The tech community, dubbed ModelScope, allows developers and researchers to download and customize pre-trained AI models, making it easier for universities and smaller companies to use AI for their research and businesses.
The platform includes Alibaba’s proprietary large pre-trained models, which power tasks such as image-captioning and visual question-answering. Alibaba unveiled ModelScope at its annual Apsara Conference in Hangzhou.
Alizila caught up with Zhou Jingren, Vice President of Alibaba DAMO Academy, to learn how these models change AI.
Q. What types of AI models can you find in the community?
A. At the beginning, there will be over 300 models in fields including computer vision, natural language processing and audio. These models include Alibaba’s proprietary models trained on relatively large datasets.
For example, we have made our proprietary AI model OFA (One-For-All) accessible, a 6 billion-parameter pre-trained model that excels at cross-modal tasks such as image captioning and visual question answering. Another proprietary large pre-trained model from Alibaba, the Tongyi text-to-image pre-trained model with 5 billion parameters, is also now available to test online.
Q. Why is DAMO Academy building the ModelScope Community?
A. We intend to build a technology community and make AI more accessible. We launched ModelScope to make hundreds of our proprietary models developed by DAMO over the past five years freely available and lowered the threshold of building and running AI models.
We hope to enable universities and businesses to use AI for research and business applications easily.
Q. Why is it important to lower the threshold for AI research?
A. Any AI-related research typically requires three prerequisites: large computing power, a sizable amount of data, and a quality algorithm. But many small companies or research institutes don’t have these resources.
With ModelScope, developers from smaller companies and research institutes can easily develop their own AI models by fine-tuning existing models on the platform and running them online with support from Alibaba Cloud’s robust computing power.
Q. Who can use the models in the community and what can they do?
A. All developers can access the models in the community. (The platform interface is currently in Chinese, but we hope to include other languages in the future). They can try the models online and get the results within minutes. They can also develop their own AI by downloading and fine-tuning existing models and running them online backed by Alibaba Cloud. They can also deploy them on other cloud platforms or in a local setting.
We hope more developers will join us to contribute to the community. Today, there are already a few dozen models at ModelScope contributed by individual developers.
Q. How will the launch of the community change the future of AI?
A. By getting more developers to join us and making more quality models available in the community, we believe ModelScope can positively impact businesses, academia and society at large.
For example, a small firm can develop and operate its own AI models for digital operation in a very cost-effective and efficient way by tapping into the community’s abundant resources.
Universities and research institutes can also leverage the wide range of AI models and Alibaba Cloud’s proven computing power to easily accelerate their research in their respective fields at a low cost.
The community launch testifies to our ongoing commitment to making cutting-edge technologies inclusive and accessible to global communities.