Closed leileilei2000 closed 3 months ago
都差不多吧
导我不管是MAX_pool3d还是max_pool3d,在yolov5都报错了max_pool3d_with_indices_backward_cuda does not have a deterministic implementation, but you set ,
这个按照他提示设置就行了,跟这些没关系
我按照设置运行了,但是就是报错max_pool3d_with_indices_backward_cuda does not have a deterministic implementation, but you set,我找到一种方法是torch.use_deterministic_algorithms(False),关掉这个决定性算法,但是有人说会非常影响结果
------------------ 原始邮件 ------------------ 发件人: "z1069614715/objectdetection_script" @.>; 发送时间: 2024年5月30日(星期四) 晚上6:23 @.>; 抄送: " @.**@.>; 主题: Re: [z1069614715/objectdetection_script] 导现在yolov5-asf里的MAX_pool3d已经改掉了,好像变成nn.functional.max_pool3d,这个了,然后在kernel_size基础上多了个input (Issue #44)
这个按照他提示设置就行了,跟这些没关系
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
没有非常影响结果的说法,解决这问题的唯一方法就是torch.use_deterministic_algorithms(False),不然就不用这个asf
好的导!
------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2024年5月30日(星期四) 晚上6:30 收件人: @.>; 抄送: " @.>; @.>; 主题: Re: [z1069614715/objectdetection_script] 导现在yolov5-asf里的MAX_pool3d已经改掉了,好像变成nn.functional.max_pool3d,这个了,然后在kernel_size基础上多了个input (Issue #44)
没有非常影响结果的说法,解决这问题的唯一方法就是torch.use_deterministic_algorithms(False),不然就不用这个asf
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
都差不多吧