为什么使用MLU270在tensorflow框架下生成MLU220离线模型输出的每层张量的值完全相同呢？

图1为生成离线模型的输出结果，图2为在cpu上运行的输出结果

2021-12-21 22:27:55.821387: I tensorflow/core/platform/cpu_feature_guard.cc:142][1541] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX512F

CNRT: 4.10.1 a884a9a

2021-12-21 22:27:55.826509: I tensorflow/core/common_runtime/mlu/mlu_device.cc:657][1541] Current MLU_Visible_Devices Count: 1

2021-12-21 22:27:55.826936: I tensorflow/stream_executor/mlu/mlu_stream_executor.cc:459][1541] tf git version: v1.4.0-2-2-g0b0dcd73003

2021-12-21 22:27:55.826953: I tensorflow/stream_executor/mlu/mlu_stream_executor.cc:461][1541] Cambricon TensorFlow version: 1.4.0 was compiled at 2021-06-30 12:22:36.474000 with Toolkit: 1.7.3-2, CNNL: 1.3.0-2, CNML: 7.10.2-2, CNPLUGIN: 1.12.1-2

2021-12-21 22:27:55.826963: I tensorflow/stream_executor/mlu/mlu_stream_executor.cc:468][1541] Current library versions: DRIVER: 409, CNRT: 4.10.1, CNNL: 1.3.0, CNML: 7.10.2

2021-12-21 22:27:55.826996: I tensorflow/core/common_runtime/mlu/mlu_device.cc:792][1541] Created TensorFlow device (/job:localhost/replica:0/task:0/device:MLU:0 with 15884 MB memory) -> physical MLU (device: 0, name: MLU270)

2021-12-21 22:27:55.827012: W tensorflow/core/common_runtime/mlu/mlu_device.cc:181][1541] The core version from user is inconsistent with the core version from device !

2021-12-21 22:27:55.827022: W tensorflow/core/common_runtime/mlu/mlu_device.cc:182][1541] Using core version from user: MLU220

WARNING:tensorflow:From tf_forward.py:148: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.

2021-12-21 22:27:56.949274: I tensorflow/compiler/tf2mlu/convert/convert_graph.cc:998][1541] MLU fusion node my_mlu_op_0 added for segment 0 consisting of 246 nodes succeeded.

CNML: 7.10.2 ba20487

2021-12-21 22:28:04.898991: I tensorflow/stream_executor/mlu/mlu_api/fusion_op/fusion_op.cc:219][1556] Save FusionOp to offline model.

2021-12-21 22:28:04.903589: [cnrtWarning] [1556] [Card : 0] The device you run is not the same as the platform in kernel!

2021-12-21 22:28:04.903617: [cnrtWarning] [1556] [Card : 0] The device you run is MLU270, but the platform in kernel is MLU220!