当前位置：网站首页>[mindspore] [mindrecord] save the reading problem after specifying the floating-point precision

[mindspore] [mindrecord] save the reading problem after specifying the floating-point precision

2022-07-20 09:28:00 【Xiaole happy】

Mindspore 1.1 CPU adopt

1.0 Ascend Problems arise

In preservation mindrecord when , Specify data float64, But the data itself is float32, stay 1.0 ascend Let's follow float32 preservation , But reading is not float64 have a problem . Possible codes are bug.

The problem is pr：fix ndarray field without type in mindrecord · Pull Request !9306 · MindSpore/mindspore - Gitee.com  It's been fixed in , Will be in API Internally, the data actually written by the user is compared with the defined schema Type conversion , In this way schema The type defined shall prevail . The above problem will not appear after reading it .

原网站

版权声明
本文为[Xiaole happy]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/201/202207190511432828.html

边栏推荐

猜你喜欢

随机推荐