This repository stores implementations of Machine Learning and Deep Learning concepts, with the only objective of better understanding them. All implementations use basic elements of foundational libraries, such as pytorch (and not its TransformerEncoderLayer module for instance).
Each folder has its own README, containing a summary of the project and associated results.
- Transformer from scratch: an encoder-decoder transformer built with basic Pytorch blocs, for machine translation
- LoRA from scratch: scripts that compare the efficiency of fine-tuning a LM's head versus fine-tuning its Q, K and V projections with LoRA
- NeRF from scratch: (wip, to come in the following days)