with (see example here)

Select higgs model(s) with h6 connector:

ggc h6

Opt a model file to interact with (see example here)

Select dia model(s) with s6 connector:

ggc s6

Opt a model file to interact with (see example here)

Select fastvlm model(s) with f9 connector:

ggc f9

Opt a model file to interact with (see example here)

Select vibevoice model(s) with v6 connector (optional: need yvoice to work; pip install yvoice):

ggc v6

Opt a model file to interact with (see example here)

Select docling model(s) with n3 connector:

ggc n3

Opt a model file to interact with (see example here)

Activate backend and frontend by (optional: need gudio to work; pip install gudio):

ggc g2

Opt a model then opt a clip (see example here)

Sketch gguf connector (optional: need dequantor to work; pip install dequantor):

ggc s8

Sketch safetensors connector (optional: need nunchaku to work; get it here):

ggc s9

Opt a model file to interact with (see example here)

Studio connector:

ggc w9

Opt a recognizor, a generator, and a transformer to interact with (see example here)

Fast sd3.5 connector:

ggc w8

Fast lumina connector:

ggc w7

Fast flux connector:

ggc w6

Frontend test.gguf.org or localhost (open a new console/terminal: ggc b)

Fast edit connector:

ggc e8

Fast plus connector:

ggc e9

Frontend gguf.org or localhost (open a new console/terminal: ggc a)

Enter the main menu for selecting a connector or getting pre-trained trial model(s).

ggc m

Include the connector selection menu to your code by:

from gguf_connector import menu

For standalone version please refer to the repository in the reference list (below).

model selector (standalone version: installable package)

cgg (cmd-based tool)

ctransformers llama.cpp

understanding gguf and the gguf-connector

gguf.org (you can download the frontend from github and host it locally; the backend is ethereum blockchain)

gguf.io (i/o is a mirror of us; note: this web3 domain might be restricted access in some regions/by some service providers, if so, visit the one above/below instead, exactly the same)

gguf.us