Stability AI Enters Language Model Race With Open-Source StableLM
Stability AI launches its first open-source language models, StableLM, marking its expansion beyond image generation into the competitive LLM market. The initial alpha releases feature 3 billion and 7 billion parameter models trained on a massive new experimental dataset.
Stability AI introduces the StableLM suite, marking its official entry into the competitive language model space currently dominated by major tech companies. Known for its open-source image generator Stable Diffusion, the company now expands its foundational AI technology to handle text and code generation tasks.
The first StableLM-Alpha models are available in 3 billion and 7 billion parameter sizes, both trained on an impressive 800 billion data tokens. Despite their relatively small size, these models perform surprisingly well in conversational and coding scenarios because they learn from a new experimental dataset that is three times larger than the popular open-source dataset called The Pile.
Stability AI plans to release larger models ranging from 15 billion to 65 billion parameters in the future. By open-sourcing StableLM, the company continues its mission to promote transparency and accessibility, allowing developers to freely inspect, use, and adapt the technology for a wide variety of downstream applications.