Cohere has teamed up with Amazon Web Services (AWS) to offer its Command R and R+ models on SageMaker JumpStart, a suite of capabilities of the SageMaker end-to-end machine learning (ML) platform. Both models are designed to excel at real-world enterprise-level applications and are optimized for retrieval-augmented generation (RAG) workflows – tasks that involve conversational interaction and lengthy contexts.
The Command R series is a powerful family of language models, known for their high precision on RAG tasks and tool-use tasks, their low latency and high throughput, their ability to comprehend 128,000-token contexts, and their language capabilities across 10 major languages.
Command R+ is the latest addition to the family, optimized for extremely efficient conversational interaction and long-context tasks. It is best-suited for workflows necessitating complex RAG functionality and multi-step tool-use, while Command R is designed for simpler RAG tasks, single-step tool-use tasks, and occasions where price is a paramount consideration.
Using SageMaker JumpStart, ML practitioners can easily deploy the Command R/R+ models to dedicated SageMaker instances. The models operate in a secure AWS environment, under the user’s virtual private cloud (VPC) controls, ensuring data security.
Users can deploy models by subscribing to the model on the AWS marketplace and following the on-screen instructions to complete the subscription process. The notebook provided in the Cohere SageMaker GitHub repository offers end-to-end guidance on how to deploy the model for inference and clean up resources.
The Command R/R+ models are optimized for performance in 10 primary languages, with pre-training data included for an additional 13 languages. They can also handle cross-lingual tasks such as translation. Moreover, the models can answer questions based on a list of supplied document snippets and cite the information source in their responses – a feature that enables highly sophisticated automation tasks.
Once users have finished exploring models, they can delete the resources they’ve created to avoid incurring unnecessary charges. The Cohere Command R and R+ models on SageMaker JumpStart offer unparalleled performance and scalability in real-world use cases, unlocking higher productivity levels and innovation in natural language processing projects.
