Adding base transformer implementation

[venelink]

Adding the base files from mini gpt (minor fixes on internal dependencies and a weird bug that I got on google collab, but not on local machine)

Adding two example notebooks: using pre-trained GPT for prompting and training a GPT model from scratch on a simple sequence prediction task inspired by POS tagging.

To do:

• Pretraining
• More use for prompting (e.g. zero-shot and/or few shot examples?)
• Add licensing stuff