StableLM

Official Website of StableLM, a Large Language Model from Stability AI

What is StableLM?

Perhaps every rise in the historical tide is accompanied by a lively scene. At the beginning of this year, the AI large language model ChatGPT ignited the market, and now another heavyweight competitor has joined the race—the Stability AI. Stability AI is a company that launched the popular AI painting domain open-source model Stable Diffusion. On April 19th, they announced the release of the open-source large language model StableLM and challenged ChatGPT. When you enter the large language model introduction page on the Stability AI website, you will first see a green parrot, showcasing the painting features of Stable Diffusion XL, followed by the introduction of the large language model. Perhaps, this is the unexpected arrival of the large language model "Stable Diffusion."

Official Website of StableLM

Download on GitHub and other open-source platforms:https://github.com/stability-AI/stableLM/

Stability AI recently released the large language model StableLM, and this website introduces the relevant parameters and download methods of the model. Currently, the Alpha version of StableLM can be downloaded from GitHub and other open-source platforms, including 30 billion and 70 billion parameter versions, with higher parameter versions to be launched later. Similar to Stable Diffusion, developers can use this model for commercial or research activities as long as they comply with the agreement.

Founded in 2019, Stability AI is one of the most well-known companies in the generative AI field. In the large language model market where ChatGPT has occupied a position, many people are looking forward to the appearance of an open-source model similar to Stable Diffusion as soon as possible. Stability AI compares the release of StableLM to Stable Diffusion, indicating that the model will also be opened to the public like Stable Diffusion. This news has attracted a lot of attention, and the start of the AI arms race may be imminent.

Therefore, Stability AI co-funded the research center Eleuther AI with several startups at the beginning of this year, and StableLM is built on the open-source model of Eleuther AI.

These models include GPT-J, GPT-NeoX, and Pythia, etc., and the training of this large model is completed on the open-source dataset The Pile, which combines text samples from multiple websites such as PubMed, StackExchange, and Wikipedia, with a scale three times larger than The Pile, containing 1.5 trillion content tags.

Although the parameter scale cannot be directly compared with GPT-3, StableLM demonstrates astonishing high performance in communication and programming, which is the benefit of diverse text datasets.