NinjaLLM: Fast, Scalable and Cost-effective RAG using Amazon SageMaker and AWS Trainium and Inferentia2
This work addresses the problem of making RAG systems more efficient and accurate for users needing scalable AI deployments, though it appears incremental as it builds on existing techniques with specific hardware optimizations.
The paper tackled improving retrieval-augmented generation (RAG) techniques by enhancing traditional methods with large language models fine-tuned on AWS Trainium and Inferentia2 chips, achieving accuracies of 62% on Natural Questions and 59% on HotPotQA, which exceeded models like DBRX and Mixtral Instruct.
Retrieval-augmented generation (RAG) techniques are widely used today to retrieve and present information in a conversational format. This paper presents a set of enhancements to traditional RAG techniques, focusing on large language models (LLMs) fine-tuned and hosted on AWS Trainium and Inferentia2 AI chips via SageMaker. These chips are characterized by their elasticity, affordability, and efficient performance for AI compute tasks. Besides enabling deployment on these chips, this work aims to improve tool usage, add citation capabilities, and mitigate the risks of hallucinations and unsafe responses due to context bias. We benchmark our RAG system's performance on the Natural Questions and HotPotQA datasets, achieving an accuracy of 62% and 59% respectively, exceeding other models such as DBRX and Mixtral Instruct.