[ONNX] RNN activation function exported to ONNX with wrong string case

[spandantiwari]

🐛 Bug

If we export nn.RNN module to ONNX, the activations attribute is written is all lower case. Several backends assume these strings to be case-sensitive, where this can be a blocking issue during runtime. There is an open issue (onnx/onnx#1934) in the ONNX repo to make this this explicit in the spec. PyTorch's ONNX exporter should also consider writing strings with correct case in activations attribute.

To Reproduce

Steps to reproduce the behavior:

1. Create a simple model with nn.RNN module.
2. Export this model to ONNX using torch.onnx.export
3. Check the activations attribute for the case of the string. As an example, the string will be 'tanh', where ideally it should be 'Tanh'.

Expected behavior

The activations attribute value should match the case specified in ONNX spec.

Environment

This can be seen in all environments.

[spandantiwari]

cc: @houseroad This is a minor issue, but one that's blocking some models from running in backends such as Onnxruntime.

Will submit a PR soon.

[spandantiwari]

#19368 is in to fix this.

[houseroad]

Close it, since the fix is merged.