A comprehensive overview of generative AI models on Amazon Bedrock
Amazon Bedrock is a fully managed service on AWS that offers a choice of high-performing foundation models (FMs) from AI21 Labs, Anthropic, Cohere, Stability AI, and Amazon. It provides a broad set of capabilities to build generative AI applications, simplifying the development while maintaining privacy and security.
AI21 Labs offers the Jurassic family of large language models (LLMs) on Amazon Bedrock. Here are the two versions of Jurassic models available:
Jurassic-2 Ultra: This is AI21’s most powerful model, offering exceptional quality. It supports a maximum of 8,192 tokens and languages including English, Spanish, French, German, Portuguese, Italian, and Dutch. The supported use cases include question answering, summarization, draft generation, advanced information extraction, and ideation for tasks requiring intricate reasoning and logic.
Jurassic-2 Mid: This is AI21’s mid-sized model, designed to strike the right balance between exceptional quality and affordability. It also supports a maximum of 8,192 tokens and the same languages as the Ultra model. The supported use cases include question answering, summarization, draft generation, and advanced information extraction.
These models are built to follow natural language instructions and are used in thousands of live applications. They are widely adopted across verticals like finance, retail, customer support, and knowledge management.
Anthropic offers the Claude family of large language models (LLMs) on Amazon Bedrock. Here are the versions of Claude models available:
Claude 2: This is Anthropic’s most capable system to date. It’s a general-purpose large language model (LLM) that supports a maximum of 100K tokens and multiple languages including English. The supported use cases include thoughtful dialogue, content creation, complex reasoning, creativity, and coding.
Claude 1.3: This is an earlier version of Anthropic’s general-purpose LLM. It also supports a maximum of 100K tokens and multiple languages including English. The supported use cases include search, writing, editing, outlining, summarizing text, coding, and providing helpful advice about a broad range of subjects.
Claude Instant: This is Anthropic’s faster, lower-priced yet very capable LLM. It supports a maximum of 100K tokens and multiple languages including English. The supported use cases include casual dialogue, text analysis, summarization, and document comprehension.
These models are based on Anthropic’s research into creating reliable, interpretable, and steerable AI systems. They are created using techniques like Constitutional AI and harmlessness training.
Stability AI offers the Stable Diffusion family of models on Amazon Bedrock. Here is the version of Stable Diffusion model available:
Stable Diffusion XL 1.0: This is Stability AI’s text-to-image tool for generating unique, realistic, high-quality images, art, logos, and designs. It supports a maximum of 100K tokens. The supported use cases include image generation from text prompts.
The Stable Diffusion models support the following controls:
Prompt strength (cfg_scale): Determines how much the final image portrays the prompt. Use a lower number to increase randomness in the generation.
Generation step (steps): Generation step determines how many times the image is sampled. More steps can result in a more accurate result.
Seed: The seed determines the initial noise setting. Use the same seed and the same settings as a previous run to allow inference to create a similar image. If you don’t set this value, it is set as a random number.
Using Stable Diffusion on Amazon Bedrock allows developers to easily access these models without leaving their AWS environment and leverage the powerful generative image capabilities without worrying about managing the infrastructure.
Amazon offers the Titan family of foundation models (FMs) on Amazon Bedrock. Here are the versions of Titan models available:
Titan Text Express (preview): This is a large language model (LLM) offering a balance of price and performance. It supports a maximum of 8K tokens and the English language. The supported use cases include retrieval augmented generation, open-ended text generation, brainstorming, summarization, code generation, table creation, data formatting, paraphrasing, chain of thought, rewrite, extraction, Q&A, and chat.
Titan Text Lite (preview): This is an affordable and compact model, ideal for basic tasks and fine-tuning. It supports a maximum of 4K tokens and the English language. The supported use cases include open-ended text generation, brainstorming, summarization, code generation, table creation, data formatting, paraphrasing, chain of thought, rewrite, extraction, Q&A, and chat.
Titan Embeddings (generally available): This is an LLM that translates text into a numerical representation. It supports a maximum of 8K tokens, over 25 languages, and 1,536 embeddings. The supported use cases include text retrieval, semantic similarity, and clustering.
These models are pretrained by AWS on large datasets, making them powerful, general-purpose models built to support a variety of use cases. They can be used as is or privately customized with your own data. Using Titan on Amazon Bedrock allows developers to easily access these models without leaving their AWS environment.
Meta offers the Llama 2 family of large language models (LLMs) on Amazon Bedrock. Here are the versions of Llama 2 models available:
Llama-2–13b-chat: This is a fine-tuned model with a parameter size of 13B. It supports a maximum of 4K tokens and the English language. The supported use cases include assistant-like chat.
Llama-2–70b-chat: This is a fine-tuned model with a parameter size of 70B. It supports a maximum of 4K tokens and the English language. The supported use cases include assistant-like chat.
These models are pretrained and fine-tuned by Meta and are optimized for dialogue use cases. They come with significant improvements over the original Llama models, including being trained on 40% more data and having a longer context length of 4,000 tokens to work with larger documents.
