Stability AI has released StableLM, its open source language model, which will allow developers and researchers to inspect, use, and adapt StableLM base models for commercial or research purposes. The model is available in a 3 billion and 7 billion parameter alpha version, with larger models from 15 billion to 65 billion parameters to come later.
StableLM is the next step in Stability AI’s mission to democratize AI technology and make it accessible to everyone. It seeks to offer a transparent, open, and scalable alternative to proprietary AI as a language model that can generate text and code, making it a powerful tool for many AI applications. It is designed to be small and efficient, and despite its small size, it offers high performance in coding and conversational tasks thanks to its experimental dataset of 1.5 trillion tokens.
StableLM Features
StableLM (already on Github) has been designed to be transparent, accessible and supportive for users. Being an open source model, developers can inspect and adapt the model to their needs without sharing their sensitive data or relinquishing control of its AI capabilities. Additional StableLM features include:
Experimental data set
StableLM is trained on an experimental dataset built on top of The Pile, but three times larger, with 1.5 trillion content tokens. The richness of this dataset provides high performance in conversational and coding tasks, demonstrating that small, efficient models can deliver high quality.
research models
Stability AI is also releasing a series of research models that have been fine-tuned via instructions. These models will use a combination of five open source AI datasets for conversational agents. These models are intended for research purposes only and are released under a CC BY-NC-SA 4.0 license which does not permit commercial use.
Accessibility
StableLM is designed to be accessible to users and built to run on local devices, which means developers can build standalone applications compatible with widely available hardware instead of relying on proprietary services from one or two companies. This helps democratize AI technology and share its economic benefits with a broad community of users and developers.
Medium
StableLM has been built to support users and not replace them. It focuses on AI efficiency, specialization, and practical performance rather than the pursuit of divine intelligence. Stability AI develops tools that help people and businesses use AI to unlock their creativity, increase their productivity, and open up new economic opportunities.
