四、日志信息: 关闭融合规则+开启混合精度的日志和代码 [日志和代码](https://e-share.obs-website.cn-north-1.myhuaweicloud.com?token=Wnjb7TvuAlOmR+uCdf5qTKGcdg3cV

44546377&Signature=/kHIXwxTw9P5AkBhFwgREaE%2B83o%3D 执行 ac -c后打印如下 2022-01-11 18:46:51 (7324) -[DEBUG]Find ['result_20220111171922

pg_type p2 ON (p1.typarray = p2.oid) WHERE p1.typarray 0 AND (p2.oid IS NULL OR p2.typelem p1.oid OR p2.typlen -1); ! oid | basetype

73660.png "npu性能.png") npu训练 URL: https://e-share.obs-website.cn-north-1.myhuaweicloud.com?token=Bh8UE6r7gI5FCDRHmTHXaIMx00Zeg

``` mkdir -p /root/rpmbuild/SOURCES/ wget https://github.com/facebook/zstd/archive/v1.3.6.tar.gz#/zstd-1.3.6.tar.gz -O /root/rpm

NYlDbyLjzR1AWoH+lZc08rMGe0705XCuqOYNJrVGxz8P9jizO8FgezX3uyOO9sivdLsIMECDi9xqPzYU/O2+9ayl7sPv3y2eFdu2Ji+gwSnkCEfd1PKctAZDvvo8cG

**日志提供方式:** NPU_break_train.log 为NPU上训练中断记录： https://smith.obs.cn-north-4.myhuaweicloud.com:443/tmp_log4issue/NPU_break_train.log

43335.png "屏幕截图.png") 但代码中并没有map、batch、map_and_batch算子；此外，代码是单P训练。 ngnn算子耗时分析结果如下：（已开启混精） ![输入图片说明](https://images.gitee.com/u

Av58NlQyhhSE6fPUsqDmbB+2Vxu7u909/TaKR9P0QJ43ng0d+lse+7Bz+F2+JvtqXHSKVFCaGuON7+zU7h0Sj0o9f8dMnxQVXESeqidpZvhVQWcZKPiWRVL1PZXRPC

-DENABLE_DVPP_INTERFACE -std=c++11 -fPIC -Wall -O2 -o ../atlas200dk//libatlasutil.so ../atlas200dk//obj/camera.o -Wl,--whole-archive -Wl,--no-whole-archive

sd_inception_v2.pb" 原始模型输入为动态shape： https://data-exchange.obs.cn-north-4.myhuaweicloud.com:443/Models/ssd_inception_v2.pb?Acce