Note that if you use custom callbacks that override `on_train_batch_end()`, you are at risk of forcing synchronous logging. Make sure to make your callbacks async-safe, and annotate them as such by setting `self.async_safe = True`

→ View original post on X — @fchollet,