Amazon Bedrock: Cohere Command Light Features

Cohere Command Light

Business applications that create text, summarize, search, cluster, categorize, and make use of Retrieval Augmented Generation (RAG) are powered by Cohere’s text generation and representation models. Amazon are pleased to announce that Cohere Command Light and Cohere Embed English and international variants are now available on Amazon Bedrock. They are now a part of the Cohere Command model.

Amazon Bedrock Access

Amazon Bedrock is a fully managed service that provides a wide range of tools to create generative AI applications, simplifying development while upholding privacy and security. It offers a selection of high-performing foundation models (FMs) from top AI firms, such as AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon. With this release, Amazon Bedrock broadens the range of model options available to assist you in developing and growing generative AI that is suited for the workplace.

Cohere’s primary text generating paradigm is called Command. It has been trained to comply with user instructions and perform well in corporate applications. A group of models called Embed have been trained to generate superior embeddings from text documents.

One of the most intriguing ideas in machine learning (ML) is embeddings. They are essential to many applications that handle search algorithms, recommendations, and natural language processing. Any kind of text, image, video, music, or document can be converted into a vector, which is just a set of numbers. Specifically, the term “embodiments” refers to the method of capturing important information, semantic linkages, or contextual qualities in data by encoding it as a vector. The reason embeddings are helpful is that the vectors they represent are “close” to one another, to put it simply. Formally speaking, embeddings convert human-perceived semantic similarity to closeness in a vector space. Generally, models or algorithms are trained to create embeddings.

A collection of models called Cohere Embed has been trained to produce embeddings from text sources. There are two versions of Cohere Embed: a multilingual model and an English language model. Both versions are now accessible on Amazon Bedrock.

Text embeddings have three primary use cases:

Semantic searches: Embeddings make it possible to search collections of documents based on their meaning, which improves upon current keyword-matching methods by allowing search systems to better take context and user purpose into account.

Text Classification: Create automated systems that identify different types of text and take appropriate action. An email filtering system might choose, for instance, to forward certain messages to sales and escalate others to tier-two support.

Retrieval Augmented Generation (RAG): Enhance the quality of text generated by a large language model (LLM) by adding contextualized data to your prompts. One can leverage a variety of data sources, like document repositories, databases, and APIs, to supplement your prompts with external data.

Let’s say you have hundreds of documents outlining the policies of your business. Owing to the restricted length of prompts that LLMs accept, you must choose pertinent passages from these materials to incorporate into questions as background. Turning all of your documents into embeddings and storing them in a vector database, like OpenSearch, is the answer.

In order to locate the most pertinent documents for a user’s query inside this corpus of documents, you convert the user’s natural language query into a vector and do a similarity search on the vector database. Next, you combine pun intended the user’s initial inquiry and the pertinent documents that the vector database surfaced into an LLM prompt. Including pertinent papers within the prompt’s context aids in the LLM’s ability to produce responses that are more precise and pertinent.

Now you can use the AWS SDKs, the AWS Command Line Interface (AWS CLI), or the Bedrock API to integrate Cohere Command Light and Embed models into your applications developed in any programming language.

Coherent Embedded at Work

Today, Amazon are introducing Cohere Command Light, Cohere Embed English, and Cohere Embed multilingual three unique versions. The process of writing code to launch Cohere Command Light is identical to that of Cohere Command, an existing feature of Amazon Bedrock. For this example, Amazon will therefore walk you through the process of writing code to communicate with Cohere Embed and go over how to use the embedding that it creates.

Accessibility

In two AWS Regions US East (North Virginia) and US West (Oregon) where Amazon Bedrock is available, the Cohere Embed models are currently accessible to all AWS users.

AWS charges for inferring models. AWS charges for each input or output token that is handled for Command Light. AWS costs for each input token for embed models. You have the option to pay for services only when they are used, with no one-time or ongoing costs. It is also possible to provide enough throughput to fulfill the performance needs of your application in return for a time-based term commitment.

With this knowledge, you can use the Cohere Embed and Amazon Bedrock text embedding models in your apps.