/flow-ai

seq2seq transformer - o9Hacks

Primary LanguagePythonMIT LicenseMIT

flow-ai

seq2seq transformer - o9Hacks

Setup

  • go to https://pytorch.org/get-started/locally/
  • install mentioned version of cuda (11.8)
  • install anaconda, create and activate a conda environment
  • install torch using the command given in step 1.
  • install rest of the packages mentioned in requirements.txt
  • update config.py file if needed
  • run train.py => for training (if u want to pause and resume the training, or loading a pretrained model => make sure to update config.py)
  • run pytorchGPUTest.py => to test CUDA/gpu is detected
  • run predict.py => for prediction (need a trained model for this to work)

Dataset
https://jsoneditoronline.org/#left=cloud.18e44cbbed4e4a6c919ec80711694750

llamascript

Setup

  • have the model running in any notebook eg Google Collab
  • create a folder context_data/data and upload the o9trainingset.txt file