Tags → #python
-
TAVA - A Novel Method for Label-Free Embedding Compression
Ongoing research on compressing text embeddings without labeled data, using a novel two-stage distillation approach with generative adversarial adapters.
-
Loadax - High-Performance Data Loading for JAX
A JAX-native dataloading library with prefetching, multi-worker support, and distributed loading for ML training pipelines.
-
Scratch
Machine learning architectures and methods implemented from scratch.
-
Menagerie
A menagerie of deep learning models and techniques.
-
Ghostwriter
Learning to mimic famous musicians with generative language models.
-
Hooper
An AI chatbot with generative UI built to discuss the NBA with realtime news and stats.
-
Universal Pretraining - Exploring the Value of Modality
Research I worked on during my graduate studies at SMU investigating the influence of modality on transformer pretraining. This work was abandoned.
-
Algorithms
Common data structures and algorithms I often end up reimplementing in python.