> Unlike GPT, which only uses encoder blocks, or BERT, which only uses decoder blocks I think you meant the reverse, autoregressive model like GPT uses only th

> Unlike GPT, which only uses encoder blocks, or BERT, which only uses decoder blocks I think you meant the reverse, autoregressive model like GPT uses only the decoder while autoencoder like BERT only uses the encoder.