多卡训练会卡吗？

[qinwangmm]

拿小数据集data_thchs30测试双卡训练，为啥workerlog.1不打日志啊？

Input size (MB): 0.55
Forward/backward pass size (MB): 67.88
Params size (MB): 134.09
Estimated Total Size (MB): 202.52
------------------------------------------------------------------------------------------------------

[2022-07-29 16:56:23.756062] 训练数据：13254
[2022-07-29 16:56:25.356384] 成功恢复模型参数和优化方法参数：models_thchs/deepspeech2/last_model

workerlog.0正常打印
 ======================================================================
[2022-07-29 17:07:20.092529] Test batch: [0/3], loss: 235.28400, cer: 0.97384
[2022-07-29 17:07:20.964818] Test epoch: 17, time/epoch: 0:01:16.713099, loss: 204.11594, cer: 0.97387
====================================================================== 

[2022-07-29 17:07:24.003397] 已保存模型：models_thchs/deepspeech2/epoch_17
[2022-07-29 17:07:27.692319] Train epoch: [18/63], batch: [0/103], loss: 227.22951, learning rate: 0.00014561, eta: 2:07:36

[yeyupiaoling]

workerlog.1 一般不会输出训练日志的，除非是对应的卡报错

[qinwangmm]

嗯呢，我测试了一下，workerlog.0结束了，GPU 占用全部也释放了，我不太确定，就问问你，谢谢