(ais_bench) [HwHiAiUser@master scripts]$ python3 -m ais_bench --model=/home/HwHiAiUser/model/resNet101/Resnet101_for_TensorFlow/resnet101_tf_aipp.om --divide_input 1 --device 1,2,3 --input=/home/HwHiAiUser/dataset/ImageNet2012/ILSVRC2012_val --loop 1 --output ./output --output_dirname val --outfmt BIN
[INFO] multidevice:[1, 2, 3] run begin
[INFO] acl init success
[INFO] open device 0 success
[INFO] load model /home/HwHiAiUser/model/resNet101/Resnet101_for_TensorFlow/resnet101_tf_aipp.om success
[INFO] create model description success
[INFO] try get model batchsize:1
[INFO] unload model success, model Id is 1
[INFO] subprocess_0 main run
[INFO] subprocess_1 main run
[INFO] subprocess_2 main run
[INFO] acl init success
[INFO] acl init success
[INFO] acl init success
[INFO] open device 1 success
[INFO] open device 2 success
[INFO] open device 3 success
[INFO] load model /home/HwHiAiUser/model/resNet101/Resnet101_for_TensorFlow/resnet101_tf_aipp.om success
[INFO] load model /home/HwHiAiUser/model/resNet101/Resnet101_for_TensorFlow/resnet101_tf_aipp.om success
[INFO] create model description success
[INFO] create model description success
[INFO] output path:./output/val/device1
[INFO] output path:./output/val/device3
[INFO] load model /home/HwHiAiUser/model/resNet101/Resnet101_for_TensorFlow/resnet101_tf_aipp.om success
[INFO] create model description success
[INFO] output path:./output/val/device2
[ERROR] arg0 tensorsize: 75264 filesize: 1204224 not match
[ERROR] subprocess run failed error_callback:
[ERROR] arg0 tensorsize: 75264 filesize: 1204224 not match
[ERROR] subprocess run failed error_callback:
[INFO] unload model success, model Id is 1
[ERROR] arg0 tensorsize: 75264 filesize: 1204224 not match
[ERROR] subprocess run failed error_callback:
[INFO] unload model success, model Id is 1
[INFO] unload model success, model Id is 1
[INFO] multidevice run end qsize:0 result:1
[INFO] summary throughput:0
[INFO] end to destroy context
[INFO] end to reset device is 0
[INFO] end to finalize acl
1.问题描述
1、ais_bench 推理 batchsize=16 的 model.om 时正常。
2、推理 batchsize=1 的 model.om 错误,此时ais_bench 使用 batch size 依然为 16, 与输入 model.om 的 batch size 不一致,导致程序报错
2.错误 log
图像大小:224, 224, 3。
2.1、使用编译 batchsize=1 的 om 模型,错误 log