Feature model inference test fails with mkl-dnn
Created by: ddokupil
System information -PaddlePaddle version 1.5 -CPU: CLX/MKLDNN version 0.18, 0.19 and 0.20 -OS Platform Ubuntu 16.04 -Python version 2.7.12 -Cmake orders -DCMAKE_BUILD_TYPE=Release -DWITH_GPU=OFF -DWITH_MKLDNN=ON -DWITH_TESTING=ON -DWITH_PROFILER=ON -DWITH_STYLE_CHECK=OFF -DON_INFER=ON -DWITH_INFERENCE_API_TEST=ON To Reproduce ./paddle/fluid/inference/tests/api/test_analyzer_image_classification --infer_model= --gtest_filter=*compare_mkldnn Describe your current behavior I0926 13:38:32.002418 47701 analysis_predictor.cc:467] ======= optimize end ======= I0926 13:38:32.004854 47701 tester_helper.h:353] Thread 0, number of threads 1, run 1 times... E0926 13:38:32.005671 47701 analysis_predictor.cc:323] feed names from program do not have name: [] from specified input PROFILE: interrupts/evictions/bytes = 136/20/6424 I0926 13:38:33.358453 47701 helper.h:322] ====== threads: 1, thread id: 0 ====== I0926 13:38:33.358496 47701 helper.h:324] ====== batch size: 1, iterations: 1, repetitions: 1 ====== I0926 13:38:33.358501 47701 helper.h:326] ====== batch latency: 1352.88ms, number of samples: 1, sample latency: 1352.88ms, fps: 0.739166, data type: float ====== F0926 13:38:33.372910 47701 tester_helper.h:121] Check failed: std::abs(pdata_ref[j] - pdata[j]) <= FLAGS_accuracy (0.0237388 vs. 0.001)
Code to reproduce the issue Other info / logs