[
https://issues.apache.org/jira/browse/BEAM-14337?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17533065#comment-17533065
]
Anand Inguva commented on BEAM-14337:
-------------------------------------
Yes, for boolean parameters, I think passing them as side inputs makes more
sense.
If there is no batch size, we assume that's it is a single sample and we add
extra dimension as you mentioned.
We always assume the batch_size to be the first dimension. I never encountered
a case where the batch size is different from the first dimension. Although,
the channel dimension may vary(but that gets handled by the user).
> Support **kwargs for PyTorch models.
> ------------------------------------
>
> Key: BEAM-14337
> URL: https://issues.apache.org/jira/browse/BEAM-14337
> Project: Beam
> Issue Type: Sub-task
> Components: sdk-py-core
> Reporter: Anand Inguva
> Assignee: Andy Ye
> Priority: P2
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> 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|https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py]
> Eg:
> [https://huggingface.co/docs/transformers/model_doc/bert#transformers.BertModel]
> {code:java}
> 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.{code}
>
> [Transformers|https://pytorch.org/hub/huggingface_pytorch-transformers/]
> integrated in Pytorch is supported by Hugging Face as well.
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
