ProGamerGov

So, MSELoss() is [implemented](https://pytorch.org/docs/stable/_modules/torch/nn/functional.html#mse_loss) as: `((input-target)**2).mean()`. When I combine `MSELoss()` with `mean(0).mean(0)` for content and style loss, I get what looks like DeepDream hallucinations. Both images had `.permute(2, 1, 0)`...

@momo-the-monster Nice work! I'm not sure about using a ton of if statements to block all the print statements. It's fine to use that once in CaffeLoader to block printing...

I'm not sure if such a tool would be possible. I would suggest finding a set of parameters you like, and then testing those parameters with a single 1920x1080 image....

@peri044 Your code fails because you are doing `x + None`. If you add an if statement to prevent that, then it will work in TorchScript.

@parmeet Ah, okay. I put together a helper function to convert the outputs to the proper format for the CLIP models: ``` def token_str_to_tensor( token_list: Union[List[str], List[List[str]]], content_length: int =...

@parmeet So, it looks like the ToTensor transform doesn't allow padding to be greater than the given list length. It would require concatinating a set of zeros (in the case...

Ideally one could do something like this to avoid having to deal with differing file paths on different computers: ``` url_path = # Download URL for bpe merges. clip_merges =...

cc @abhinavarora

If this is a bug, it seems like it goes beyond unicode. I fed the TorchText and OpenAI CLIP tokenizers the bpe merges file as an input, and they deviate...

I was looking at only the unicode differences, and the found the following. The single character 'ij', is encoded by OpenAI's tokenizer with the same tokens for both an `i`...