Stability AI Launches Open-Source StableLM Language Models
Stability AI introduces StableLM, a new open-source language model suite available in 3 billion and 7 billion parameter versions to democratize AI access. The models show strong performance in conversational and coding tasks despite their smaller size.
Stability AI releases StableLM, a new suite of open-source language models designed to drive innovation in text and code generation. The Alpha version currently includes 3 billion and 7 billion parameter models, and the company plans to launch larger versions ranging from 15 billion to 65 billion parameters in the future. This release marks Stability AI's expansion from image generation into the competitive language model space.
The new models train on a massive dataset of 1.5 trillion tokens, which allows the relatively small models to achieve high performance in conversational and coding tasks. Stability AI makes the base models freely available under a CC BY-SA-4.0 license to democratize access to AI technology. Additionally, the company provides research models fine-tuned for instruction using a mix of popular open-source datasets, though these specific variations remain under a non-commercial license for research purposes.
StableLM enters a rapidly growing AI landscape alongside rivals like OpenAI's GPT-4, Meta's LLaMA, and Google's PaLM. By open-sourcing these models, Stability AI enables researchers to thoroughly examine model performance, develop interpretability techniques, and identify potential risks. The company states that language models will serve as the backbone of the digital economy, emphasizing its belief that everyone deserves a voice in their design.