My personal notes on sequence modelling using deep learning. Contains a mixture of theory in notes
and implementations in src
and/or notebooks
.
src
: Python modules containing models and helper functionsnotebooks
: Implementations and demos. For more complex architectures, such as the transformer, notebooks can be considered as entry points to the code insrc
, and so may be quite minimal. Seesrc
for implementation details, andnotes
for theory.notes
: My personal notes on sequence modelsfigures
: Pictures fornotes
andnotebooks
This project's environment was set up for the CPU only, as I'm focusing on understanding architecture rather than performance here.
If you're running on Windows, you can try using the environment.yml
included with the project. This can be a bit flaky with complicated dependencies, at least in my experience. Otherwise, go to the pytorch installation GUI for the appropriate command, and then once pytorch is installed, then use the environment.yml
as a guide for the rest of the dependencies. You can delete problematic dependencies if you encounter errors. Use mamba!
conda env export --from-history > environment.yml