Instructions to use cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4 with Transformers:

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

pipe = pipeline("image-text-to-text", model="cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4")
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 AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4")
model = AutoModelForMultimodalLM.from_pretrained("cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4")
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?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4",
		"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/cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4

SGLang

How to use cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4 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 "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4" \
    --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": "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4",
		"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 "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4" \
        --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": "cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4",
		"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"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4 with Docker Model Runner:
```
docker model run hf.co/cyankiwi/Qwen3.5-27B-AWQ-BF16-INT4
```

Model size (28GB) is abnormal for an INT4 model

by S567901 - opened Feb 25

Discussion

S567901

Feb 25

Thanks for your work on quantizing this model! However, I noticed that the total size of the .safetensors files is around 28 GB.Or maybe I've misunderstood something ?

A 28GB model cannot be loaded on a standard 24GB VRAM GPU and defeats the purpose of INT4 quantization. Could you please double-check the export script and consider re-uploading a packed version?

Thanks again!

azidanit

Feb 25

looks like it skip some layer (leave as bf16)

would be great if we got recipe :D
thanks cyankiwi!

cpatonn

cyankiwi org Feb 26

Thank you for your interest and using the model!

And yes, it is due to leaving linear attention layers at BF16 to achieve a more optimal accuracy, as linear attention is more prone to quantization errors. And as linear attention accounts for a huge portion of the model weights, leaving it at BF16 leads to a large memory size of the quantized model.

I also release cyankiwi/Qwen3.5-27B-AWQ-4bit, where linear attention is also quantized :)

BingoBird

Mar 2

Your work is getting more popular - recent reddit post generated 500+ upvotes.

Great job.

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