Instructions to use stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ with Transformers:

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

pipe = pipeline("text-generation", model="stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ")
model = AutoModelForCausalLM.from_pretrained("stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ", 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 stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ

SGLang

How to use stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ 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 "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ" \
    --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": "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ",
		"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 "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ" \
        --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": "stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ with Docker Model Runner:
```
docker model run hf.co/stelterlab/Qwen3-30B-A3B-Instruct-2507-AWQ
```

Qwen3.5-35B-A3B AWQ quant planned?

by amidwestnoob - opened Mar 6

Discussion

amidwestnoob

Mar 6

Hey — I'm running your Qwen3-30B-A3B-Instruct-2507-AWQ on dual 3090s
with vLLM and it's been rock solid. Thank you for the clean quant.

Are you planning an AWQ quant of Qwen3.5-35B-A3B? If you are, I'll
wait for yours rather than using another quantizer. If not, no worries
— just want to plan my upgrade path.

Thanks!
— @amidwestnoob

stelterlab

Owner Mar 10

Hi!

Not yet. The last quants I made, I made because no one else had made them at that time. ;-) Like the nemotron.

I think the cyanwiki team did already a AWQ quant. Those are pretty solid. And the are normally faster than me.

I only tested Qwen3.5-35B-A3B yet on a Spark/GB10 as FP8 and "pure" as BF16 on a dual L40 config. The last vLLM versions made trouble with AWQ under Blackwell. There a still a lot of open issues.

I could try a llm-compressor run as soon as I have free GPU resources.

Kind regards, cos

amidwestnoob

Mar 11

Thanks cos — appreciate the quick reply and the context on the Blackwell/AWQ issues. Your nemotron quant is actually the checker in a dual-brain setup on my rig — Qwen3.5 GGUF as the primary builder on one 3090, your nemotron AWQ as an independent fact-checker on the other. Different training lineage catches things the builder misses. Working great. I'll try cyankiwi for the 3.5 AWQ — currently on GGUF via llama.cpp since the AWQ had a VL config loading issue on vLLM 0.17.0. If you do end up running one later I'd be happy to test it on dual 3090s. Cheers.

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