Script for distilling zero-shot classifier to more efficient student (#10244)
* add zero-shot distillation script
* readme wordsmithing
* clean up code
* add multi-gpu teacher inference
plus tidying up more code
* add use_fast_tokenizer arg
* update results in readme
* more readme wordsmithing
* style
* Add handle to readme
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* fix code block
* add error+docs about distributed & tpu
* add @sgugger format requests
* xla -> tpu
* support fp16 for teacher preds
* no checkpoint by default
* add demo colab link
* add model sharing prompt + model link
* correct resulting acc of example
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>