Instructions to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF 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 yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

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 yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

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 yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

Use Docker

docker model run hf.co/yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yethdev/qwen3.5-0.8b-manumit-v2-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "yethdev/qwen3.5-0.8b-manumit-v2-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

Ollama
How to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF with Ollama:
```
ollama run hf.co/yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF with Docker Model Runner:
```
docker model run hf.co/yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M
```

Lemonade

How to use yethdev/qwen3.5-0.8b-manumit-v2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.qwen3.5-0.8b-manumit-v2-GGUF-Q4_K_M

List all available models

lemonade list

Atomic Chat

Qwen3.5-0.8B, manumit v2, GGUF

GGUF builds of qwen3.5-0.8b-manumit-v2 for llama.cpp, Ollama, LM Studio, and anything else that reads GGUF. Same weights as the safetensors release, just quantized so they run on CPU or a small GPU.

manumit finds the directions in the residual stream that carry refusal and projects them out of the weights, then heals the model back on ordinary data so the ablation does not cost you the model. The full method and the measurement setup are on the main model card.

The multi-token-prediction head is not in these files. It only ever fed self-speculative decoding, and the release weights ship without it, so nothing is lost for normal generation.

Files

file	quant	size	pick this if
`qwen3.5-0.8b-manumit-v2-Q4_K_M.gguf`	Q4_K_M	0.5 GB	smallest, runs almost anywhere
`qwen3.5-0.8b-manumit-v2-Q5_K_M.gguf`	Q5_K_M	0.6 GB	a bit larger, a bit closer to full precision
`qwen3.5-0.8b-manumit-v2-Q8_0.gguf`	Q8_0	0.8 GB	closest to the original weights

If you are not sure, take Q4_K_M. Move up to Q5_K_M or Q8_0 if you have the memory and want the last bit of quality.

Numbers

Refusal is the keyword refusal rate on held-out harmful prompts, AdvBench-test and JailbreakBench. Ability is MMLU-Pro at n=500, base measured the same way.

	this model	base
AdvBench refusal	0.0%	high
JailbreakBench refusal	0.0%	high
MMLU-Pro	15.2%	12.3%

Refusal is essentially gone and MMLU-Pro landed at or above the base. That is the result worth having.

Run it

llama.cpp:

llama-cli -m qwen3.5-0.8b-manumit-v2-Q4_K_M.gguf -p "Your prompt here"

Ollama:

ollama run hf.co/yethdev/qwen3.5-0.8b-manumit-v2-GGUF:Q4_K_M

Stated plainly

There is no safety layer left and no guard model watching the output. Whatever you generate is yours to answer for, and you still have to follow the law and the base model's terms. manumit takes the refusal behaviour out, it does not put anything back.