model infer时infer_results中出现nan
Created by: neozhangthe1
使用自有数据训练的模型,在预测过程中时常报以下错误
* Aborted at 1516804178 (unix time) try "date -d @1516804178" if you are using GNU date *
PC: @ 0x0 (unknown)* SIGFPE (@0x7fe13bdd2f36) received by PID 8361 (TID 0x7fe12114b700) from PID 1004351286; stack trace: *
经过debug发现问题出在https://github.com/PaddlePaddle/DeepSpeech/blob/dad1c2727e375beb05dd747b06b04d0438e4a78d/model_utils/model.py#L197
对于特定的utterance,infer_results中会出现nan,具体表现是infer得到的向量值随着time step减小,最终变成nan
其中每行为time step以及对应的向量