spencerbraun

Hi @raphaelreinauer, thanks for the suggestion, the training script is certainly incomplete. Your comment makes sense, and I agree my loss scheme is incorrect. With regards to the stop gradient,...

@sappersapper Thank you for the suggestion. I believe that will work, though might need to think about whether the memory overhead can be avoided.

Sorry as noted this is a work in progress and my focus has been on the model. The training script is just a sketch right now, but hope to remedy...

Hi, wondering if there is any update or fix here?