Instructions to use Sehyo/Qwen3.5-122B-A10B-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Sehyo/Qwen3.5-122B-A10B-NVFP4 with Transformers:

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

pipe = pipeline("image-text-to-text", model="Sehyo/Qwen3.5-122B-A10B-NVFP4")
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("Sehyo/Qwen3.5-122B-A10B-NVFP4")
model = AutoModelForMultimodalLM.from_pretrained("Sehyo/Qwen3.5-122B-A10B-NVFP4", 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 Sehyo/Qwen3.5-122B-A10B-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Sehyo/Qwen3.5-122B-A10B-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Sehyo/Qwen3.5-122B-A10B-NVFP4",
		"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/Sehyo/Qwen3.5-122B-A10B-NVFP4

SGLang

How to use Sehyo/Qwen3.5-122B-A10B-NVFP4 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 "Sehyo/Qwen3.5-122B-A10B-NVFP4" \
    --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": "Sehyo/Qwen3.5-122B-A10B-NVFP4",
		"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 "Sehyo/Qwen3.5-122B-A10B-NVFP4" \
        --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": "Sehyo/Qwen3.5-122B-A10B-NVFP4",
		"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 Sehyo/Qwen3.5-122B-A10B-NVFP4 with Docker Model Runner:
```
docker model run hf.co/Sehyo/Qwen3.5-122B-A10B-NVFP4
```

MTP Added - Re-download

by Sehyo - opened Mar 1

Discussion

Sehyo

Owner Mar 1

Have added MTP layers / support.
Please redownload :)

mtcl

Mar 1

Alright doing it now, can you please add a sglang tool support startup command?

scottgl

Mar 2

•

edited Mar 2

Thanks! I also uploaded the MTP weights here if needed: https://huggingface.co/scottgl/Qwen3.5-122B-A10B-MTP-NVFP4 (NVFP4 quantized).

Sebesky

Mar 2

Has anyone got this to work with vllm? I used the nightly build. Im able to serve the model, but performance is suboptimal and during decode it gets caught in infinite reasoning loops. Currently using the shell script below on a GB10 and OpenWebUI. If anybody has had success please share your exact configuration, it would help me greatly.

#!/bin/bash

Configuration

CONTAINER_NAME="vllm_instance"
VLLM_IMAGE="vllm/vllm-openai:cu130-nightly" # Adjust image tag if you compiled locally

Use vLLM nightly docker until 0.17.0 is released.

docker run --gpus all
-p 8000:8000
--ipc=host
-v ~/.cache/huggingface:/root/.cache/huggingface
--name $CONTAINER_NAME
-e HF_TOKEN=$HF_TOKEN

$VLLM_IMAGE $MODEL
--tensor-parallel-size 1
--gpu-memory-utilization 0.9
--speculative-config '{"method": "mtp", "num_speculative_tokens": 2}'
--enable-auto-tool-choice
--tool-call-parser qwen3_coder
--reasoning-parser qwen3
--max-model-len 131072 \

Sehyo

Owner Mar 2

For method you should use qwen3_next_mtp and not "mtp". num_speculative_tokens works best with 3 in my testing.

unoid

Mar 3

@sebesky ,re-pull vllm:nightly, or cu130-nightly, MTP tool parser was fixed today march 3rd.

scottgl

Mar 5

•

edited Mar 5

Did you try different backends? Which did you get the best performance with?

Unfortunately this doesn't have all the GB10 specific workaround patches for NVFP4 to work well. It works, but I'm only getting 10-15 tokens per sec with cu130-nightly.

Do you know if the nightly is built from https://github.com/vllm-project/vllm? Or from somewhere else?

dfk1352

May 14

This is the first model I got that has MTP support actually working, thank you for making this. Getting around 20 ~ 30 t/s on GB10, nothing crazy but sufficient for agentic workflows.

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