Pinned Loading
-
Rolling-Window-Normalization
Rolling-Window-Normalization PublicCompanion code for a paper showing a statistically significant GPT-2 perplexity-S&P 500 correlation is spurious -- and how rolling-window normalization eliminates this correlation with no retraining.
Python
-
Thorngrad
Thorngrad PublicA NumPy-backed autograd engine with broadcasting-aware gradients, built from scratch as an extension of Karpathy's micrograd. Employs tensors instead of scalars, the use of optimizers, activations,…
Python
-
Transformer-From-Scratch
Transformer-From-Scratch PublicA full encoder-decoder Transformer (Vaswani et al. 2017) built from PyTorch primitives. Includes Pre-LN vs. Post-LN experiment on a sequence-reversal toy task.
Python
Something went wrong, please refresh the page to try again.
If the problem persists, check the GitHub status page or contact support.
If the problem persists, check the GitHub status page or contact support.