LaDyS
LaDyS is a PyTorch package for benchmarking new models against standard dynamical latent variable models like GPFA, LFADS, and CASSM. Unfortunately, the Neural Latents Benchmark was sunsetted for this task in January 2026, and benchmarked on just four small-scale datasets. Thus, the code was a black box, preventing researchers from testing their models on new, larger datasets or debugging on synthetic datasets. We are paying out code bounties. We welcome any new task suggestions and are willing to pay out bounties for tasks you think are useful. CURRENTLY RELEASED.