Scalable Acquisition, Synthesis and Curation of Data for Indic LLMs
Date:
Delivered an invited research talk on scalable data pipelines for acquisition, curation, and synthetic data generation for Indic Large Language Models.
- Discussed imbalances between typical pretraining data scales and alignment data requirements.
- Detailed the imperative for generating high-quality, high-EQ (Emotional Quotient) synthetic datasets.
- Audience included professors and graduate researchers in mathematical sciences and machine learning.