☁️ Cloud Embeddings: evaluate, finetune, deploy and store state-of-the-art pretrained embeddings 🔢
A repository for tackling cloud text pre-trained embeddings, from evaluation to deployment, including fine-tuning and vector stores, with an AWS cloud lens, with pretrained HuggingFace 🤗 embeddings and AWS.
A series of blog posts is coming soon to give more contexts to this part 👷🏻
Why do embeddings matter?
They're at the backbone of multiple ML systems we encounter every day; plus, as LLM encounter increasing popularity, the use-case of retrieval augmented generation (RAG) is a professional use of GenAI that heavily relies on embeddings.
Objective of this repository.
- This collection of code showcases an end-to-end guide from selection, evaluation, finetuning, deployment to storage and retrieval embeddings with a cloud lens.
- This collection will hopefully allow you to get some perspective on every step, allowing your organization to embark in a seamless embedding journey, ensuring production readiness at every step.
Repository structure
Evaluate
Thanks to the evaluation part of this repository, you can evaluate SOTA embeddings with MTEB and SageMaker Processing according to your needs and delights.
Finetune
Finetune part is about modern, LoRA finetuning embeddings with 🤗 HuggingFace and SageMaker Training.
Deploy
Deployment. Automated pretrained embedding deployment with AWS CDK and SageMaker Model Hosting, in a Serverless way.
Store
Deployment contains a ready-made DB instance with RDS!
🚧🚧 In construction 🚧🚧 We'll add resources on how to store and retrieve data, but you can already find excellent resources in here for instance.