Closed reborn-113 closed 3 years ago
Hi, I have forked this repository from https://github.com/chaitjo/character-level-cnn The original repo is https://github.com/mhjabreel/CharCnn_Keras Please raise this issue there, they might be better suited to answer your question.
Hi,author: When I recreated your network, I found, In GPU FP32 mode, loss does not converage occasionally during training. Afier the first epoch, it lasted at 1.3864. Then, the loss jitter occurs.