Code and data of the EMNLP 2022 Main Conference paper "Reduce Catastrophic Forgetting of Dense Retrieval Training with Teleportation Negatives".
An unofficial extended version for ICLR 2021"Gradient Descent on Neural Networks Typically Occurs at the Edge of Stability"
PyTorch NaNs are silent killers. This hook catches them at the exact layer and batch — with ~3 ms overhead vs ~7 ms for set_detect_anomaly.
RVAV: a physics-informed PyTorch optimizer for energy-stable, high-LR training—with a simple closure API, tests, CI, and quickstart.
Benchmarking GAN optimizers (Adam, RMSprop, SGD, Lookahead) on CIFAR-10 using WGAN-GP and FID evaluation.
A parameterized, drop-in activation function for stabilizing deep Batch-Normalization-free networks in micro-batch and Federated Learning settings.
Drift-Aware Adaptive Aggregation (DAA) for federated learning on CIFAR-10 under heterogeneous client partitions.
Add a description, image, and links to the training-stability topic page so that developers can more easily learn about it.
To associate your repository with the training-stability topic, visit your repo's landing page and select "manage topics."