Instructions to use Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57 with Transformers:

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

pipe = pipeline("image-text-to-text", model="Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57")
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("Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57")
model = AutoModelForMultimodalLM.from_pretrained("Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57", 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 Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57",
		"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/Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57

SGLang

How to use Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57 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 "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57" \
    --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": "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57",
		"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 "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57" \
        --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": "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57",
		"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 Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57 with Docker Model Runner:
```
docker model run hf.co/Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57
```

MindCube — Plain CogMap + Free-Form Reasoning (checkpoint-57)

Full-parameter SFT of Qwen/Qwen2.5-VL-3B-Instruct on the plain_cgmap_ffr_out setting of MindCube: the model first emits a plain cognitive map of the scene, then free-form reasoning, then the answer.

This repo holds the checkpoint-57 snapshot of that run.

📄 Paper: Spatial Mental Modeling from Limited Views (arXiv:2506.21458)
🏠 Project page: https://mind-cube.github.io/
📦 Dataset: MLL-Lab/MindCube

Training setup


Base model	`Qwen/Qwen2.5-VL-3B-Instruct`
Task / dataset	`plain_cgmap_ffr_out`
Learning rate	1e-5
Epochs	3
Max sequence length	8192
Image pixels	min 784 / max 90000
Precision	bfloat16

Usage

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor

model_id = "Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_id, torch_dtype="bfloat16", device_map="auto"
)
processor = AutoProcessor.from_pretrained(model_id, min_pixels=784, max_pixels=90000)

Inference follows the standard Qwen2.5-VL multi-image chat format; see the MindCube repo for the exact prompts and evaluation scripts used for this setting.

Citation

@article{yin2025mindcube,
  title   = {Spatial Mental Modeling from Limited Views},
  author  = {Yin, Baiqiao and Wang, Qineng and Zhang, Pingyue and Zhang, Jianshu and
             Wang, Kangrui and Wang, Zihan and Zhang, Jieyu and Chandrasegaran, Keshigeyan and
             Liu, Han and Krishna, Ranjay and Xie, Saining and Li, Manling and
             Wu, Jiajun and Fei-Fei, Li},
  journal = {arXiv preprint arXiv:2506.21458},
  year    = {2025}
}

Downloads last month: 11

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57

Base model

Qwen/Qwen2.5-VL-3B-Instruct

Finetuned

(847)

this model

Dataset used to train Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57

Paper for Inevitablevalor/MindCube-plain-cgmap-ffr-out-ckpt57

Spatial Mental Modeling from Limited Views

Paper • 2506.21458 • Published Jun 26, 2025 • 14