Upload loss_vs_step.png

Update README.md

Create README.md

Upload V2: stackformer GPT-2 + sparse cross-attention vision model (128 visual tokens)

initial commit