projects

Research projects with code. Each card links to its repository.

research

Translated BabyLM

BabyLM-scale training corpora for Hindi and Telugu built by machine translation, with GPT-BERT and GPT-Wee models trained on them. With Prof. Libby Barak.

NLPlabMSU/babylm-translation-scripts

Dravidian LMs

Five GPT-2 models trained from scratch to compare dedicated monolingual and joint multilingual modeling for Tamil, Telugu, Kannada, and Malayalam.

vishnup22/dravidian-lm-research

Fairness Drift

Fairness regression testing across LLM releases: 4 providers, 7 bias benchmarks, and 8 release transitions under a fixed evaluation procedure.

pulipakav1/fairness-drift-llms

ongoing

Telugu Dialects

How regional and dialect variation in Telugu and Hindi affects tokenization, representations, and transfer in language models.

vishnup22/telugu-dialect-nlp