Instructions to use INCModel/Qwen3-30B-A3B-12L-MXFP8-test with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use INCModel/Qwen3-30B-A3B-12L-MXFP8-test with Transformers:

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

pipe = pipeline("text-generation", model="INCModel/Qwen3-30B-A3B-12L-MXFP8-test")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("INCModel/Qwen3-30B-A3B-12L-MXFP8-test")
model = AutoModelForCausalLM.from_pretrained("INCModel/Qwen3-30B-A3B-12L-MXFP8-test", 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 INCModel/Qwen3-30B-A3B-12L-MXFP8-test with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "INCModel/Qwen3-30B-A3B-12L-MXFP8-test"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "INCModel/Qwen3-30B-A3B-12L-MXFP8-test",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/INCModel/Qwen3-30B-A3B-12L-MXFP8-test

SGLang

How to use INCModel/Qwen3-30B-A3B-12L-MXFP8-test 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 "INCModel/Qwen3-30B-A3B-12L-MXFP8-test" \
    --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": "INCModel/Qwen3-30B-A3B-12L-MXFP8-test",
		"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 "INCModel/Qwen3-30B-A3B-12L-MXFP8-test" \
        --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": "INCModel/Qwen3-30B-A3B-12L-MXFP8-test",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use INCModel/Qwen3-30B-A3B-12L-MXFP8-test with Docker Model Runner:
```
docker model run hf.co/INCModel/Qwen3-30B-A3B-12L-MXFP8-test
```

Qwen3-30B-A3B-12L-MXFP8-test

This is a 12-layer test checkpoint derived from a Qwen3-30B-A3B AutoRound MXFP8 checkpoint. It is intended for vLLM loading and inference tests, including MXFP8 linear and fused MoE coverage. It is not intended for quality evaluation or production use.

Configuration

Architecture: Qwen3MoeForCausalLM
Transformer layers: 12 (the first 12 layers of the source checkpoint)
Weight format: MXFP8, group size 32
Activation format: dynamic MXFP8, group size 32
Packing format: auto_round:llm_compressor
AutoRound version: 0.14.2

The tokenizer, embedding, final normalization, and language model head are retained. Quantization metadata is trimmed to the retained layers.

vLLM test

This checkpoint is prepared for the following vLLM test model identifier:

INCModel/Qwen3-30B-A3B-12L-MXFP8-test

The test uses eager execution and generates eight tokens from the prompt The capital of France is.

Downloads last month: 19

Safetensors

Model size

8B params

Tensor type

BF16

F8_E4M3

Model tree for INCModel/Qwen3-30B-A3B-12L-MXFP8-test

Base model

Qwen/Qwen3-30B-A3B-Base

Finetuned

Qwen/Qwen3-30B-A3B

Quantized

(144)

this model