amazon-bedrock-with-builder-and-command-patterns VS llm-awq

The second service is what’s going to make our application come alive and give it the AI functionality we need and that service is AWS Bedrock which is their new generative AI service launched in 2023. AWS Bedrock offers multiple models that you can choose from depending on the task you’d like to carry out but for us, we’re going to be making use of Meta’s Llama V2 model, more specifically meta.llama2-70b-chat-v1.

Amazon Bedrock

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon. Each model is accessible through a common API which implements a broad set of features to help build generative AI applications with security, privacy, and responsible AI in mind.

For those keeping track, Amazon Bedrock became generally available in September of 2023. My team had access to a preview, so when the AWS Comprehend entity analysis did not lend itself well to my use case; and I didn't feel like training a model, I started to get familiar with Bedrock. The following post is a follow-on to the Community article above and fleshes out a few details that will help those newer to Amazon Bedrock navigate the product.

In this blog you will learn how to use Amazon Cognito credentials and IAM Roles to invoke Amazon Bedrock API in a react-based application with JavaScript and the CloudScape design system. You will deploy all the resources and host the app using AWS Amplify.

Working with the AWS Bedrock API is an exhilarating experience! I came across an interesting business case where I needed to develop an AI MVP. The MVP generates JSON data based on a prompt and utilizes the anthropic.claude-v2:1 model in AWS Bedrock.

TinyChat is an efficient, lightweight, Python-native serving framework for 4-bit LLMs by AWQ. It delivers 2.3x generation speed up on RTX4090.

Code: https://github.com/mit-han-lab/llm-awq/tree/main/tinychat

I am having trouble finding any 8bit GPTQ models at all, there don't seem to be any on HF it's almost all 4bit with the odd 3bit of the big ones. Suspect I will have to make my own for eval purposes but it's lower priority on my list then finding a 4bit that's GPU friendly but doesn't have such a performance penalty... Looking at AWQ they have 3 and 4bit versions.

Using the currently popular gptq the 3bit quantization hurts performance much more than 4bit, but there's also awq (https://github.com/mit-han-lab/llm-awq) and squishllm (https://github.com/SqueezeAILab/SqueezeLLM) which are able to manage 3bit without as much performance drop - I hope to see them used more commonly.

Better quantization would have a direct and meaningful impact for everyone running local LLMs. The technique has already been applied to both Vicuna and the multimodal LLaMA variant LLaVA.

https://github.com/mit-han-lab/llm-awq

GitHub: https://github.com/mit-han-lab/llm-awq