Instructions to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S 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 YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
# Run inference directly in the terminal:
llama cli -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
# Run inference directly in the terminal:
llama cli -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

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 YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
# Run inference directly in the terminal:
./llama-cli -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

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 YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Use Docker

docker model run hf.co/YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

LM Studio
Jan

vLLM

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Ollama
How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with Ollama:
```
ollama run hf.co/YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
```
Unsloth Desktop

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with Docker Model Runner:
```
docker model run hf.co/YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S
```

Lemonade

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Run and chat with the model

lemonade run user.Qwopus3.5-27B-v3-Abliterated-TQ3_4S-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

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 YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S

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 "YTan2000/Qwopus3.5-27B-v3-Abliterated-TQ3_4S" \
  --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"

VRAM usage, quantization & Context length

by cnsiva - opened Apr 23

Discussion

cnsiva

Apr 23

Hi YTan2000,
Thank you for your work. I tried couple of you models before in my environment, I didn't the great success. The response is garbled out in some cases. So, I have the following question.

Do I need to quantize in my environment using llama-quantize (built it from llama.cpp-tq3) ?
Can this model fit on my VRAM (4079 Ti Super 16GB)
What is max context length I can use?
Is any other llama-server param I should be using?

MySystem Configuration:
OS: Ubuntu 26.04
CPU: AMD 7950X3D
RAM: 64 GB DDR5 6000 MHz
GPU: 4070 Ti Super 16 GB

awesome747

Apr 23

16gb is plenty. Can you try the latest Qwen3.6-35B? 35b can fit 128k but you can always try higher or lower with -ngl 99 or less 99 is max.

and try 27b which can take 32k.

Quantization sometimes remove the quality especially the abilterated one.

You may want to try cuda 12 as some version of cuda version has contributed to garbage. Check one of the discussion recently who had the same problem

awesome747

Apr 23

https://huggingface.co/YTan2000/Qwen3.5-27B-TQ3_4S/discussions/2#69d805a99f1a86ffe01a0735

All my model can work with 16gb as my card is 16gb

cnsiva

Apr 24

I am interested to run the dense model (Qwen3.6-27B or Gemma4-31B) with TQ3. will this work in 16 GB VRAM with minimum 64K context?

YTan2000

Owner Apr 24

27b 32-48k, if you are machine is headless then 64k no problem full in GPU. If you offload to CPU (ngl <99) , you can fit more but slower
Gemma4 similar.
Qwen3.6-A35b is a lot better, can go to max context 256k. I only use 128k and more than enough for a lot of task. If you use open code or other cli, it auto compact the context for you.

cnsiva

Apr 24

Thank you. I have a headless Ubuntu server, and I can dedicate all resources to the llama.cpp server. I tried both Qwen3.6-27B and Qwen3.6-A35b with TQ3, but neither of them worked. Both of them produced garbled output. So, my question is, do I need to quantize the model in my server environment and use it? If you have any llama-server startup parameters that I can use, it would be great if you could share them. I would like to continue using llama.cpp-tq3.

MySystem Configuration:
OS: Ubuntu 26.04
CPU: AMD 7950X3D
RAM: 64 GB DDR5 6000 MHz
GPU: 4070 Ti Super 16 GB

awesome747

Apr 24

•

edited Apr 24

Sorry, away for two weeks. Try load other models that is non tq3_4s. See that you have the same problem.

Ensure you get the latest from main.
Also check cuda version. As mentioned, certain cuda version has issue. Try 12.9

Settings: -ngl 99 -ctk q4_0 -ctv tq3_0 -fa on --cache-ram 0 --jinja --reasoning off --reasoning-budget 0 --reasoning-format deepseek -c 32768

awesome747

Apr 24

https://huggingface.co/YTan2000/Qwen3.5-27B-TQ3_4S/discussions/2#69d805a99f1a86ffe01a0735

Check this thread

awesome747

Apr 24

https://huggingface.co/YTan2000/Qwopus3.5-27B-v3-TQ3_4S/discussions/3#69e3407ae1003d7517ddc548

Also check this. Sorry can't help as I am on the plane. Good luck. May be as chatgpt or claude

gdevenyi

Apr 24

Take a look at https://github.com/gdevenyi/huggingface-estimate for a way to calculate memory requirements. It properly supports TurboQuant weights during estimation.

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