Quest: Curriculum and Preprocessing for Translation Targets

[gkielian]

This would start with pretraining on multilingual corpus, comparing phonemization with regular tiktoken.

• phonemization and train/val.bin with pre-training datasets
• phonemization and train/val.bin with of fine-tuning
• tiktoken of train/val.bin
• tiktoken of fine-tuning

For now let us target English <-> Japanese language training pairs.