Jiaojiao Fan

Hi! I find there could be a possible bug in the function [generate_samples()](https://github.com/lucidrains/DALLE2-pytorch/blob/v0.26.1/train_decoder.py#L159) of train_decoder.py file. If text embeddings exist, then this would input text embedding in the decoder sampling...

Hi! You mentioned the numpy array in image embedding `.npy` files generated from [clip_retrieval](https://github.com/rom1504/clip-retrieval) lost the index correspondence to the original image `.tar` files. I wonder are they completely irrelevant...