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

Libraries

How to use URSA-MATH/URSA-8B with Transformers:

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

pipe = pipeline("image-text-to-text", model="URSA-MATH/URSA-8B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModelForSeq2SeqLM
model = AutoModelForSeq2SeqLM.from_pretrained("URSA-MATH/URSA-8B", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use URSA-MATH/URSA-8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "URSA-MATH/URSA-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": "URSA-MATH/URSA-8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/URSA-MATH/URSA-8B

SGLang

How to use URSA-MATH/URSA-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 "URSA-MATH/URSA-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": "URSA-MATH/URSA-8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "URSA-MATH/URSA-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": "URSA-MATH/URSA-8B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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

URSA-8B

URSA-8B is the first small-sized MLLM specifically focused on Chain-of-thought multimodal mathematical reasoning.

Installation

from huggingface_hub import snapshot_download

repo_id = "URSA-MATH/URSA-8B"
local_dir = YOUR_LOCAL_PATH  

snapshot_path = snapshot_download(
    repo_id=repo_id,
    local_dir=local_dir,
    revision="main", 
    cache_dir=None, 
)

Inference

We have adapted vLLM for URSA-8B. Please refer to the GitHub repository for quick inference implementation.

Citation

If you find our paper, model, or data helpful, please give this repo a star 🌟 and cite our article ✏️.

@article{luo2025ursa,
  title={URSA: Understanding and Verifying Chain-of-thought Reasoning in Multimodal Mathematics},
  author={Luo, Ruilin and Zheng, Zhuofan and Wang, Yifan and Yu, Yiyao and Ni, Xinzhe and Lin, Zicheng and Zeng, Jin and Yang, Yujiu},
  journal={arXiv preprint arXiv:2501.04686},
  year={2025}
}

Downloads last month: 8

Safetensors

Model size

8B params

Tensor type

F32

Model tree for URSA-MATH/URSA-8B

Finetunes

1 model

Dataset used to train URSA-MATH/URSA-8B

Paper for URSA-MATH/URSA-8B

URSA: Understanding and Verifying Chain-of-thought Reasoning in Multimodal Mathematics

Paper • 2501.04686 • Published Jan 8, 2025 • 53