GitLab

* use accelerator to check mixed_precision

* default `mixed_precision` to `None`

* pass mixed_precision to accelerate launch