Support **kwargs for PyTorch models.

[damccorm]

Some models in Pytorch instantiating from torch.nn.Module, has extra parameters in the forward function call. These extra parameters can be passed as Dict or as positional arguments. 

Example of PyTorch models supported by Hugging Face -> https://huggingface.co/bert-base-uncased

Some torch models on Hugging face

Eg: https://huggingface.co/docs/transformers/model_doc/bert#transformers.BertModel

inputs = {
     input_ids: Tensor1,
     attention_mask: Tensor2,
     token_type_ids: Tensor3,
}

model = BertModel.from_pretrained("bert-base-uncased") # which is a  
# subclass of torch.nn.Module

outputs
= model(**inputs) # model forward method should be expecting the keys in the inputs as the positional
arguments.

 

Transformers integrated in Pytorch is supported by Hugging Face as well. 

 

Imported from Jira BEAM-14337. Original Jira may contain additional context.
Reported by: Anand Inguva.
Subtask of issue #21435

[damccorm]

Unable to assign user @yeandy. If able, self-assign, otherwise tag @damccorm so that he can assign you. Because of GitHub's spam prevention system, your activity is required to enable assignment in this repo.

[yeandy]

@damccorm I'm unable to self-assign. Is it because I don't have write permissions? Is there a GH action that can allow me to self-assign future issues?

[damccorm]

One of the unfortunate downsides of moving to issues is that non-committers can't self-assign/triage issues that they didn't create by default. I think that's a workflow we'll need to support going forward though, I'll look into it. For the moment, I'd recommend commenting/declaring intent to work on any issues.

I have access through

beam/.asf.yaml

Line 39 in 4dce7b8

- damccorm

which gives triage permissions, unfortunately that's limited to 20 people though

[damccorm]

Once #21719 is in, you should be able to do this with chat-op commands

[yeandy]

Got it, thanks!

[tvalentyn]

we decided to move forward with a separate param: #21806