AttributeError: 'tuple' object has no attribute 'size'

[lan2720]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Steps to reproduce the behavior:

from transformers import BertModel
from torchinfo import summary

bert_base_path =  '/my/local/path/to/bert-base-chinese'

teacher_encoder = BertModel.from_pretrained(bert_base_path)
teacher_encoder.config.output_hidden_states = True
# teacher_encoder.config.num_labels = len(config["ner_labels"])
summary(teacher_encoder, input_size=[(2,512), (2,512), (2,512)],
        dtypes=['torch.IntTensor', 'torch.IntTensor', 'torch.IntTensor'],
        device='cpu')

• Version torchinfo 1.5.4(pip install)

[TylerYep]

This is happening because torchinfo currently does not have a sophisticated way of handling dicts containing tensors.

Would appreciate PRs improving this functionality! The calculate_size function is where the changes should occur.

[scaomath]

torchinfo/torchinfo/layer_info.py

Lines 127 to 134 in ed0a7aa

	elif isinstance(inputs, dict):
	# TODO avoid overwriting the previous size every time
	size = []
	elem_bytes = list(inputs.values())[0].element_size()
	for _, output in inputs.items():
	size = list(output.size())
	if batch_dim is not None:
	size = [size[:batch_dim] + [1] + size[batch_dim + 1 :]]

@TylerYep What is the current dev roadmap for this function calculate_size? Would you take a PR that adds this functionality supporting dicts containing lists or tuples?

[TylerYep]

Yes, PRs are welcome (but check the current Pull Requests first, there may be other contributors working on the exact same functionality)