Instructions to use DavidAU/Qwen3-30B-A6B-16-Extreme with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DavidAU/Qwen3-30B-A6B-16-Extreme with Transformers:

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

pipe = pipeline("text-generation", model="DavidAU/Qwen3-30B-A6B-16-Extreme")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DavidAU/Qwen3-30B-A6B-16-Extreme")
model = AutoModelForCausalLM.from_pretrained("DavidAU/Qwen3-30B-A6B-16-Extreme", 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 DavidAU/Qwen3-30B-A6B-16-Extreme with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DavidAU/Qwen3-30B-A6B-16-Extreme"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DavidAU/Qwen3-30B-A6B-16-Extreme",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/DavidAU/Qwen3-30B-A6B-16-Extreme

SGLang

How to use DavidAU/Qwen3-30B-A6B-16-Extreme 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 "DavidAU/Qwen3-30B-A6B-16-Extreme" \
    --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": "DavidAU/Qwen3-30B-A6B-16-Extreme",
		"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 "DavidAU/Qwen3-30B-A6B-16-Extreme" \
        --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": "DavidAU/Qwen3-30B-A6B-16-Extreme",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use DavidAU/Qwen3-30B-A6B-16-Extreme with Docker Model Runner:
```
docker model run hf.co/DavidAU/Qwen3-30B-A6B-16-Extreme
```

Please make an A12B and A24B up to whatever makes sense to you

by TimeLordRaps - opened Jun 8, 2025

Discussion

TimeLordRaps

Jun 8, 2025

This is the easily the longest traces Ive seen from a model of anywhere near this active size with the model clear reasoning steps. It's also seemingly responding fine to adult entertainment industry questions, which I find useful, not sure how more or different experts would fair, would be interesting to see if you could prune experts, or abliterate only parts of certain experts, based on their expertise, so to speak of course. Please, please make more, or better yet, an interactive space for abliteration/expert activation via standard methods, or advanced ones you can think of that I can't. I think you seem most poised for developing something like this, if you dont have it for yourself already, which would surprise me, depending on the degree to which you already have something like this implemented on your end. You make such great models, and make it feel like an art.

DavidAU

Owner Jun 8, 2025

I hear you ;

RE: ADult / Pruned
There is an adult version (30B-A3B), as well as a pruned version (16B-A3B) :

"30B-A3B-abliterated-erotic"
and
"Qwen3-16B-A3B"

You may want to contact the maker of the pruned version directly; it is also somewhat discussed at his repo.
There is also an abliterated version of the "16B-A3B".

Currently building a Qwen3 MOE "from scratch" (using Mergekit) is not possible yet, I am waiting on this currently.
Hopefully this will be soon... which will allow 0.6Bs (and other sizes) to make fully custom Qwen3 MOEs.

I have tested Qwen3s with the Brainstorm adapter, it works - but needs adjustments before release.

The current batch of 30B-A3Bs / 16B-A3Bs -> I can do some work with them, just have not started this yet.

I have already made 14Bs into 22B (Qwen3) with reasoning intact here:
https://huggingface.co/DavidAU/Qwen3-The-Xiaolong-Josiefied-Omega-Directive-22B-uncensored-abliterated-GGUF
(3 versions)

As well as 18 Qwen 3 - 4Bs , here:
https://huggingface.co/collections/DavidAU/d-au-qwen3-4b-fiction-on-fire-merge-pruning-series-683ff98167c6fae4bec5d1b6

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