Cohere, in collaboration with Amazon, has made available its foundational models of Command R and R+ on Amazon SageMaker JumpStart, a system allowing users to deploy and run inferences. These models are the latest in retrieval augmented generation (RAG), delivering enterprise-grade workloads through efficient, accurate, and scalable language models. The tool supports multiple languages, including but not limited to English, French, German, Spanish, and Japanese.
The Command R model excels in RAG applications and tool use tasks, with high accuracy and a token context length of 128,000. The more recent Command R+ model is designed for extended conversational interaction and complex RAG functionality tasks, ideal for multi-step workflows. It is also noteworthy that users seeking single-step tasks or focused on price considerations may find the original Command R more suitable.
SageMaker JumpStart offers ML practitioners model deployment and customization features. Foundational models can be implemented on dedicated SageMaker instances through a network-isolated environment. Additionally, the Python SDK allows you to experience these models programmatically. The Cohere Command R/R+ models can now be deployed with ease using Amazon SageMaker Studio, while ensuring data security within your AWS secure environment.
Users can explore available models via the SageMaker JumpStart through the SageMaker Studio UI and the SageMaker Python SDK. The Cohere Command R and R+ models can be found in the Cohere hub, and are deployed on Amazon Elastic Compute Cloud instances powered by NVIDIA Tensor Core GPUs.
For deploying a model, users are required to subscribe to the model on AWS Marketplace by following the on-screen instructions. Once subscribed, the deployment initiates, guided by model performance, MLOps controls, and SageMaker features such as Pipelines and Debugger.
The functionality of Command R/R+ specialty in performing well in 10 key languages and optimized to perform well in 13 others. It is trained to respond in the language of the user and perform cross-lingual tasks, such as translation or answering questions about content in other languages.
Command R/R+ can use document references and conduct advanced tasks through APIs. Its multi-step mode can call a sequence of tasks and quickly adapt based on information coming from external sources. Cleaning up resources post-exploration is key to avoid unwanted charges.
In conclusion, the aim of this post is to walk users through using SageMaker JumpStart to discover and deploy Cohere’s next-gen language models. Command R and R+ are built for enterprise use and will allow you to unlock new levels of productivity and innovation in your natural language processing tasks.
