Search Results for "position"

...We replace the old BERT vocabulary with a larger one of size 51271 built from the training data, in which we 1) add missing 6800+ Chinese characters (most of them are traditional Chinese characters); 2) remove redundant tokens (e.g. Chinese character tokens with ## prefix); 3) add some English tokens to reduce OOV. Position Embeddings We extend the max_position_embeddings from 512 to 1024. We initialize the new version of models with the old version of checkpoints with vocabulary alignment. Token embeddings found in the old checkpoints are copied. And other newly added parameters are randomly initialized. We further train the new CPT & Chinese BART 50K steps with batch size 2048, max-seq-length 1024, peak learning rate 2e-5, and warmup ratio 0.1. ...

Implementation of RQ Transformer, which proposes a more efficient way of training multi-dimensional sequences autoregressively. This repository will only contain the transformer for now. You can use this vector quantization library for the residual VQ. This type of axial autoregressive transformer should be compatible with memcodes, proposed in NWT. It would likely also work well with multi-headed VQ. I also think there is something deeper going on, and have generalized this to any number of...