Instructions to use bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE with Transformers:

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

pipe = pipeline("image-text-to-text", model="bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE")
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("bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE")
model = AutoModelForMultimodalLM.from_pretrained("bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE", 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 bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE",
		"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/bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE

SGLang

How to use bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE 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 "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE" \
    --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": "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE",
		"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 "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE" \
        --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": "bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE",
		"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 bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE with Docker Model Runner:
```
docker model run hf.co/bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE
```

Qwen3.8-Flash-Next W4A8 (int4 PLE sidecar included)

GPTQ W4A8 quantization of the of Qwen3.8-Flash-Next (qwen4_exp architecture): 4-bit weights, int8 activations, symmetric, group size 128, desc_act=false, true-sequential, activation-group-aware. Quantized with gptqmodel 7.6.0 on a custom calibration corpus. Tokenizer, chat template and configs are carried over unchanged from the original checkpoint.

What changed vs the Original checkpoint

The bf16 PLE n-gram embedding table (95.4 GiB, upstream model-00002-of-00020.safetensors) is removed from the checkpoint. It ships instead as a group-16 int4 PLE sidecar in ple-int4/ (128 shards + META.json, ~30 GB): group-16 RTN, fp16 scales, fixed zero-point 8, low-nibble-first — byte-compatible with the PLE CPU-offload worker contract.
The small per-layer PLE projection tensors remain in-line in the model shards (they were never part of the externalized table).
19 safetensors shards, ~75 GB payload.

Scope of quantization

All 74,028 language-model linears are W4A8 (83.8% of payload bytes are packed 4-bit weights

2.6% fp16 scales). Left at bf16, by design (~16% of payload):

MTP drafter module (~5.2 GB) — kept full precision for draft acceptance;
embed_tokens + lm_head (~2.5 GB, 248320x2560 each);
vision tower + merger (~0.9 GB);
norms/biases and the 9 small PLE projection tensors.

Serving

Coming Soon

License

Inherits the original release's license: qwen-community-1.0 (see LICENSE).

Credits

Special Thanks to Rob for performing the quantizaiton process.

Downloads last month: -

Safetensors

Model size

129B params

Tensor type

BF16

I32

I64

Model tree for bradharrison/Qwen3.8-Flash-Next-W4A8-INT4-PLE

Base model

Qwen/Qwen3.8-Flash-Next

Quantized

(546)

this model