Instructions to use openbmb/NOSA-8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use openbmb/NOSA-8B with Transformers:

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

pipe = pipeline("text-generation", model="openbmb/NOSA-8B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("openbmb/NOSA-8B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("openbmb/NOSA-8B", trust_remote_code=True)
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

vLLM

How to use openbmb/NOSA-8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "openbmb/NOSA-8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/NOSA-8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/openbmb/NOSA-8B

SGLang

How to use openbmb/NOSA-8B 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 "openbmb/NOSA-8B" \
    --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": "openbmb/NOSA-8B",
		"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 "openbmb/NOSA-8B" \
        --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": "openbmb/NOSA-8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use openbmb/NOSA-8B with Docker Model Runner:
```
docker model run hf.co/openbmb/NOSA-8B
```

nielsr HF Staff commited on Jan 30

Commit

a32f39d

verified ·

1 Parent(s): 31e8b22

Add pipeline tag and library name

Browse files

This PR improves the model card by adding relevant metadata to enhance discoverability and Hub integration:
- `pipeline_tag: text-generation`: Correctly categorizes the model for text generation tasks.
- `library_name: transformers`: Enables the "Use in Transformers" button, as the model is compatible with the library via `trust_remote_code=True`.

Additionally, I have:
- Fixed the links in the "Models" table to ensure they are clickable.
- Added a direct link to the research paper on the Hub.

Files changed (1) hide show

README.md +10 -10

README.md CHANGED Viewed

@@ -1,11 +1,14 @@
 ---
-license: apache-2.0
 datasets:
 - openbmb/InfLLM-V2-data-5B
 language:
 - en
 - zh
 ---
 <div align="center">
 <h1>NOSA: Native and Offloadable Sparse Attention</h1>
@@ -33,26 +36,23 @@ language:
 **NOSA** is a trainable sparse attention mechanism designed for KV-cache offloading with an explicit locality constraint, paired with an inference system (**NOSI**) to realize its efficiency. It improves long-context/long-generation quality over prior offloading baselines while boosting decoding throughput by up to **5.04×** vs **FullAttn**, **1.92×** vs **InfLLMv2**, and **1.83×** vs **ShadowKV** on **1B/3B/8B** LLMs.
 ## Models
-We train 1B, 3B, and 8B models  FullAttn, InfLLMv2, DMA, and NOSA, resulting in a total of 12 models. The following models have been released on Hugging Face.
 |Model|Link|
 |:-:|:-:|
-|NOSA-1B | [NOSA-1B](huggingface.co/openbmb/NOSA-1B) |
-|NOSA-3B | [NOSA-3B](huggingface.co/openbmb/NOSA-3B) |
-|NOSA-8B | [NOSA-8B](huggingface.co/openbmb/NOSA-8B) |
 Please reach out to us if additional baseline models (FullAttn, InfLLMv2, or DMA) are needed. You may open an issue or contact us directly via email (our email addresses are provided in the paper).
 ## Citation
-```
 @article{huang2025nosa,
   title={NOSA: Native and Offloadable Sparse Attention},
   author={Huang, Yuxiang and Wang, Pengjie and Han, Jicheng and Zhao, Weilin and Su, Zhou and Sun, Ao and Lyu, Hongya and Zhao, Hengyu and Wang, Yudong and Xiao, Chaojun and Han, Xu and Liu, Zhiyuan},

 ---
 datasets:
 - openbmb/InfLLM-V2-data-5B
 language:
 - en
 - zh
+license: apache-2.0
+pipeline_tag: text-generation
+library_name: transformers
 ---
 <div align="center">
 <h1>NOSA: Native and Offloadable Sparse Attention</h1>
 **NOSA** is a trainable sparse attention mechanism designed for KV-cache offloading with an explicit locality constraint, paired with an inference system (**NOSI**) to realize its efficiency. It improves long-context/long-generation quality over prior offloading baselines while boosting decoding throughput by up to **5.04×** vs **FullAttn**, **1.92×** vs **InfLLMv2**, and **1.83×** vs **ShadowKV** on **1B/3B/8B** LLMs.
+This repository contains the weights for the NOSA-3B model. For more details, please refer to the paper: [NOSA: Native and Offloadable Sparse Attention](https://huggingface.co/papers/2510.13602).
 ## Models
+We train 1B, 3B, and 8B models (FullAttn, InfLLMv2, DMA, and NOSA), resulting in a total of 12 models. The following models have been released on Hugging Face:
 |Model|Link|
 |:-:|:-:|
+|NOSA-1B | [NOSA-1B](https://huggingface.co/openbmb/NOSA-1B) |
+|NOSA-3B | [NOSA-3B](https://huggingface.co/openbmb/NOSA-3B) |
+|NOSA-8B | [NOSA-8B](https://huggingface.co/openbmb/NOSA-8B) |
 Please reach out to us if additional baseline models (FullAttn, InfLLMv2, or DMA) are needed. You may open an issue or contact us directly via email (our email addresses are provided in the paper).
 ## Citation
+```bibtex
 @article{huang2025nosa,
   title={NOSA: Native and Offloadable Sparse Attention},
   author={Huang, Yuxiang and Wang, Pengjie and Han, Jicheng and Zhao, Weilin and Su, Zhou and Sun, Ao and Lyu, Hongya and Zhao, Hengyu and Wang, Yudong and Xiao, Chaojun and Han, Xu and Liu, Zhiyuan},