AWS Widens Amazon Bedrock Platform With Custom Model Imports, Evaluation, Choices, and Content Filtering
Amazon Web Services (AWS) has upgraded and expanded its generative AI app-building platform Amazon Bedrock with new features and versions of its proprietary Titan large language model (LLM) portfolio as well as the new Meta Llama 3 model as it strives to compete with Microsoft, OpenAI, Google, and other rivals in the space. The enhanced Amazon Bedrock includes custom model importing, model evaluation to match them to tasks, and new content safeguards.
Building on Bedrock
The new Custom Model Import feature stands out as the biggest upgrade for Amazon Bedrock. Organizations can integrate their own custom models into the Bedrock platform, providing more flexibility and choice in building generative AI applications with abilities like Retrieval Augmented Generation (RAG). The custom models operate like any other API, streamlining development from their creation to deployment, starting with open models like Flan-T5, Llama 3, and Mistral, with plans for additional support in the future.
“Customers can take models that they customized on Amazon SageMaker, or other tools, and easily add them to Amazon Bedrock. After an automated validation, they can seamlessly access their custom model, as with any other model in Amazon Bedrock. They get all the same benefits, including seamless scalability and powerful capabilities to safeguard their applications, adherence to responsible AI principles – as well as the ability to expand a model’s knowledge base with RAG, easily create agents to complete multi-step tasks, and carry out fine tuning to keep teaching and refining models. All without needing to manage the underlying infrastructure,” AWS Vice President of Data and AI Dr. Swami Sivasubramanian explained in a blog post. “With this new capability, we’re making it easy for organizations to choose a combination of Amazon Bedrock models and their own custom models while maintaining the same streamlined development experience.”
Evaluate and Filter AI
Amazon Bedrock augmented the Custom Model Import feature with its new Model Evaluation tool. As the name implies, Model Evaluation is designed to assist customers in assessing and comparing different generative AI models available on the Bedrock platform. Amazon wants the tool to address the challenge of selecting the most suitable model for specific use cases. Customers can leverage predefined evaluation criteria and built-in datasets to evaluate model performance efficiently. The range of models available has also grown with not only Meta’s Llama 3 but also the new and improved Amazon Titan Image Generator and Amazon Titan Text Embeddings V2 models.
The text-to-image creator turns written prompts into new visuals or edits existing ones. Now, it automatically includes an invisible watermark that makes it easy to check if Titan generated any given image. Text Embeddings V2 will roll out soon and is optimized for RAG use cases like search and chatbots that answer questions and make recommendations. The RAG approach also means Text Embeddings V2 has a range of sizes to embed in applications that don’t necessarily need the maximum amount of computing power and storage available. Amazon suggested using the model would cut storage needs to a quarter of the usual while maintaining 97% accuracy.
With the renewed push in the generative AI platform space, AWS is also looking to avoid and limit any controversy around the content produced with generative AI models. That’s where its new Guardrails for Amazon Bedrock tool comes into play. The feature provides built-in safeguards to filter out content the developer may believe is too sensitive or harmful. That could include personal information, profanity, or even specific words. Amazon claims it will block up to 85% of what developers want to prevent, thanks to natural-language descriptions and configurable thresholds for denied topics and content types.
Follow @voicebotai Follow @erichschwartz
AWS Augments Amazon Bedrock With New Features, Titan Models, Agents
