Hi there
I have been trying to train a model as per the [Robust Speech Event](https://github.com/neelan-elucidate-ai/transformers/tree/main/examples/research_projects/robust-speech-event)
Training begins but seems to fill up /dev/sdb1
location which has a total of 119 GB. This happens even when i force the cache to be on a different drive as suggested here : both manually in the functions and via env var.
This happens regardless of the dataset used (at fist I thought it was an issue with "load_dataset("mozilla-foundation/common_voice_7_0", "en")
I suspect the very large model and data need a very big tmp location but it should be clearing the tmp not filling it up.
Any ideas @patrickvonplaten @osanseviero @anton-l ?