cache text embedding in inference

[RalphHan]

@GuyTevet I've done a little bit change to the code to cache text embedding in inference

[GuyTevet]

Thanks! While this should not affect training mode, for readability purposes, I would also wrap this with if not self.training and add a comment that explains this mechanism. https://discuss.pytorch.org/t/check-if-model-is-eval-or-train/9395