Stable is available for developers for both commercial and research purposes on GitHub.
Alongside the new large language model, Stability AI has also released a set of research models with finely tuned instructions. (Image Source: Stability AI)
Stability AI, the company that brought us the popular text-to-image generator Stable Diffusion recently launched a new open-sourced large language model called StableLM, which is available on GitHub.
In a recent blog post, the company announced that the alpha version of StableLM is now available in 3 billion and 7 billion parameters, which will soon be followed by 15 billion and 65 billion. The new large language model will be available to developers for both commercial and research purposes.
Stability AI has trained StableLM on a new experimental dataset based on ‘The Pile’ but with three times more tokens of content. According to the company, StableLM, despite having fewer parameters (3-7 billion) compared to other large language modes like GPT-3 (175 billion), offers high performance when it comes to coding and conversations.
StableLM when asked for an alternate title for how to add contacts to an Android phone. (Express Photo)
Also Read |Startup behind Stable Diffusion releases AI system for generating videos from text
Interested users can check out the alpha version of the large language model by searching for StableLM on Hugging Face. When we tried StableLM, it was slow to respond and most of the time, came up with an answer that was completely unrelated to the query.
For example, when asked to suggest an alternate title for ‘How to add contacts on your Android device’, it said that users can make use of the contacts app on the phone to add new contacts. It looks like StableLM still has a long way to go before it can compete with the likes of ChatGPT.
Alongside the new large language model, Stability AI has also released a set of research models with finely tuned instruction which uses conversational agents like, GPT4All, Dollt, ShareGPT, Alpaca and HH. However, these models are only for research purposes and are unavailable for commercial use.
Source:indianexpress.com