Paper: https://arxiv.org/html/2404.13950v1
As an MLE, I want to use SPLATE to get the same quality of results with less latency.
This will involve several tasks:
- Create a new Index type for sparse indexing.
- Create a new candidate generator using SPLATE.
- Create a new retriever based on SPLATE.