Implement Vision Transformer + MAE Pretraining in PyTorch
Hardmodern-architecturesauto-graded
Build ViT with masked autoencoder pretraining — randomly mask patches, encode visible ones, decode to reconstruct.
Solve it
Check your answer
The grader verifies properties of your implementation, so a correct solution written differently from ours still passes.
pip install torchleet
from torchleet import check
check("vit-mae", PatchEmbedding, ViT, MAE)