Building Generative AI applications has never been this easy.
From comparing model performance to building RAG based application, Amazon SageMaker Canvas provides a No Code capability democratizing Generative AI across the enterprise. You can now develop Gen AI application in a span of minutes
In the past couple of posts, we discussed how Amazon Bedrock as a Generative AI platform enables enterprises to deliver Gen AI solutions at scale in a secure and cost optimized manner. Amazon SageMaker Canvas build on top of the Bedrock Gen AI platform for No Code Gen AI application development
Let us dive a bit deeper in to SageMaker Canvas Gen AI capabilities
Access to Amazon Bedrock and SageMaker Models
Using SageMaker Canvas, you can now access 3 categories of models.
- First party Models — Models built by Amazon / AWS hosted as serverless endpoints on Amazon Bedrock
- Proprietary Models — Models build by partners of AWS hosted securely in an AWS environment available as Bedrock serverless endpoint
- Open-Source Models — Models available from Model Hubs like Hugging face hosted as a SageMaker end point
Chat using the foundation Models: You can now select any of the foundation models and start prompting these models.
Compare Model Performance
You can now compare performance of these models in the chat interface. Select the models that you intend to compare, send a prompt to these models and compare the results of the completions between the models. At any point of time, you will be able to compare across 3 different models. In the below example we are comparing against Claude Instant and Claude V2 for their completions for a prompt.
Developing RAG applications:
Building a RAG application requires orchestration across your retrieval system and generation system. This includes aggregating the retrieved context with the prompt and creating a completion. Currently SageMaker Canvas supports Amazon Kendra as a retrieval system. To use this feature
1. Index your data in Amazon Kendra using 40+ built in connector provided by Amazon Kendra
2. Enable the IAM role to connect to Amazon Kendra from SageMaker Canvas. You can do this when you create a SageMaker Domain
Within SageMaker Canvas, just select / enable Query Documents option and select the Kendra index to query from. In the below example, AWS Well architected framework documentation has been indexed in Amazon Kendra under the index name, aws-index
Compare Generation Models for RAG applications
You can now combine the RAG capability and Model Comparison capability of SageMaker Canvas to select the best fit generation model for your use case.
Cannot wait to see what new features are going to be release in 4 weeks at this year’s ReInvent. if you have not registered for the reinvent, you can get registered here — https://reinvent.awsevents.com/