An instruction-tuned version of our hybrid SSM-Transformer Jamba model, Jamba-Instruct is built for reliable commercial use, with best-in-class quality and performance.
May 2nd, 2024
Following our recent release of our groundbreaking Jamba open model—built from a novel hybrid SSM-Transformer architecture—Jamba’s outstanding quality, performance, and unprecedented context window length are now available for commercial use.
Jamba-Instruct, the aligned version of Jamba, is in public preview on the AI21 Platform. Packaged with additional training, chat capabilities, and safety guardrails to make it suitable for immediate and secure enterprise use, Jamba-Instruct is the model of choice for companies looking to use GenAI to solve and run critical enterprise workflows.
Jamba-Instruct outperforms all other instruction-tuned competitors on quality and cost, making it ideal for enterprises looking to get the most value for their money.
Not only does Jamba-Instruct boast the largest context window in its size class, its cost optimizations means that this 256K context window is more economically accessible than ever before, placing a wide range of long context enterprise use cases within reach.
Additionally, with a context window that large, Jamba-Instruct is able to intake the equivalent of a 400 page novel (or all of a company’s financial filings for a given fiscal year, for example) and answer questions and produce summaries that are grounded in this content—instead of manually segmenting document sections to fit the much smaller context windows of our competitors.
As a result, Jamba-Instruct is ideal for some of the most common use cases enterprises can encounter, such as enabling accurate QA (question answering) on earnings call transcripts, thoroughly summarizing key points from legal documents, or building chatbots that can coherently sustain long and reference-based conversations with customers.
Jamba-Instruct maintains the impressive milestones achieved by the powerful Jamba base model, while including new features that make it suitable for reliable commercial use.
In addition to upholding Jamba’s impressive 256K context window and leading performance across common quality benchmarks, Jamba-Instruct comes ready with safety guardrails, chat capabilities, and enhanced command comprehension—all essential features for reducing the total cost of model ownership, mitigating risk, and improving overall performance.
Whereas Jamba, as an open model, requires extensive independent fine-tuning to ensure output that’s safe, accurate, and relevant, we’ve already done the heavy lifting for Jamba-Instruct to make it ready for immediate implementation.
The result? An accelerated time to production for your next GenAI enterprise application.
Jamba-Instruct is available in public preview from the AI21 Platform. Sign into your account (or create a new one for free) to get started. We’re looking forward to hearing what you build with this groundbreaking technology.