Amid a comparatively quiet period for OpenAI, Rival Anthropocene has made headlines with the discharge of its latest Claude 3 family of huge language models (LLMs). But there's one other base model vendor to control that released some big news about generative AI this week: Decision.
VentureBeat last reported on the Israeli startup in fall 2023, when it released its open source models DeciDiffusion and DeciLM 6B, that are fine-tuned variants of Stability's Stable Diffusion 1.5 and Meta's LLaMA 2 7B – each also open source , that are faster and require less computing resources than their original source models. Since then, Deci has been released DeciCodera code completion LLM and Decidiffusion 2.0.
Now the corporate is release a brand new, even smaller and fewer computationally intensive LLM, Deci-Nano, which is closed source, in addition to a whole Gen-AI development platform for corporations and programmers, one other paid product. Deci-Nano is initially available exclusively as a part of the Deci Gen AI Development Platform.
Away from open source?
The company appears to be moving towards a more business or mixed open source/closed source model mix, much like what we saw with Mistral with its controversial partnership with Microsoft.
Do Decis and Mistral's moves toward closed-source AI models indicate a waning enthusiasm for open-source AI? After all, every private company has to generate profits by some means…
Rachel Salkin, vice chairman of promoting at Deci, told VentureBeat via email that:
“
Salkin also noted that:
..”, regardless that their demo areas have been paused.
Performance at a (low cost) price…
If Deci is indeed moving in a more business direction, because it seems, then the corporate appears to be making it easier for users and customers to enter this phase of its existence.
Deci-Nano provides language understanding and reasoning with ultra-fast inference speed, generating 256 tokens in only 4.56 seconds on NVIDIA A100 GPUs.
The company posted charts on its blog announcing Deci-Nano, showing it outperforming the Mistral 7B-Instruct and Google's Gemma 7B-it models.
Additionally, Deci-Nano is priced very aggressively in comparison with $0.1 per 1 million (input) tokens $0.50 for OpenAI's GPT-3.5 Turbo And $0.25 for the brand new Claude 3 Haiku.
“Deci-Nano embodies our production-focused approach, which incorporates a commitment not only to quality, but additionally to efficiency and cost-effectiveness,” said Yonatan Geifman, co-founder and CEO of Deci a post on his LinkedIn page. “We develop architectures and software solutions that get the utmost computing power out of existing GPUs.”
But it stays closed source. And Deci hasn't publicly shared what number of parameters it has. Salkin told VentureBeat:
AutoNAC is a technology developed by Deci that goals to scale back model size by analyzing an existing AI model and making a series of small models “whose overall functionality could be very near the unique model,” in accordance with a Deci white paper on technology.
From financial and legal evaluation to copywriting and chatbots, Deci-Nano's affordability and superior capabilities open up latest possibilities for corporations seeking to innovate without the burden of excessive costs.
Deci offers customers a variety of deployment options, either on serverless instances for simplicity and scalability or on dedicated instances for fine-tuning and enhanced data protection. The company says this flexibility ensures corporations can scale their AI solutions based on their needs and seamlessly switch between deployment options without compromising performance or security.
A brand new platform is born
Although the majority of Deci's announcement this week focused on Deci-Nano, the larger news (no pun intended) stands out as the company's move to supply what it describes as a whole generative AI platform in a single Press release as a “comprehensive solution that meets the efficiency and data protection requirements of corporations.”
What exactly do users of the platform get? “A brand new set of proprietary, fine-tunable Large Language Models (LLMs), an inference engine and an AI inference cluster management solution,” said Deci.
The first proprietary model to be offered through the platform is in fact Deci-Nano. But apparently Deci plans to supply more deals based on the wording of those marketing materials, a fact confirmed by Salkin, who wrote to us:
“”
The inference engine allows users to deploy Deci-Nano in accordance with their specifications by either connecting to Deci's API and servers, running Deci-Nano in the shopper's virtual private cloud, or deploying it on-premises to the shopper's server .
For customers who wish to manage Deci-Nano themselves in a virtual private cloud (VPC), Deci simply provides their very own container model. The Company also carries out managed interventions in its customer transactions on behalf of the Customer Kubernetes cluster.
Finally, Deci's Genartive AI Platform provides a whole on-premise deployment solution for patrons who want the technology of their data center somewhat than within the cloud. Deci provides these customers with a virtual container that incorporates each the Deci Nano model and Deci's Infery software development kit, allowing the shopper to construct the model into apps and experiences for patrons, employees, or other end users.
Pricing for the Deci Generative AI Platform and its various installation offerings haven’t been publicly listed, but we’ll update as soon as we receive this information.
