Instructions to use ZJUVAI/GenesisGeo-2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ZJUVAI/GenesisGeo-2B with Transformers:

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

pipe = pipeline("image-text-to-text", model="ZJUVAI/GenesisGeo-2B")
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 AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("ZJUVAI/GenesisGeo-2B")
model = AutoModelForMultimodalLM.from_pretrained("ZJUVAI/GenesisGeo-2B", device_map="auto")
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?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ZJUVAI/GenesisGeo-2B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ZJUVAI/GenesisGeo-2B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ZJUVAI/GenesisGeo-2B",
		"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/ZJUVAI/GenesisGeo-2B

SGLang

How to use ZJUVAI/GenesisGeo-2B 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 "ZJUVAI/GenesisGeo-2B" \
    --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": "ZJUVAI/GenesisGeo-2B",
		"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 "ZJUVAI/GenesisGeo-2B" \
        --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": "ZJUVAI/GenesisGeo-2B",
		"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 ZJUVAI/GenesisGeo-2B with Docker Model Runner:
```
docker model run hf.co/ZJUVAI/GenesisGeo-2B
```

GenesisGeo-2B

📃 Paper • 📚 GitHub

This model is specialized in automated geometric theorem proving, capable of proposing auxiliary constructions to solve challenging geometry problems. It forms the neural component of the GenesisGeo project—a neuro-symbolic system that combines a vision-language model with the DDAR symbolic deduction engine.

It is built upon Qwen3-VL-2B-Instruct and trained on synthetic geometry data for predicting auxiliary constructions from formal problem statements, with or without diagrams.

Model Description

Architecture: Vision-language model
Base Model: Qwen3-VL-2B-Instruct
Training Data: One million synthetic multimodal geometry records for pretraining, followed by auxiliary construction data for supervised fine-tuning
Training: Pretraining on complete solution records containing formal problems, diagrams, auxiliary constructions, and proof traces; supervised fine-tuning for auxiliary construction prediction with and without diagrams
Fine-tuning: The vision encoder is frozen, while the language model and visual aligner are trained
Purpose: Proposing auxiliary constructions in geometric proofs within a neuro-symbolic reasoning loop

Performance

The integrated GenesisGeo neuro-symbolic system achieves the following paper-reported results, also listed in the GitHub README:

Variant	IMO-30	IMO-95	HAGeo-409
Text	28/30	59/95	270/409
Vision + Text	29/30	63/95	278/409

Paper-reported results with a 32 × 512 × 4 search budget and a 60-minute time limit per problem. These results measure the complete system, including symbolic deduction and proof search.

Usage

Input and output

The model takes a formal geometry problem in the predicate DSL, enclosed in <problem>...</problem>, optionally accompanied by a diagram. It predicts auxiliary constructions in an <aux>...</aux> block.

Use the problem serialization and prompting implemented in the GenesisGeo repository. Generated constructions are checked by the symbolic deduction engine as part of the complete proof-search system.

For proof search and benchmark evaluation, follow the evaluation instructions in the GenesisGeo repository.

License

Apache License 2.0. The model is derived from Qwen3-VL-2B-Instruct.

Downloads last month: -

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for ZJUVAI/GenesisGeo-2B

Base model

Qwen/Qwen3-VL-2B-Instruct

Finetuned

(269)

this model

Paper for ZJUVAI/GenesisGeo-2B

GenesisGeo: Technical Report

Paper • 2509.21896 • Published Sep 26, 2025