Instructions to use saik0s/comfy_backup with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use saik0s/comfy_backup with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="saik0s/comfy_backup",
	filename="ComfyUI/models/text_encoders/gemma-3-12b-it-q2_k.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use saik0s/comfy_backup with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf saik0s/comfy_backup:Q4_K_S
# Run inference directly in the terminal:
llama cli -hf saik0s/comfy_backup:Q4_K_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf saik0s/comfy_backup:Q4_K_S
# Run inference directly in the terminal:
llama cli -hf saik0s/comfy_backup:Q4_K_S

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf saik0s/comfy_backup:Q4_K_S
# Run inference directly in the terminal:
./llama-cli -hf saik0s/comfy_backup:Q4_K_S

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf saik0s/comfy_backup:Q4_K_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf saik0s/comfy_backup:Q4_K_S

Use Docker

docker model run hf.co/saik0s/comfy_backup:Q4_K_S

LM Studio
Jan
Ollama
How to use saik0s/comfy_backup with Ollama:
```
ollama run hf.co/saik0s/comfy_backup:Q4_K_S
```

Unsloth Studio

How to use saik0s/comfy_backup with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for saik0s/comfy_backup to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for saik0s/comfy_backup to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for saik0s/comfy_backup to start chatting

How to use saik0s/comfy_backup with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saik0s/comfy_backup:Q4_K_S

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "saik0s/comfy_backup:Q4_K_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use saik0s/comfy_backup with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saik0s/comfy_backup:Q4_K_S

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default saik0s/comfy_backup:Q4_K_S

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use saik0s/comfy_backup with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf saik0s/comfy_backup:Q4_K_S

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "saik0s/comfy_backup:Q4_K_S" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use saik0s/comfy_backup with Docker Model Runner:
```
docker model run hf.co/saik0s/comfy_backup:Q4_K_S
```

Lemonade

How to use saik0s/comfy_backup with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull saik0s/comfy_backup:Q4_K_S

Run and chat with the model

lemonade run user.comfy_backup-Q4_K_S

List all available models

lemonade list

comfy_backup / custom_nodes /ComfyUI-INT8-Fast /Speed.md

saik0s

Add files using upload-large-folder tool

8d5e143 verified 22 days ago

preview code

Raw

History Blame Contribute Delete

1.91 kB

	# Speed:

	Measured on a 3090 at 1024x1024, 26 steps with Flux2 Klein Base 9B.

	\| Format \| Speed (s/it) ↓ \| Relative Speedup \|
	\|-------\|--------------\|------------------\|
	\| bf16 \| 2.07 \| 1.00× \|
	\| bf16 compile \| 2.24 \| 0.92× \|
	\| fp8 \| 2.06 \| 1.00× \|
	\| int8 \| 1.64 \| 1.26× \|
	\| int8 compile ★\| 1.04 \| 1.99× \|
	\| gguf8_0 compile \| 2.03 \| 1.02× \|

	3090, Qwen Image 2512.

	\| Format \| Speed (s/it) ↓ \|
	\|-------\|--------------\|
	\| Nunchaku INT4 Best Quality \| 1.21 \|
	\| Nunchaku INT4 with R128 Lora \| 1.36 \|
	\| INT8 ConvRot compile \| 1.26 \|
	\| INT8 Row compile ★\| 1.18 \|
	\| INT8 R128 Lora \| No slowdown, except if dynamic. \|

	I would also like to point out that we beat Nunchaku INT4 on every quality measurement in the [Quality Metrics](Metrics.md)

	Additionally, the quality of loras applied with [this nunchaku lora node](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader) appears to be degraded.

