Keeping the vocab to 99.

Update the tokenizer for 4.17.

Adding tf weights.

Removing all merges.

Going the other way and simplifying the tokenizer instead.

Revert "Making the vocab_size match the tokenizer vocab_size."

Making the vocab_size match the tokenizer vocab_size.

crop_size is after resize ??

Adding pipeline_tag.

Making processor + model match.

initial commit