Closed zhangming8 closed 1 year ago
Thank you for your question. Yes, we believe it should be possible to use the weights trained by the official repository. Although we have added some adjustable parameters to our code, we have not modified the model architecture or retrained the official weights. Therefore, the official weights should still be compatible with our code.
Yeah, got it, thanks
Hi, thanks for your greate work. I found that your "aot" folder is different with official repo (https://github.com/yoxu515/aot-benchmark/tree/main) Such as, your add "max_len_long_term" param in your repo. So, can the weight trained by the official repo be used here?