How Dendrite can support viewing and editing of ONNX with Transformer model architecture

[GeorgeS2019]

Refer to these links
fel88/Dagre.NET#2

[fel88]

@GeorgeS2019 Do you know where I can download seq2seq ONNX? Because BERT doesn't have a decoder part

I've implement some experimental logic to group nodes via prefix

Sample 1 (groups: bert/embeddings, attention, output, intermediate):

Sample 2 (groups: bert/embeddings, bert/encoder/layer_):

[GeorgeS2019]

I hope I mark the 12 encoder blocks.

Suggestion: Create a subgraph or block that organize the 12 encoder blocks (make them equal size) within one block.

One step at a time. I too am learning :-)

FYI: use this as a reference: Explanation of BERT Model – NLP