Join leaders in Boston on March 27th for a special evening of networking, insight, and conversation.request an invitation here.
While OpenAI is in a relatively quiet period, rivals human stole the headlines with the release of new claude 3 family of Large-Scale Language Models (LLM). But there is another notable underlying model provider that announced important generative AI news this week. Desi.
venture beat The last time we covered Israeli startups was in the fall of 2023. When we released open source models for DeciDiffusion and DeciLM 6B, these were fine-tuned variants Stability's Stable Diffusion 1.5 and Meta's LLaMA 2 7B, both open source and faster than the original source models. is designed to require fewer computing resources.Desi has since been released Descicodercode completion LLM, and Decidiffusion 2.0However, the latter, like many of Desi's other models, is suspended in Hugging Face.
Now, that company is release Deci-Nano, a new, even smaller and less computationally intensive closed source LLM, and the complete Gen AI development platform for enterprises and programmers, another paid product. Deci-Nano is currently only available as part of the Deci Gen AI development platform.
Are you moving away from open source?
The company appears to be moving toward a more fully commercial model, or a mixed model of open source and closed source, similar to what we've seen in the past. Mistral continues controversial partnership with Microsoft.
Does Deci and Mistral’s move to closed-source AI models signal that enthusiasm for open-source AI is waning? After all, every private company needs to make money somehow…
Performance at a (low) price…
If Deci is indeed moving in a more commercial direction as it appears, the company looks set to ease users and customers into this phase of its existence.
Deci-Nano provides language understanding and inference at lightning-fast inference speeds, generating 256 tokens in just 4.56 seconds on NVIDIA A100 GPUs.
On its blog announcing the Deci-Nano, the company posted a graph showing it outperforming the Mistral 7B-Instruct and Google's Gemma 7B-it models.
![](https://venturebeat.com/wp-content/uploads/2024/03/Screen-Shot-2024-03-15-at-1.24.04-PM.png?resize=936%2C379&strip=all)
Furthermore, Deci-Nano is priced very bullishly at $0.1 per million (input) tokens. OpenAI's GPT-3.5 Turbo for $0.50 and $0.25 for new Claude 3 Haiku.
“Deci-Nano embodies our production-oriented approach that focuses not only on quality but also on efficiency and cost-effectiveness,” said Yonatan Geifman, Co-Founder and CEO of Deci. states. Posts on his LinkedIn page. “We are building architectures and software solutions that extract the maximum computational power from existing GPUs.”
But it remains closed source. And Deci does not share the number of parameters publicly. VentureBeat reached out to the company's advisor and was told: „This model is actually closed source, and Deci has chosen not to release any information about its size or architecture. This is intended to generate buzz about the new Gen AI development platform.“ is leaving. ”
From financial and legal analysis to copywriting and chatbots, Deci-Nano's affordability and superior features unlock new possibilities for businesses looking to innovate without incurring excessive costs.
Deci offers numerous options for customers to deploy, either on serverless instances for ease and scalability, or dedicated instances for fine-tuning and increased privacy. The company says this flexibility allows businesses to scale their AI solutions as their needs evolve and seamlessly move between deployment options without sacrificing performance or security.
A new platform is born
While most of Deci's announcements this week focused on Deci-Nano, the bigger news (I kid you not) may be the company's move to offer a complete generative AI platform. news release as a „comprehensive solution designed to meet enterprise efficiency and privacy needs.“
What exactly will users of the platform get? “A new series of unique fine-tunable large-scale language models (LLMs), an inference engine, and an AI inference cluster management solution,” Deci said.
The first unique model offered through this platform is, of course, the Deci-Nano. But clearly Deci plans to offer other products based on the wording of these marketing materials.
The inference engine allows users to specify Deci-Nano by connecting to Deci's APIs and servers, running Deci-Nano on a customer's virtual private cloud, or deploying it on-premises on a customer's servers. You can expand it according to your needs.
For customers looking to manage Deci-Nano themselves in a virtual private cloud (VPC), Deci simply offers its own containerization model. The company also performs customer managed interventions on behalf of its customers. Kubernetes cluster.
Finally, Deci's Genartive AI Platform provides a complete on-premises deployment solution for customers who require technology in the data center rather than in the cloud. Deci provides these customers with a virtual container that houses both her Deci-Nano model and Deci's Infery software development kit, so the customer can use that model to create applications and applications for customers, employees, and other end users. You can build on the experience.
Pricing for the Deci Generative AI platform and its various installed products has not been made public, but we will update the information as it becomes available.
VentureBeat's mission will be a digital town square for technical decision makers to gain knowledge and transact on transformative enterprise technologies. Please see the briefing.