Fix setting correct repo id when pushing dataset to hub

[chrislee973]

Description

In the dataset preprocessing step, pass the dataset hash as the config_name parameter in push_to_hub if push_dataset_to_hub is set. The config_name parameter is used for handling different versions/configurations of a dataset, which fits our case here.

Motivation and Context

The push_dataset_to_hub config setting takes a value in the form <user>/<dataset_name>, but the previous implementation was pushing to hub with the id f"{cfg.push_dataset_to_hub}/{ds_hash}", which was causing an error because <user>/<dataset_name>/<ds_hash> is an invalid hf repo id.

How has this been tested?

• With push_dataset_to_hub set, ran the dataset preprocessing step several times with different dataset settings each time.
• Confirmed that this successfully pushed different variations of the dataset to hf hub (screenshot)
• On subequent dataset preprocessing runs with the same settings, confirmed that we successfully pulled down the variation of that dataset from hf hub.
  

Types of changes

• Dataset preprocessing step

[winglian]

Thanks @chrislee973 The original intention was to "cache" the pre-processed dataset to its own dataset, but this is much better. Were you able to verify that it loads the subdirectory only when there are 2 or more datasets uploaded and you remove the local dataset cache ?

[winglian]

@chrislee973 Thanks for the PR. Just tested. good to go!