AI21 debuts Jamba 1.5, boosting hybrid SSM transformer model to enable agentic AI

Transformers are the cornerstone of the fashionable generative AI period, but it surely’s not the one technique to construct a mannequin.

AI21 is out at the moment with new variations of its Jamba mannequin, which mixes transformers with a Structured State Area (SSM) mannequin method. The brand new Jamba 1.5 mini and Jamba 1.5 massive construct on the preliminary improvements the corporate debuted with the discharge of Jamba 1.0 in March. Jamba makes use of an SSM method often known as Mamba. Jamba’s purpose is to deliver the perfect attributes of transformers and SSM collectively. The title Jamba is definitely an acronym that stands for Joint Consideration and Mamba (Jamba) structure. The promise of the mixed SSM Transformer structure is healthier efficiency and accuracy than both method can present by itself.

“We received superb suggestions from the neighborhood, as a result of this was mainly the primary and nonetheless is likely one of the solely Mamba primarily based manufacturing scale fashions that we received,” Or Dagan, VP of product at AI21 informed VentureBeat. “It’s a novel structure that I believe began some debates about the way forward for structure in LLMs and whether or not transformers are right here to remain or do we’d like one thing else.”

With the Jamba 1.5 collection AI21 is including extra capabilities to the mannequin together with operate calling, JSON mode, structured doc objects and quotation mode. The corporate hopes that the brand new additions make the 2 fashions very best for crafting agentic AI programs. Each fashions even have a big context window of 256K and are Combination-of-Specialists (MoE) fashions. Jamba 1.5 mini gives 52 billion whole and 12 billion lively parameters. Jamba 1.5 massive has 398 billion whole parameters and 94 billion lively parameters. 

Each Jamba 1.5 fashions can be found below an open license. AI21 additionally gives industrial help and companies for the fashions. The corporate additionally has partnerships with AWS, Google Cloud, Microsoft Azure, Snowflake, Databricks and Nvidia.

What’s new in Jamba 1.5 and the way it will speed up agentic AI

Jamba 1.5 Mini and Massive introduce a lot of new options designed to fulfill the evolving wants of AI builders:

• JSON mode for structured information dealing with
• Citations for enhanced accountability
• Doc API for improved context administration
• Perform calling capabilities

Based on Dagan, these additions are significantly essential for builders engaged on agentic AI programs. Builders extensively use JSON (JavaScript Object Notation) to entry and construct utility workflows.

Dagan defined that including JSON help allows builders to extra simply construct structured enter/output relationships between completely different components of a workflow. He famous that JSON help is essential for extra complicated AI programs that transcend simply utilizing the language mannequin by itself. The quotation characteristic however, works at the side of the brand new doc API. 

“We will educate the mannequin that while you generate one thing and you’ve got paperwork in your enter, please attribute the related components to the paperwork,” Dagan mentioned. 

How quotation mode is completely different than RAG, offering an built-in method for agentic AI

Customers shouldn’t confuse quotation mode with Retrieval Augmented Technology (RAG), although each approaches floor responses in information to enhance accuracy.

Dagan defined that the quotation mode in Jamba 1.5 is designed to work at the side of the mannequin’s doc API, offering a extra built-in method in comparison with conventional RAG workflows. In a typical RAG setup, builders join the language mannequin to a vector database to entry related paperwork for a given question or activity.The mannequin would then have to study to successfully incorporate that retrieved data into its technology.

In distinction, the quotation mode in Jamba 1.5 is extra tightly built-in with the mannequin itself. This implies the mannequin is skilled to not solely retrieve and incorporate related paperwork, but in addition to explicitly cite the sources of the knowledge it makes use of in its output. This gives extra transparency and traceability in comparison with a conventional LLM workflow, the place the mannequin’s reasoning could also be extra opaque.

AI21 does help RAG as properly. Dagan famous that his firm presents its personal end-to-end RAG answer as a managed service that features the doc retrieval, indexing, and different required elements.

Wanting ahead, Dagan mentioned that AI21 will proceed to work on advancing its fashions to serve buyer wants. There may even be a continued give attention to enabling agentic AI.

“We additionally perceive that we have to function and push the envelope with agentic AI programs and the way planning and execution is dealt with in that area,” Dagan mentioned.