Stability AI, a company renowned for commercializing Stable Diffusion, is taking strategic action to prioritize language models in an effort to improve its financial situation. The company has recently introduced its first model in the ‘Stable LM 2’ series, called ‘Stable LM 2 1.6B,’ which is a powerful language model despite its compact size of 1.6 billion parameters.
Developing Stable LM 2 1.6B involved training it on a vast amount of data, consisting of two trillion tokens, over two training cycles. This multilingual model was trained on data from seven languages, including English, Spanish, and French.
One of the main goals in creating Stable LM 2 1.6B is to lower the barriers for developers. Stability AI aims to enable more developers to participate in the generative AI ecosystem by providing them with access to this powerful language model. In fact, this model has outperformed other models with fewer than 2 billion parameters, surpassing well-known systems like Microsoft’s Phi-1.5, TinyLlama 1.1B, and Falcon 1B in various tasks.
With the release of Stable LM 2 1.6B, Stability AI is striving to empower developers and model creators by offering complete transparency on the training details. The company has not only introduced the base model but also an instruction-tuned version. It has made the data details and the final pre-training checkpoint available to facilitate developers in seamlessly continuing pre-training and fine-tuning processes.
In recent months, Stability AI has been gradually shifting its focus towards language models. This change in direction may be attributed to the company facing increased pressure from investors due to financial difficulties. Reports have even suggested that Stability AI is considering a sale, with potential buyers like Cohere and Jasper expressing interest. While the company continues to heavily invest in computing costs and salaries, its revenue remains relatively low. However, Stability AI is determined to prioritize research and development.
Emphasizing language-based systems aligns well with Stability AI’s existing capabilities in image and video generation. This strategic shift also positions the company as a competitor to providers of text-based models.
Access to Stable LM 2 1.6B is available for both commercial and non-commercial purposes, albeit with certain restrictions. Non-commercial usage requires a free Stability AI Membership, while a professional tier is available for creators and developers who meet specific criteria. Negotiations with Stability AI are necessary for enterprise-level access.
While Stable LM 2 1.6B offers significant capabilities, it is important to note that it may still have certain limitations, such as the potential for outputting toxic language. Stability AI urges the community to use the model responsibly and take appropriate measures to ensure ethical development practices.
These recent developments from Stability AI demonstrate the company’s determination to leverage language models for its future growth and success in the AI industry.
The source of the article is from the blog reporterosdelsur.com.mx