Complete improvements of download script.

[populated]

I modified the script to utilize data classes, JSON serialization, and the tqdm library, ensuring a seamless and informative data download process. It also offers options to specify data sizes, splits, and target example counts. (cool, cool!)

Little list of changes:

1. Added a data class (ChatData) for structuring GPT-related data.
2. Implemented a JSON encoder (ChatDataEncoder) for custom serialization.
3. Created a class (GPTData) to manage data download, processing, and saving.
4. Introduced methods for validating data sizes and splits.
5. Utilized tqdm for a progress bar during data download.
6. Provided options for truncating data based on a target example count.

Usage (I thought this was necessary, soooo):

gpt_data = GPTData(target_examples=None)
gpt_data.download_and_save_data(data_size_fn='webtext', split_fn='train')

Testing:

It works perfectly—I've tested all sizes and splits. I also tried various example sizes and all in general. It worked flawlessly on my local machine (Linux).