search

microsoft / nnfusion

A flexible and efficient deep neural network (DNN) compiler that generates high-performance executable from a DNN model description.
MIT License
958 stars 162 forks source link

[BUG] Unsupported Operator Prod, Einsum, and BroadcastTo in Tensorflow Frozen PB model format #425

Open donglinz opened 2 years ago

donglinz commented 2 years ago

🐛 Bug

NNFusion failed in optimizing BERT model from both Huggingface and TFHub. Missing "Prod" operator support from Huggingface BERT:

image

Missing "Einsum" and "BroadcastTo" operator support from TFHub BERT model:

image

To Reproduce Steps to reproduce the behavior:

  1. Froze BERT model from Huggingface(https://huggingface.co/bert-base-uncased) or TFHub(https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12)
  2. Optimize it with NNfusion