DistilBERT preprocessor bug - Githubissues

jeongukjae / huggingface-to-tfhub

Converting Huggingface Models to TFHub

5 stars 0 forks source link

DistilBERT preprocessor bug #4

Closed jeongukjae closed 3 years ago

jeongukjae commented 3 years ago

If seq_length argument is passed, token_type_ids that DistilBert does not require is returned.

Reproduce: https://colab.research.google.com/drive/1RSpk-A5s2Ogna6_w4RzRODeYTJshkhqY?usp=sharing

[x] fix bug https://github.com/jeongukjae/huggingface-to-tfhub/commit/5e65fedfc4f21ff491ec46d06ad9355f6aa06315
[x] upload to tfhub https://github.com/tensorflow/tfhub.dev/pull/94

jeongukjae commented 3 years ago

This issue is fixed by this commit (https://github.com/jeongukjae/huggingface-to-tfhub/commit/5e65fedfc4f21ff491ec46d06ad9355f6aa06315) and tested by this script (https://github.com/jeongukjae/huggingface-to-tfhub/blob/main/test_exported_bert_like_models.py)