Stability AI, the firm behind the AI-powered Stable Diffusion image generator, has published StableLM, an open-source suite of large language models (LLMs). In addition, the company revealed in a blog post that its models are now available for developers to use and change on GitHub.

“Artificial intelligence and large-scale models should be open to the public, and only when the threshold is so low that everyone can use them conveniently, can there be a real large-scale outbreak of creativity.” – Wu Tian, Vice President of Baidu

Stability AI was the driving force behind the 2022 public release of Stable Diffusion, a ground-breaking picture model that is a transparent, adaptable, and open replacement for proprietary AI. With the introduction of the StableLM family of models, Stability AI is advancing the availability of fundamental AI technology to all users. Their models can produce text and code and will power various applications further down the line. In addition, they show how practical training can enable compact, efficient models to achieve outstanding performance. 

Open-source language model 

The release of StableLM extends the open-sourcing of previous language models by the non-profit research centre EleutherAI. These language models consist of the Pile open-source dataset-trained GPT-J, GPT-NeoX, and Pythia suite. Recent open-source language models, such as Cerebras-GPT and Dolly-2, continue to build on these initiatives.

The experimental dataset used to train StableLM is three times the size of the Pile and contains 1.5 trillion tokens of content. According to the source, the researchers will reveal the dataset's specifics in due time. The richness of this dataset enables StableLM to perform remarkably well in conversational and coding tasks.

Features

• Transparent - The researchers release their models as open source to promote transparency and build trust. Researchers can develop techniques for interpretability, identify potential hazards, and aid in developing safeguards. Organizations in the public and private sectors can modify (or "tune") these open-source models for their applications without disclosing sensitive information or relinquishing control over their AI capabilities.
• Accessible - Researchers design for the edge so familiar users can execute their models on local devices. Using these models, developers can create independent applications compatible with broadly available hardware instead of relying on a few companies' services. Thus, a large community of consumers and developers share the economic benefits of artificial intelligence. 
• Supportive - Researchers create models to assist their users, not to supplant them. The researchers are focused on developing efficient, specialized, and applicable AI instead of seeking godlike intelligence. They create tools that help common individuals and businesses use AI to unleash their creativity, increase their productivity, and create new economic opportunities.  

Conclusion

Like its competitor ChatGPT, StableLM is made to make text and code quickly. It was trained on a bigger version of the open-source dataset called the Pile, which includes information from many sites like Wikipedia, Stack Exchange, and PubMed. 

Furthermore, While StableLM relies on the open-source language models developed by Stability AI in conjunction with the organization EleutherAI, it also continues Stability AI's objective to make AI tools more accessible, as it did with Stable Diffusion. 

The models are now accessible in their repository on GitHub.

Image source: Unsplash