complete torch model training program

[danaiamirali]

Completed model training program with:

• [x] Dataset cleaning (removing comments), filtering (ignoring non data science *.py files), and fetching
• [x] Dataset iterator that supports chunking large files
• [x] Model training that supports saving / loading from checkpoints
• [x] Retraining tokenizer
• [x] Model evaluation using text generation metrics like perplexity and more conventional metrics like loss, accuracy, and F1
• [x] Command line args for size of dataset to use, training split, number of epochs, checkpointing, whether to use retrained tokenizer

[danaiamirali]

@USSiamaboat @michaeljcliao @Maxim12313 Merge your work into this branch to combine it once it's completed

[danaiamirali]

merging dev-2 instead of dev