Silicon Valley-Based Startup ‘Cerebras’ Introduces Open Source ChatGPT-Like Models
An American artificial intelligence company called Cerebras Systems, used to build computer systems for complex AI-based deep learning models, recently announced to release an open-source ChatGPT-like model.
These models will be free of cost to the research and business communities to promote greater collaboration.
This Silicon Valley-based startup released seven 7 AI models trained on its supercomputer Andromeda, including larger language models with 13 billion parameters and smaller models with 111 million parameters. Models with more parameters can perform more complex AI generative functions.
Last year the OpenAI ChatGPT was released with around 175 billion parameters having the ability to generate poetry and research content which influenced interest and investment in the AI field.
The startup claims that small language models can be used on smartphones and other smart devices while larger models run on PCs and servers. However, complex activities like summarizing lengthy passages call for larger models.
A chip consultant at Cambrian AI added that a smaller model could be more accurate than a larger one if you train it more.
Further, he adds that most AI models are being trained on Nvidia company processors, but new startups like Cerebras are trying to make space in that market.
The founder and CEO of Cerebras, Andrew Feldman, states that their startup training can also be utilized on Nvidia systems for further training or customization.