When we add new tokens, this method automatically adds embedding using torch nn.Embedding.
https://huggingface.co/transformers/_modules/transformers/modeling_utils.html#PreTrainedModel.resize_token_embeddings
The documentation says the resized embeddings are nn. Embedding, which said they by default initialize weights from N(0, 1) (https://pytorch.org/docs/stable/generated/torch.nn.Embedding.html). But I have checked that the resized embedding weights are almost N(0, 0.01) or N(0, 0.02)? Can I check the true distribution of resized embedding weights ?