Environment info

• transformers version: 4.11.2
• Platform: Linux-5.4.0-66-generic-x86_64-with-glibc2.10
• Python version: 3.8.8
• PyTorch version (GPU?): 1.9.1+cu102 (True)
• Tensorflow version (GPU?): not installed (NA)
• F

Currently max_epochs defaults to 1000:

If both max_epochs and max_steps aren't specified, max_epochs will default to 1000. To enable infinite training, set max_epochs = -1.

As a user, though, I would expect that if I don't specify a specific ending point, the training would continue indefinitely. In my own experiments, when the training cut off at 999 epochs, I was confused, and googling t

Change tensor.data to tensor.detach() due to
pytorch/pytorch#6990 (comment)
tensor.detach() is more robust than tensor.data.

🙌🙌🙌🙌🙌

Bug Report

Describe the bug

An ONNX model with QuantizeLinear (opset 13) passes shape inference when it has no input for zero_point, but fails if the input is the empty string. Trailing optional inputs should be able to be expressed in either way.

onnx version 1.10.1

Is your feature request related to a problem? Please describe.
I typically used compressed datasets (e.g. gzipped) to save disk space. This works fine with AllenNLP during training because I can write my dataset reader to load the compressed data. However, the predict command opens the file and reads lines for the Predictor. This fails when it tries to load data from my compressed files.

Motivated by huggingface/transformers#12789 in Transformers, one welcoming change would be replacing assertions with proper exceptions. The only type of assertions we should keep are those used as sanity checks.

Currently, there is a total of 87 files with the assert statements (located under datasets and src/datasets), so when working on this, to manage the PR s