使用Ascend PyTorch Profiler接口采集不到trace数据
收藏回复举报
使用Ascend PyTorch Profiler接口采集不到trace数据
t('forum.solved') 已解决
新人帖
发表于2024-05-11 10:31:09
0 查看

采集代码如下:

 

 # 配置采集参数 

    experimental_config = torch_npu.profiler._ExperimentalConfig( 

        export_type=torch_npu.profiler.ExportType.Text, 

        aic_metrics=torch_npu.profiler.AiCMetrics.PipeUtilization, 

        profiler_level=torch_npu.profiler.ProfilerLevel.Level1, 

        l2_cache=False, 

        data_simplification=False 

    ) 

    with torch_npu.profiler.profile( 

            activities=[torch_npu.profiler.ProfilerActivity.CPU, torch_npu.profiler.ProfilerActivity.NPU], 

            schedule=torch_npu.profiler.schedule(wait=0, warmup=1, active=1, repeat=1, skip_first=0), 

            on_trace_ready=torch_npu.profiler.tensorboard_trace_handler("./result_{}".format(epoch)), 

            record_shapes=True, 

            profile_memory=True, 

            with_stack=True, 

            with_flops=False, 

            with_modules=False, 

            experimental_config=experimental_config 

    ) as prof: 

        # for i, (images, target) in enumerate(train_loader)及其包含的代码向后缩进 

        for i, (images, target) in enumerate(train_loader): 

......

运行日志如下:

[W AclInterface.cpp:181] Warning: 0Failed to find function aclrtCreateEventExWithFlag (function operator()) 

Epoch: [0][ 0/67]       Time 30.974 ( 0.000)    Data 16.008 ( 0.000)    Loss 6.9359e+00 (0.0000e+00)    Acc@1   0.20 (  0.00)   Acc@5   0.20 (  0.00) 

Epoch: [0][ 1/67]       Time 22.673 ( 0.000)    Data 22.030 ( 0.000)    Loss 6.9379e+00 (0.0000e+00)    Acc@1   0.00 (  0.00)   Acc@5   0.00 (  0.00) 

2024-05-11 02:22:01 [INFO] [523971] profiler.py: Start parsing profiling data: /home/yhui/ModelZoo-PyTorch/PyTorch/built-in/cv/classification/ResNet50_for_PyTorch/result_0/master_510670_20240511022107971_ascend_pt 

2024-05-11 02:22:03 [WARNING] [523971] profiler.py: 442 memory record(s) are incomplete. 

2024-05-11 02:22:07 [ERROR] [524621] profiler.py: Failed to get acl to npu flow events. 

2024-05-11 02:22:08 [ERROR] [524882] profiler.py: Failed to get acl to npu flow events. 

2024-05-11 02:22:11 [INFO] [524058] profiler.py: CANN profiling data parsed in a total time of 0:00:09.193205 

2024-05-11 02:22:11 [ERROR] [525290] profiler.py: Failed to get acl to npu flow events. 

2024-05-11 02:22:11 [INFO] [523971] profiler.py: All profiling data parsed in a total time of 0:00:10.011566 

我要发帖子