Shark-NLP / DiffuSeq

[ICLR'23] DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models
MIT License
711 stars 87 forks source link

BERT parameter #65

Open X-fxx opened 10 months ago

X-fxx commented 10 months ago

In your code, I found that the value of the parameter init_pretrained is "no". Does this mean that during the model training process, you did not use the pre-trained parameters of the BERT model, but trained a BERT model from scratch? If you are not using the BERT model and I need to use the parameters of the BERT model, what modifications do I need to make?