Deci, the cutting-edge deep learning company, has launched DeciCoder, a groundbreaking generative AI model that is set to revolutionize code generation for developers. Powered by a 1 billion-parameter Large Language Model (LLM) and equipped with a wide 2048-context window, DeciCoder surpasses previous models in code generation and establishes new standards of efficiency.
Unlike its counterparts, DeciCoder boasts unparalleled throughput and a low memory footprint. This allows development teams to optimize code generation with minimal latency and migrate workloads to more affordable GPUs, resulting in substantial cost savings. In fact, when compared to established code LLMs like SantaCoder, DeciCoder demonstrated an impressive 22% increase in throughput, a significant reduction in memory usage, and a 1.5-2.4 percentage point improvement in accuracy based on the HumanEval benchmark. Furthermore, when DeciCoder is combined with Deci’s LLM inference acceleration library, Infery, its throughput outperforms that of SantaCoder by a remarkable 350%.
DeciCoder was constructed using Deci’s proprietary Automated Neural Architecture Construction (AutoNAC) engine. This powerful technology, based on Neural Architecture Search (NAS), enables DeciCoder to strike an optimal balance between accuracy and processing speed. It tailors the architecture of the model to suit specific data features, tasks, performance objectives, and inference environments. Deci’s AutoNAC has previously been instrumental in generating highly efficient computer vision and Natural Language Processing (NLP) models, such as YOLO-NAS, DeciBERT, and DeciSeg.
This release of DeciCoder is just the beginning of a series of highly anticipated offerings from Deci in the field of Generative AI. Developers can access DeciCoder and its pre-trained weights under the permissive Apache 2.0 License, granting them broad usage rights and facilitating its integration into real-world, commercial applications.
Q: What is DeciCoder?
A: DeciCoder is a powerful generative AI model developed by Deci, designed to revolutionize code generation for developers.
Q: How is DeciCoder different from other models?
A: DeciCoder boasts impressive throughput and a low memory footprint, allowing for extensive code generation with minimal latency. It outperforms established models in terms of throughput, memory usage, and accuracy.
Q: What technology is used to create DeciCoder?
A: DeciCoder is constructed using Deci’s Automated Neural Architecture Construction (AutoNAC) engine, which leverages Neural Architecture Search (NAS) to optimize the model’s architecture for specific tasks and environments.
Q: Can developers use DeciCoder for commercial applications?
A: Yes, developers have broad usage rights under the permissive Apache 2.0 License, allowing them to integrate DeciCoder into real-world commercial applications.