[TRACKER] Billion-Scale Data Generation

[jinsolp]

Goal is to create synthetic datasets that mimic the distribution of real world datasets at scale.
Main Related Issue: #1699

Current Status

• We were able to approximate 960M x 1024 Falcon dataset

TODOs

Evaluating how to decide parameters for scale up

• Investigate good PCA heuristics
• Investigate how sample size and number of clusters need to scale for approximating large dataset