Add mechanisms to train.py so it can skip the pre-defined paramsif the pretrained model is provided (pretrain.py, train.py)

Merge pull request #9 from Mikyx-1/vietlh/change_split_mechanism_in_pretraining

Change the train/val split mechanism in pretrain.py (#8)

Merge pull request #7 from Mikyx-1/vietlh/add_pretraining

Make the pretrain.py minimal and move the tokenizer training to pretokenize.py

Add pretokenize.py to build .bin files for faster loading in pretraining stage

Simplify pretrain.py

Add *.zip to .gitignore

Add pretrain.py

Merge pull request #5 from Mikyx-1/vietlh/refactor_code

refactor the tokenizer.py

Delete the project folder structure in README.md

1st Refactorisation attempt

Update the docstring (model.py)

Merge pull request #4 from Mikyx-1/vietlh/kv_caching