📥 Model Download |
⚡ Quick Start |
📜 License |
📖 Citation
We provide an efficient MindSpore implementation of JanusPro. This repository is built on the models and code released by DeepSeek. We are grateful for their exceptional work and generous contribution to open source.
2025.02.10: MindSpore implementation of Janus-Pro is released, supporting both multimodal understanding and visual generation on Ascend NPU.
Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling
Janus-Pro is an advanced version of the previous work Janus. Specifically, Janus-Pro incorporates (1) an optimized training strategy, (2) expanded training data, and (3) scaling to larger model size. With these improvements, Janus-Pro achieves significant advancements in both multimodal understanding and text-to-image instruction-following capabilities, while also enhancing the stability of text-to-image generation.
JanusPro is available to the public to support a broader and more diverse range of research within both academic and commercial communities. Please note that the use of this model is subject to the terms outlined in License section. Commercial usage is permitted under these terms.
| Model | Sequence Length | Download |
|---|---|---|
| Janus-Pro-1B | 4096 | 🤗 Hugging Face |
| Janus-Pro-7B | 4096 | 🤗 Hugging Face |
You can download by:
# with revision .safetensors can be downloaded
huggingface-cli download deepseek-ai/Janus-Pro-1B --revision refs/pr/6 --local-dir ckpts/Janus-Pro-1B
huggingface-cli download deepseek-ai/Janus-Pro-7B --revision refs/pr/110 --local-dir ckpts/Janus-Pro-7BThe code is tested in the following environments
| mindspore | ascend driver | firmware | cann tookit/kernel |
|---|---|---|---|
| 2.5.0 | 24.1.0 | 7.35.23 | 8.0.RC3.beta1 |
On the basis of Python >= 3.8 environment, install the necessary dependencies by running the following command:
pip install -e .python inference.py \
--image images/doge.png \
--question "explain this meme"python generation_inference.py \
--prompt "A stunning princess from kabul in red, white traditional clothing, blue eyes, brown hair"For the remote gradio demo, you can run with the following command:
On NPU server:
pip install -e .[gradio]
python demo/app_januspro.pyOn local terminal, run ssh -L 37906:localhost:37906 user_name@server_ip, then open localhost:37906 on the web.
Have Fun!
Experiments are tested on ascend 910* with mindspore 2.5.0 graph mode:
| model | # card(s) | image size | attn. type | throughput (token/s) |
|---|---|---|---|---|
| Janus-Pro-1B | 1 | 384x384 | Eager | 16.6 |
| Janus-Pro-7B | 1 | 384x384 | Eager | 12.2 |
Experiments are tested on ascend 910* with mindspore 2.5.0 pynative mode:
| model | # card(s) | image size | attn. type | throughput (token/s) |
|---|---|---|---|---|
| Janus-Pro-1B | 1 | 384x384 | Eager | 5.88 |
| Janus-Pro-7B | 1 | 384x384 | Eager | 3.30 |
Experiments are tested on ascend 910* with mindspore 2.5.0 graph mode:
| model | # card(s) | batch Size | image size | attn. type | throughput (token/s) | s/img |
|---|---|---|---|---|---|---|
| Janus-Pro-1B | 1 | 1 | 384x384 | Eager | 16.2 | ~ 40 |
| Janus-Pro-7B | 1 | 1 | 384x384 | Eager | 11.9 | ~ 52 |
Experiments are tested on ascend 910* with mindspore 2.5.0 pynative mode:
| model | # card(s) | batch size | image size | attn. type | throughput (token/s) | s/img |
|---|---|---|---|---|---|---|
| Janus-Pro-1B | 1 | 1 | 384x384 | Eager | 4.52 | ~ 127 |
| Janus-Pro-7B | 1 | 1 | 384x384 | Eager | 3.56 | ~ 162 |
- All the performances are tested with KV-Cache enabled.
This code repository is licensed under the MIT License. The use of Janus models is subject to DeepSeek Model License.
@article{chen2025janus,
title={Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling},
author={Chen, Xiaokang and Wu, Zhiyu and Liu, Xingchao and Pan, Zizheng and Liu, Wen and Xie, Zhenda and Yu, Xingkai and Ruan, Chong},
journal={arXiv preprint arXiv:2501.17811},
year={2025}
}