	Klein 9B, Measured on an 8gb 5060, same settings as the 3090 run:

	\| Format \| Speed (s/it) ↓ \| Relative Speedup \|
	\|-------\|--------------\|------------------\|
	\| fp8 \| 3.04 \| 1.00× \|
	\| fp8 fast \| 3.00 \| 1.00× \|
	\| fp8 compile \| couldn't get to work \| ??× \|
	\| int8 \| 2.53 \| 1.20× \|
	\| int8 compile ★\| 2.25 \| 1.35× \|

	8gb RTX 5060, Anima, Comfy version from 2026-05-02, Pytorch 2.11+CU13.0, latest kitchen triton and everything else

	\| Format \| Speed (it/s) ↑ \|
	\|-------\|--------------\|
	\| bf16 \| 0.78 \|
	\| INT8 ConvRot \| 1.12 \|
	\| INT8 Row \| 1.24 \|
	\| INT8 ConvRot Compile \| 1.47 \|
	\| MXFP8 \| 0.89 \|
	\| MXFP8 --fast \| 0.93 \|
	\| MXFP8 + Compile \| Still failing. \|

	Finally have gotten compile with --fast to work with mxfp8, PyTorch 2.13.0.dev20260511+cu132, RTX5060 same as before.

	Quality results for this run, can be found here: [Anima Results](Metrics.md#anima-on-a-5060)

	\| Format \| Speed (it/s) ↑ \|
	\|-------\|--------------\|
	\| MXFP8 --fast + Compile \| 1.37it \|
	\| INT8 ConvRot + Compile \| 1.47it \|

	# Speed:

	Measured on a 3090 at 1024x1024, 26 steps with Flux2 Klein Base 9B.

	\| Format \| Speed (s/it) ↓ \| Relative Speedup \|
	\|-------\|--------------\|------------------\|
	\| bf16 \| 2.07 \| 1.00× \|
	\| bf16 compile \| 2.24 \| 0.92× \|
	\| fp8 \| 2.06 \| 1.00× \|
	\| int8 \| 1.64 \| 1.26× \|
	\| int8 compile ★\| 1.04 \| 1.99× \|
	\| gguf8_0 compile \| 2.03 \| 1.02× \|

	3090, Qwen Image 2512.

	\| Format \| Speed (s/it) ↓ \|
	\|-------\|--------------\|
	\| Nunchaku INT4 Best Quality \| 1.21 \|
	\| Nunchaku INT4 with R128 Lora \| 1.36 \|
	\| INT8 ConvRot compile \| 1.26 \|
	\| INT8 Row compile ★\| 1.18 \|
	\| INT8 R128 Lora \| No slowdown, except if dynamic. \|

	I would also like to point out that we beat Nunchaku INT4 on every quality measurement in the [Quality Metrics](Metrics.md)

	Additionally, the quality of loras applied with [this nunchaku lora node](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader) appears to be degraded.

	Klein 9B, Measured on an 8gb 5060, same settings as the 3090 run:

	\| Format \| Speed (s/it) ↓ \| Relative Speedup \|
	\|-------\|--------------\|------------------\|
	\| fp8 \| 3.04 \| 1.00× \|
	\| fp8 fast \| 3.00 \| 1.00× \|
	\| fp8 compile \| couldn't get to work \| ??× \|
	\| int8 \| 2.53 \| 1.20× \|
	\| int8 compile ★\| 2.25 \| 1.35× \|

	8gb RTX 5060, Anima, Comfy version from 2026-05-02, Pytorch 2.11+CU13.0, latest kitchen triton and everything else

	\| Format \| Speed (it/s) ↑ \|
	\|-------\|--------------\|
	\| bf16 \| 0.78 \|
	\| INT8 ConvRot \| 1.12 \|
	\| INT8 Row \| 1.24 \|
	\| INT8 ConvRot Compile \| 1.47 \|
	\| MXFP8 \| 0.89 \|
	\| MXFP8 --fast \| 0.93 \|
	\| MXFP8 + Compile \| Still failing. \|

	Finally have gotten compile with --fast to work with mxfp8, PyTorch 2.13.0.dev20260511+cu132, RTX5060 same as before.

	Quality results for this run, can be found here: [Anima Results](Metrics.md#anima-on-a-5060)

	\| Format \| Speed (it/s) ↑ \|
	\|-------\|--------------\|
	\| MXFP8 --fast + Compile \| 1.37it \|
	\| INT8 ConvRot + Compile \| 1.47it \|