Pinned Loading

1. Lora-Without-RegretLora-Without-RegretPublic

   Reproducing Lora Without Regret blog from scratch + Researching the ratio between LoRA and FullFT learning rates

   Python 13 2

2. Adversarial-Training-Reduces-Superposition-Through-Non-Robust-Feature-PruningAdversarial-Training-Reduces-Superposition-Through-Non-Robust-Feature-PruningPublic

   code for my most recent research studying why adversarial training reduces superposition as first shown by Gorton & Lewis (2025).

   Python 1

3. Transformers-on-MajorityTransformers-on-MajorityPublic

   An empirical study that builds on Merrill et al. theoretical construction to solve majority. A closer look into transformers reaction to training dynamics and data corruption.

   Python 6

4. Robustness-Superposition-causalityRobustness-Superposition-causalityPublic

   Replication of "Adversarial Examples Are Not Bugs, They Are Superposition" paper.

   Python 6

5. Vit-on-small-dataVit-on-small-dataPublic

   The Lightest Vision Transformer (ViT) trained from scratch out there to achieve 93.37 ± 0.07%” top-1 accuracy on CIFAR-10 within just 50 epochs.

   Python 8

6. training_models_from_scratchtraining_models_from_scratchPublic

   Training tiny models from scratch using NumPy in code and linear algebra on a piece of paper.

   Python 20 1