Instructions to use Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview with Transformers:

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

pipe = pipeline("text-generation", model="Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview")
model = AutoModelForCausalLM.from_pretrained("Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview", 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 Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview

SGLang

How to use Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview 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 "Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview" \
    --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": "Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview",
		"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 "Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview" \
        --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": "Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview with Docker Model Runner:
```
docker model run hf.co/Alibaba-Apsara/DASD-30B-A3B-Thinking-Preview
```

Can somebody please make GGUF or MLX version quickly?

by Narutoouz - opened Jan 14

Discussion

Narutoouz

Jan 14

•

edited Jan 14

Chinese AI models are pushing the AI Industry forward. Though I must say it was done due to the data obtained from GPT OSS 120b model from OpenAI. As an AI student, more over a consumer of AI, thanks for democratizing this by open sourcing this, Alibaba-Aspara. Again, thanks to OpenAI for Opensourcing GPT OSS 120b, it is still an excellent model, even though it been over 6 months since it was released. Sam Altman's promise to open source was real and thanks for that.

The benchamarks are already SOTA in the preview version compared to other models in this parameter range, I can't wait to see the instruct and thinking variants of this model. Waiting excitely for those!

owao

Jan 14

Narutoouz:

you can easily produce a GGUF version, without even needing to download the model and run the command by yourself, using https://huggingface.co/spaces/ggml-org/gguf-my-repo as Qwen3MoeForCausalLM (you can find the architecture in config.json) has already been supported for some times now
I don't know how you missed it, but this is indeed a thinking model, as per the README and the name of the model!
beware that this is currently a text only model, it hasn't been trained for tool usage, so even if the base model was I guess you shouldn't expect anything

MegatronMyLeader

Jan 27

•

edited Jan 27

Chinese AI models were pushing the frontier long before GPT OSS was released.

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