I spend more time with tensors than I do with my friends. 👾
Used to reverse engineer multimodal LLMs at Aleph-Alpha.
📜 Recent work:
- AtMan - Understanding Transformer Predictions Through Memory Efficient Attention Manipulation
- DORA - method to explore outlier representations in Deep Neural Networks
⚡ Other stuff:
- Google Summer of Code 2020 @ INCF
- Intern at RunwayML
- Torch-dreams - Making neural networks more interpretable, for research and art.
- Devolearn - Data driven research on embryos with deep learning models.
Contact:
- Telegram
- Or drop an email: mayukhmainak2000@gmail.com