paddle.logsumexp对于大矩阵存在bug

rm5edbpk  于 2022-12-31  发布在  其他
关注(0)|答案(2)|浏览(137)

bug描述 Describe the Bug

  • paddle版本develop。

在jupyter notebook中初始化一下这个tensor!矩阵太大了需要分配一段时间。

import paddle, time
tensor = paddle.randn((1, 4096, 1024))

下面的代码在jupyter notebook中循环点击运行N次,发现结果正常。

def logsumexp(x, axis=0, keepdim=False):
    return paddle.log(x.exp().sum(axis=axis, keepdim=keepdim))

start = time.time()
for i in range(100):
    logsumexp(tensor, axis=1, keepdim=True)
end = time.time()
print("custom logsumexp", end-start)

下面的代码在jupyter notebook中循环点击运行N次,发现程序卡住了?感觉好像卡在显存分配上????前几步训练还正常,但是你循环运行N次后代码就挂了。

start = time.time()
for i in range(100):
    paddle.logsumexp(tensor, axis=1, keepdim=True)
end = time.time()
print("paddle logsumexp", end-start)

其他补充信息 Additional Supplementary Information

No response

iecba09b

iecba09b1#

您好,我们已经收到了您的问题,会安排技术人员尽快解答您的问题,请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时,您也可以通过查看 官网API文档常见问题历史IssueAI社区 来寻求解答。祝您生活愉快~

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the APIFAQGithub Issue and AI community to get the answer.Have a nice day!

uplii1fm

uplii1fm2#

已反馈给相关同学。

相关问题