6acc103ea1f132846020209c3eff5646077e55cd,src/plugin.py,EmbeddingRegularizer,__init__,#EmbeddingRegularizer#,10

Before Change

            self.dim = emb_dim

            if bert is not None:
                if not isinstance(bert, (tuple, list)) or len(bert) != 2:
                    raise ValueError("`bert` should be a tuple/list of config and fine-tuned model "
                                     "such as (\"bert-base-uncased\", \"fine-tuned-model.pth\").")
                self.emb_table = BertEmbeddingPredictor(
                    bert[0], text_encoder, bert[1])
            else:
                self.emb_table = nn.Embedding.from_pretrained(

After Change

        if enable:
            if bert is not None:
                self.use_bert = True
                if not isinstance(bert, str):
                    raise ValueError("`bert` should be a str specifying bert config such as \"bert-base-uncased\".")
                self.emb_table = BertEmbeddingPredictor(bert, tokenizer, src)
                vocab_size, emb_dim = self.emb_table.model.bert.embeddings.word_embeddings.weight.shape
                vocab_size = vocab_size-3 // cls,sep,mask not used
                self.dim = emb_dim
            else:
                self.use_bert = False

Instances