Created by: LilyCaiZL

为什么我用autodl的时候，跑出来的loss和acc是一样的值？同样的代码我放在aistudio上面，完全没问题，acc和loss是不一样的，acc大，loss就小很多，请问能否帮忙解决，我是昨天下的最新的paddlepaddle和paddlehub [2020-05-27 23:57:20,855] [ TRAIN] - step 3320 / 3429: loss=0.95312 acc=0.95312 [step/sec: 3.04] [2020-05-27 23:57:24,215] [ TRAIN] - step 3330 / 3429: loss=0.96875 acc=0.96875 [step/sec: 3.04] [2020-05-27 23:57:27,576] [ TRAIN] - step 3340 / 3429: loss=0.97188 acc=0.97188 [step/sec: 3.04] [2020-05-27 23:57:31,104] [ TRAIN] - step 3350 / 3429: loss=0.98958 acc=0.98958 [step/sec: 3.00] [2020-05-27 23:57:34,466] [ TRAIN] - step 3360 / 3429: loss=0.96250 acc=0.96250 [step/sec: 3.04] [2020-05-27 23:57:37,835] [ TRAIN] - step 3370 / 3429: loss=0.97500 acc=0.97500 [step/sec: 3.03] [2020-05-27 23:57:41,198] [ TRAIN] - step 3380 / 3429: loss=0.97813 acc=0.97813 [step/sec: 3.04] [2020-05-27 23:57:44,691] [ TRAIN] - step 3390 / 3429: loss=0.96875 acc=0.96875 [step/sec: 3.25] [2020-05-27 23:57:48,049] [ TRAIN] - step 3400 / 3429: loss=0.97813 acc=0.97813 [step/sec: 3.04] [2020-05-27 23:57:51,471] [ TRAIN] - step 3410 / 3429: loss=0.97188 acc=0.97188 [step/sec: 2.99]

`==> round0/log-0.info <== [2020-05-27 14:17:51,179] [ EVAL] - best model saved to ckpt_20200527140507/best_model [best acc=0.31413] [2020-05-27 14:17:52,471] [ INFO] - Load the best model from ckpt_20200527140507/best_model [2020-05-27 14:17:52,758] [ INFO] - Evaluation on test dataset start [2020-05-27 14:17:56,026] [ EVAL] - [test dataset evaluation result] loss=0.39270 acc=0.39270 [step/sec: 5.92] [2020-05-27 14:17:56,027] [ INFO] - Saving model checkpoint to ckpt_20200527140507/step_4100 [2020-05-27 14:17:57,099] [ INFO] - PaddleHub finetune finished. [2020-05-27 14:17:57,100] [ INFO] - Try loading checkpoint from ckpt_20200527140507/ckpt.meta [2020-05-27 14:17:57,397] [ INFO] - PaddleHub model checkpoint loaded. current_epoch=51, global_step=4100, best_score=0.31413 [2020-05-27 14:17:57,398] [ INFO] - Evaluation on dev dataset start [2020-05-27 14:18:00,638] [ EVAL] - [dev dataset evaluation result] loss=0.31413 acc=0.31413 [step/sec: 5.97]

==> round0/log-10.info <== [2020-05-27 18:05:31,993] [ EVAL] - best model saved to ckpt_20200527172834/best_model [best acc=0.83658] [2020-05-27 18:05:33,126] [ INFO] - Load the best model from ckpt_20200527172834/best_model [2020-05-27 18:05:33,401] [ INFO] - Evaluation on test dataset start [2020-05-27 18:05:36,953] [ EVAL] - [test dataset evaluation result] loss=0.88527 acc=0.88527 [step/sec: 2.60] [2020-05-27 18:05:36,954] [ INFO] - Saving model checkpoint to ckpt_20200527172834/step_5075 [2020-05-27 18:05:37,949] [ INFO] - PaddleHub finetune finished. [2020-05-27 18:05:37,949] [ INFO] - Try loading checkpoint from ckpt_20200527172834/ckpt.meta [2020-05-27 18:05:38,226] [ INFO] - PaddleHub model checkpoint loaded. current_epoch=146, global_step=5075, best_score=0.83658 [2020-05-27 18:05:38,226] [ INFO] - Evaluation on dev dataset start [2020-05-27 18:05:41,438] [ EVAL] - [dev dataset evaluation result] loss=0.83658 acc=0.83658 [step/sec: 2.90] `