DeciCoder

Deci, an AI company, has introduced DeciCoder, an open-source Large Language Model (LLM) for code generation. DeciCoder aims to provide efficient and accurate code generation, addressing the increasing demand for such models. It leverages AutoNAC™, a proprietary Neural Architecture Search technology, to automate the process of generating optimal architectures. 

DeciCoder's architecture incorporates Grouped Query Attention and achieves a balance between precision and efficiency. In comparison to existing models like SantaCoder, DeciCoder offers higher throughput, fewer layers, more heads, and a parallel embedding size. DeciCoder not only focuses on efficiency gains but also on sustainability, reducing carbon emissions by 324 kg CO2 per model instance. This model is part of Deci's holistic approach to AI efficiency, and they are also planning to release a generative AI SDK. DeciCoder's permissive licensing allows developers to integrate it into projects with minimal constraints, promoting innovation and growth across industries.

Overall, DeciCoder represents AI efficiency's potential and contributes to a more sustainable and efficient AI future.