Instructions to use unsloth/Qwen3.6-27B-MTP-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use unsloth/Qwen3.6-27B-MTP-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="unsloth/Qwen3.6-27B-MTP-GGUF")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("unsloth/Qwen3.6-27B-MTP-GGUF", dtype="auto")

llama-cpp-python

How to use unsloth/Qwen3.6-27B-MTP-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="unsloth/Qwen3.6-27B-MTP-GGUF",
	filename="BF16/Qwen3.6-27B-BF16-00001-of-00002.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 unsloth/Qwen3.6-27B-MTP-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama-cli -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama-cli -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

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 unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

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 unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

LM Studio
Jan

vLLM

How to use unsloth/Qwen3.6-27B-MTP-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "unsloth/Qwen3.6-27B-MTP-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3.6-27B-MTP-GGUF",
		"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/unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

SGLang

How to use unsloth/Qwen3.6-27B-MTP-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "unsloth/Qwen3.6-27B-MTP-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3.6-27B-MTP-GGUF",
		"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 images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "unsloth/Qwen3.6-27B-MTP-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3.6-27B-MTP-GGUF",
		"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"
						}
					}
				]
			}
		]
	}'

Ollama
How to use unsloth/Qwen3.6-27B-MTP-GGUF with Ollama:
```
ollama run hf.co/unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
```

Unsloth Studio

How to use unsloth/Qwen3.6-27B-MTP-GGUF 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 unsloth/Qwen3.6-27B-MTP-GGUF 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 unsloth/Qwen3.6-27B-MTP-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

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

How to use unsloth/Qwen3.6-27B-MTP-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

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": "unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use unsloth/Qwen3.6-27B-MTP-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

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 unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Docker Model Runner
How to use unsloth/Qwen3.6-27B-MTP-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL
```

Lemonade

How to use unsloth/Qwen3.6-27B-MTP-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.Qwen3.6-27B-MTP-GGUF-UD-Q4_K_XL

List all available models

lemonade list

the trade off is not good (new update)

#28

by rosspanda0 - opened 19 days ago

Discussion

rosspanda0

19 days ago

•

edited 4 days ago

prompt processing speed is as important as token output speed in coding agent like Claude Code, while decreasing prompt processing by half, reducing available vram so get reduced context length, 1.5x speed up to output to me is actually not an improvement. it's a downgrade.
03-06:
for multiple GPUs users, the tricky part of llama.cpp utilizing MTP is llama.cpp offloads the additional VRAM requirement fully to the last GPU!!! see the post below.

keick

19 days ago

What hardware are you running on? I'm seeing a ~9% drop in PP and a ~20% increase in TG on a single V100 (32GB). I have to be very careful with RAM, the MTB is running two models so it's easy for one of the models to drop to CPU and not notice - so I adjusted context down to compensate on the MTB model. Otherwise the 2nd MTB model drops to system memory and TG drops to ~16.

Also I did notice that I can not run spec-draft-n-max at the recommended (6) - for some reason it's very CPU intensive even when it fits in RAM. However running at (4) is no problem.

Qwen3.6-27B:Q6_K (100k context)
./llama-server -hf unsloth/Qwen3.6-27B-GGUF:Q6_K --temp 0.6 --top-p 0.95 --top-k 20 --port 8001 --host 0.0.0.0 --reasoning off -c 102400 -b 2048 -ub 2048 --flash-attn on --cache-type-k q8_0 --cache-type-v q8_0 --split-mode none --main-gpu 0

Results: pp: 488, tg: 21

Qwen3.6-27B-MTB:Q6_K (60k context)
./llama-server -hf unsloth/Qwen3.6-27B-MTP-GGUF:Q6_K --temp 0.6 --top-p 0.95 --top-k 20 --port 8001 --host 0.0.0.0 --reasoning off -c 61440 -b 2048 -ub 2048 --flash-attn on --cache-type-k q8_0 --cache-type-v q8_0 --split-mode none --main-gpu 0 --spec-type draft-mtp --spec-draft-n-max 4

Results: pp: 447, tg: 26

rosspanda0

18 days ago

6 years old hardwares but pretty good with vanilla version for productivities. I used to have 200000K context length, when I said 1.5 x faster, it's not honest. on my machines, the token speed is improved only a bit , maybe not, while in-taking speed plunge, the whole system become unusable.

hellork

14 days ago

I've been noticing inconsistent results with prompt-processing. Sometimes it's fast. Sometimes it's slow.
Going to try reverting this KV cache reuse regression, since 5/19 https://github.com/ggml-org/llama.cpp/issues/23589

rosspanda0

4 days ago

turn out llama.cpp offload the additional vram requirement for MTP to the LAST GPU, the -ts parameter has to be very carefully tuned, especially big reduction of the vram estimation for LAst GPU. previously slow performance was mainly caused by the cache was not fully offload to last GPU which I didn't notice. Today I retest it, found out this tricky thing. now MTP working as expected.

rosspanda0 changed discussion title from the trade off is not good to the trade off is not good (new update) 4 days ago

shimmyshimmer changed discussion status to closed 2 days ago

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