Instructions to use realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6 with Transformers:

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

pipe = pipeline("image-text-to-text", model="realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6")
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("realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6")
model = AutoModelForMultimodalLM.from_pretrained("realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6", device_map="auto")
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 realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6",
		"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/realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6

SGLang

How to use realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6 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 "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6" \
    --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": "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6",
		"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 "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6" \
        --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": "realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6",
		"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 realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6 with Docker Model Runner:
```
docker model run hf.co/realderpz/Swift-1.5-Qwen3.8-27B-PARO-MXFP6
```

realderpz commited on 7 days ago

Commit

6346316

verified ·

1 Parent(s): dc49e38

Link ParoQuant paper

Browse files

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -16,7 +16,7 @@ tags:
 # Swift-1.5-Qwen3.8-27B-PARO-MXFP6
-ParoQuant MXFP6 quant of [`ukisai/Swift-1.5-Qwen3.8-27b`](https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b). Unofficial.
 - Scheme: MXFP6 E2M3 weights (W6A8), ParoQuant rotations (krot 8, group 128), `quant_method: paroquant_mxfp6`
 - Method: rotations from [`z-lab/Qwen3.8-27B-PARO`](https://huggingface.co/z-lab/Qwen3.8-27B-PARO) (trained on base Qwen3.8-27B, frozen), then round-to-nearest to MXFP6. Stage-2 fine-tune skipped for time.

 # Swift-1.5-Qwen3.8-27B-PARO-MXFP6
+[ParoQuant](https://arxiv.org/abs/2511.10645) MXFP6 quant of [`ukisai/Swift-1.5-Qwen3.8-27b`](https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b). Unofficial.
 - Scheme: MXFP6 E2M3 weights (W6A8), ParoQuant rotations (krot 8, group 128), `quant_method: paroquant_mxfp6`
 - Method: rotations from [`z-lab/Qwen3.8-27B-PARO`](https://huggingface.co/z-lab/Qwen3.8-27B-PARO) (trained on base Qwen3.8-27B, frozen), then round-to-nearest to MXFP6. Stage-2 fine-tune skipped for time.