Building blocks for foundation models.

FlashFFTConv: Efficient Convolutions for Long Sequences with Tensor Cores

Repo for "Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture"

Code release for the paper Perfectly Balanced: Improving Transfer and Robustness of Supervised Contrastive Learning

Fast and memory-efficient exact attention

Interactive Model Iteration with Weak Supervision and Pre-Trained Embeddings

More interactive weak supervision with FlyingSquid

Compositional Video Event Specification