Andrei Moraru

What exactly have you changed in the code? Be wary of erasing things like ```python global best_bleu4, epochs_since_improvement, checkpoint, start_epoch, fine_tune_encoder, data_name, word_map ``` PEP will mark those as warnings,...

Have you tried [this fix](https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Image-Captioning/pull/125/commits/a94a8391cfb3ba9957f0c0f7ba13f2e468a3246c) instead?

That's true. They should be the same in the loss by using `.data`. Curios, is your loss just not decreasing, or is it getting worse?

Well, it's empty, so there are no complete sequences. You could catch it with an if to add the last incomplete sequence: ```python if len(complete_seqs) == 0: complete_seqs.extend(seqs.tolist()) # (s,...

@Leo-Thomas , yes the link to MSCOCO seems to be broken indeed. Here is how you can get the 2014 dataset You can run this in a bash script: ```bash...

Edit: I think there is an exact fix for this [here](https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Image-Captioning/pull/125/commits/a94a8391cfb3ba9957f0c0f7ba13f2e468a3246c)