Instructions to use TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast with Transformers:

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

pipe = pipeline("image-text-to-text", model="TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast")
model = AutoModelForMultimodalLM.from_pretrained("TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast", 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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast",
		"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/TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast

SGLang

How to use TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast 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 "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast" \
    --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": "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast",
		"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 "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast" \
        --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": "TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast",
		"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 TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast with Docker Model Runner:
```
docker model run hf.co/TyroneNel/Swift-Qwen3.8-27B-Uncensored-W4A16-fast
```

Swift-Qwen3.8-27B-Uncensored-W4A16-fast / NOTICE

TyroneNel

Upload Swift-Qwen3.8-27B-Uncensored-W4A16-fast

f1944d3 verified 18 days ago

Raw History Blame Contribute Delete

5.1 kB

	Swift-Qwen3.8-27B-Uncensored-W4A16 and Swift-Qwen3.8-27B-Uncensored-W4A16-fast
	Quantization and serving preparation: Copyright 2026 TyroneNel

	This work incorporates the Swift Contribution, licensed under the Swift Open
	License v1.0 (see LICENSE):

	Swift-Qwen3.8-27B, Copyright 2026 UkisAI.
	https://huggingface.co/ukisai/Swift-Qwen3.8-27b
	Derivative of Qwen3.8-27B, Copyright 2026 Alibaba Cloud, Apache License 2.0.

	Provenance chain:

	1. Base Model: Qwen/Qwen3.8-27B
	https://huggingface.co/Qwen/Qwen3.8-27B
	Copyright 2026 Alibaba Cloud. Licensed under the Apache License,
	Version 2.0. See LICENSE-APACHE-2.0.

	2. Swift Contribution: ukisai/Swift-Qwen3.8-27b
	https://huggingface.co/ukisai/Swift-Qwen3.8-27b
	Copyright 2026 UkisAI. Licensed under the Swift Open License v1.0.
	See LICENSE. A reasoning-efficiency LoRA adapter merged into the Base
	Model weights (see UkisAI's NOTICE and model card).

	3. Uncensoring: d0xin/Swift-Qwen3.8-27B-Uncensored-BF16
	https://huggingface.co/d0xin/Swift-Qwen3.8-27B-Uncensored-BF16
	A Derivative Work of (2): rank-1 directional residual-stream ablation
	at layer 38 (131 of 1,199 tensors modified, vision tensors unchanged).
	Methodology and validation: ABLITERATION.json,
	STRUCTURAL_VALIDATION.json and INTELLIGENCE_VALIDATION.json in that
	repository.

	4. This work: a Derivative Work of (3). Changes (Apache License 2.0,
	Section 4(b), and Swift Open License v1.0 change notice):
	- model-*.safetensors, model_extra_tensors.safetensors,
	model.safetensors.index.json: the decoder linear layers of (3) are
	quantized to 4-bit integers (W4A16, group size 128, symmetric) with
	Intel AutoRound (128 calibration samples x 2048 tokens). The vision
	tower and the GatedDeltaNet in_proj_a/in_proj_b layers stay BF16.
	embed_tokens is quantized to int8 (group 128, symmetric).
	Swift-Qwen3.8-27B-Uncensored-W4A16: lm_head and the MTP module's
	linear layers are quantized to int8 (group 128, symmetric).
	Swift-Qwen3.8-27B-Uncensored-W4A16-fast: lm_head and the MTP
	module's linear layers are quantized to int4 (group 128,
	symmetric) with GPTQ, calibrated on hidden states captured from
	this model's own generations.
	Both: a 40,960-row draft head (mtp.draft_lm_head.*) is sliced
	from the quantized lm_head for MTP speculative decoding.
	- config.json, quantization_config.json: quantization metadata
	added. generation_config.json: identical to (3).
	- mtp_draft_vocab_ids.pt, draft_vocab_ids.json: added (the token ids
	of the draft head, counted over this model's own outputs).
	- chat_template.jinja: two changes to the Qwen3.8 template.
	(a) Reasoning-effort translation: the OpenAI names "minimal",
	"high" and "max" map to the template's low / xhigh levels, and an
	unknown value no longer raises an error.
	(b) Tool calls whose arguments are a JSON string instead of an
	object render as one <parameter=arguments> block instead of
	failing.
	- README.md: replaced.
	- tokenizer.json, tokenizer_config.json, preprocessor_config.json,
	processor_config.json: re-saved from (3) by transformers 5 (the
	chat template moved to chat_template.jinja); same vocabulary,
	merges and special tokens.
	Tools: the syv-ai/HyperQwen pipeline (run_quant.sh, prepare/,
	drafter/), https://github.com/syv-ai/HyperQwen.

	Per Section 4(e) of the Swift Open License v1.0, this distribution includes
	a copy of the Base Model License (LICENSE-APACHE-2.0) next to the Swift Open
	License v1.0 (LICENSE), because this work incorporates portions of the Base
	Model.

	Attribution notices from the NOTICE file of ukisai/Swift-Qwen3.8-27b,
	reproduced as Section 4(d) of the Swift Open License v1.0 requires:

	--------------------------------------------------------------------------
	Swift-Qwen3.8-27B
	Copyright 2026 UkisAI

	UkisAI's contribution (the "Swift Contribution") is licensed under the
	Swift Open License v1.0. See LICENSE.

	This model is a Derivative Work of Qwen3.8-27B
	https://huggingface.co/Qwen/Qwen3.8-27B
	Copyright 2026 Alibaba Cloud
	Licensed under the Apache License, Version 2.0. See LICENSE-APACHE-2.0.

	Changes made by UkisAI (Apache License 2.0, Section 4(b) change notice):
	- model-*.safetensors, model.safetensors.index.json: model weights were
	fine-tuned by UkisAI (LoRA adapter trained by UkisAI and merged into the
	Base Model weights).
	- generation_config.json: added "min_p": 0 and "repetition_penalty": 1.0.
	- README.md: replaced. ukisai-banner.png and swift-speed-demo.mp4 added.
	- All other files (config.json, chat_template.jinja, tokenizer.json,
	tokenizer_config.json, vocab.json, merges.txt, preprocessor_config.json,
	video_preprocessor_config.json) are unmodified from Qwen3.8-27B and
	remain under the Apache License, Version 2.0.
	--------------------------------------------------------------------------