Deci Open Source Base Model, Surpassing Meta's LLaMA and Stable Diffusion

2023-09-15

DeepMind recently launched its generative AI base model, which outperforms Meta's algorithm by an astonishing 15 times. DeciDiffusion, DeciLM 6B, and its new Inference Software Development Kit (SDK) are setting new standards for performance and cost-effectiveness in the field of generative AI. Unlike closed-source API models, Deci provides unrestricted access to its models, allowing for self-hosting anywhere.

The computational requirements for training and inference of generative AI models have hindered teams from launching and scaling generative AI applications in a cost-effective manner. This Israeli-based company directly addresses this challenge with its latest product release, making inference scaling efficient, cost-effective, and suitable for enterprise integration.

Better and More Accurate

By using Deci's open-source models and Infery LLM, AI researchers can reduce inference computation costs by up to 80%. They can utilize existing and widely available GPUs such as NVIDIA A10. Deci's models are applicable to various applications, including content and code generation, image creation, and chat applications, among others.

One of the models, "DeciDiffusion 1.0," is a high-speed text-to-image model. It can generate high-quality images in less than a second, outperforming competitors' Stable Diffusion 1.5 model by three times.

DeciLM 6B, with its 5.7 billion parameters, stands out with its high-speed inference, being 15 times faster than Meta LLaMA 2 7B. Additionally, there is "DeciCoder," a code generation LLM with 1 billion parameters, which not only has excellent inference speed but also maintains or exceeds accuracy standards.

Yonatan Geifman, CEO and co-founder of Deci, emphasizes the necessity of control over model quality, inference process, and costs in the field of generative AI.

These models are built using Deci's proprietary Neural Architecture Search (AutoNAC) technology. In addition to the base models, Deci also introduces Infery LLM - an inference SDK that enables developers to significantly improve the performance of existing LLMs while maintaining the desired accuracy.

Prof. Ran El Yaniv, Chief Scientist and co-founder of Deci, said, "Through Deci's solution, companies not only gain enterprise-level quality and control but also have the flexibility to customize models and inference processes according to their needs."

"DeciLM-6B has set a new gold standard, with throughput astonishingly 15 times higher than Llama 2 7B. This achievement is attributed to Deci's cutting-edge Neural Architecture Search engine, AutoNAC," tweeted Akshay Pacchar, Chief Data Scientist at TomTom.

RECENT AI NEWS

RECENT AI TOOLS