Training lstm and transformer models. cs182

env creation for running locally:

  • recommend running in google colab so we don't have to create a local environment, the instructions below will get you most of the way there if you do choose to use a local environment, but there may be some additional packages that need to be installed

  • install pytorch and torchvision from pytorch channel

  • then install (some redundancy with pytorch but is okay): numpy matplotlib pillow scipy h5py scikit-image cython imageio jupyter nltk

  • I used mamba, and python 3.11 (installed by pytorch by default)

potential issues

  • running these notebooks is quite slow depending on hardware

  • if certain data files are missing, may need to run certain bash scripts in the /datasets directories

  • for cython issues, run the following from /deeplearning directories and restart kernel: python setup.py build_ext --inplace

  • for character encoding issues visit charset_decoder.py file