Instructions to use cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit with Transformers:

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

pipe = pipeline("text-generation", model="cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit")
model = AutoModelForCausalLM.from_pretrained("cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit"
# 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-Next-80B-A3B-Instruct-AWQ-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit

SGLang

How to use cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit 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-Next-80B-A3B-Instruct-AWQ-4bit" \
    --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-Next-80B-A3B-Instruct-AWQ-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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-Next-80B-A3B-Instruct-AWQ-4bit" \
        --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-Next-80B-A3B-Instruct-AWQ-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit with Docker Model Runner:
```
docker model run hf.co/cyankiwi/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit
```

gibberish still persists?

by Geximus - opened Sep 16, 2025

Discussion

Geximus

Sep 16, 2025

And i have problems - model answers is hallucinating - i have very emotional answers with different themes mixed out but i have standart assistent promt. Do you have same problem ?

cpatonn

cyankiwi org Sep 16, 2025

Thank you for your input! This might be the result of the active paramsshared_expert being quantized into 4bit, which is due to vllm not being able to load load the model if shared_expert is not quantized.

My apologies for this, but there will be an update to the model soon in the next few days to improve the model accuracy.

Geximus

Sep 16, 2025

thanks for you work looking forward to use your AWQ instruct model!

Geximus

Sep 18, 2025

Thank you for your input! This might be the result of the active paramsshared_expert being quantized into 4bit, which is due to vllm not being able to load load the model if shared_expert is not quantized.

My apologies for this, but there will be an update to the model soon in the next few days to improve the model accuracy.

Hey! Any updates ? What is your plans?

cpatonn

cyankiwi org Sep 19, 2025

Hi @Geximus , thank you for your interest. There already has been a vllm update, which should improve every qwen3next models. Regarding this specific model, the update will be available in the next 36 hours.

cpatonn

cyankiwi org Sep 20, 2025

Hi @Geximus , thank you for downloading my models :)

I have just released a recent update with active params being in bf16 and not quantized. It should match the original bf16 model more closely.

Please let me know if the problem still persists.

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