Instructions to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller",
	filename="Huihui-Qwen3.6-27B-abliterated-i1-IQ4_XS-FFN-IQ3.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": [
				{
					"type": "text",
					"text": "Describe this image in one sentence."
				},
				{
					"type": "image_url",
					"image_url": {
						"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
					}
				}
			]
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller 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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
# Run inference directly in the terminal:
llama cli -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
# Run inference directly in the terminal:
llama cli -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
# Run inference directly in the terminal:
./llama-cli -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
# Run inference directly in the terminal:
./build/bin/llama-cli -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

Use Docker

docker model run hf.co/lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

LM Studio
Jan

vLLM

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

Ollama
How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with Ollama:
```
ollama run hf.co/lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
```

Unsloth Studio

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller 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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller 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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller to start chatting

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

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": "lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

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 lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with Docker Model Runner:
```
docker model run hf.co/lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS
```

Lemonade

How to use lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull lemonyins/Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller:IQ4_XS

Run and chat with the model

lemonade run user.Qwen3.6-27B-uncensored-abliterated-i1-IQ4_XS-GGUF-Smaller-IQ4_XS

List all available models

lemonade list

out of memory at 160k

by starskyzheng - opened May 20

Discussion

starskyzheng

May 20

By using Huihui-Qwen3.6-27B-abliterated-i1-IQ4_XS-FFN-IQ3.gguf (13Gb), I fail to perform with 160k content length.
The example was Qwen3.6-27B-abliterated-i1-IQ4_XS-FFN-IQ3_S.gguf, was it the same model?

lemonyins

Owner May 21

Yes, the KV needs to be set to turbo4.

MachineCat

26 days ago

•

edited 26 days ago

Same here, cannot get that much of content length.
/path/to/llama-cpp-turboquant/build/bin/llama-server -m /path/to/models/Huihui-Qwen3.6-27B-abliterated-i1-IQ4_XS-FFN-IQ3.gguf -c 102400 -ngl 99 --flash-attn on --cache-type-k turbo4 --cache-type-v turbo4 --host 0.0.0.0

I did not try the exact threshold above 100kB, but it did fail on me for 128kB

Running on a laptop with 5080 16GB VRAM, Kubuntu 26.04

lemonyins

Owner 23 days ago

Is there another application that is using the GPU memory? My monitor uses an integrated graphics card, and an independent graphics card can be fully used for loading large models.

MachineCat

23 days ago

Emm...not sure if it because of the llama-server behave differently on Windows and Linux?
Here are something I found:

The Linux server I ran actually automatically upgraded the k cache: "llama_kv_cache: auto-asymmetric: GQA ratio 6:1 (n_head=24, n_head_kv=4) — upgrading K from turbo4 to q8_0 to prevent quality degradation. Disable with TURBO_AUTO_ASYMMETRIC=0"
It is running 4 slots in parallel by default. Is your Windows running only 1 slot?

Anyway, with those two fixes, now I can get 160k context length.
But this actually makes me thinking, maybe I should keep the q8 K for better quality as 100k context length is already good enough for what I am trying to do 😀

lemonyins

Owner 22 days ago

So that's how it is, by setting TURBO_AUTO_ASYMMETRIC=0, the KV Cache uses the turbo4 format instead of the auto-upgraded q8_0.

Or use the released version fixed by this issue from https://github.com/lemonyins/llama-cpp-turboquant-mtp

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment