mindstudio-probe 1.0.3__py3-none-any.whl → 1.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/LICENSE +201 -201
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/METADATA +36 -34
- mindstudio_probe-1.1.0.dist-info/RECORD +287 -0
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/WHEEL +1 -1
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/entry_points.txt +1 -0
- msprobe/README.md +131 -237
- msprobe/__init__.py +16 -1
- msprobe/{config/config.json → config.json} +47 -49
- msprobe/core/advisor/advisor.py +124 -124
- msprobe/core/advisor/advisor_const.py +58 -59
- msprobe/core/advisor/advisor_result.py +58 -58
- msprobe/core/common/const.py +402 -318
- msprobe/core/common/exceptions.py +99 -99
- msprobe/core/common/{file_check.py → file_utils.py} +523 -283
- msprobe/core/common/inplace_op_checker.py +38 -0
- msprobe/core/common/inplace_ops.yaml +251 -0
- msprobe/core/common/log.py +86 -69
- msprobe/core/common/utils.py +371 -616
- msprobe/core/common_config.py +78 -71
- msprobe/core/compare/acc_compare.py +472 -298
- msprobe/core/compare/check.py +180 -95
- msprobe/core/compare/compare_cli.py +69 -49
- msprobe/core/compare/highlight.py +259 -222
- msprobe/core/compare/multiprocessing_compute.py +174 -149
- msprobe/core/compare/npy_compare.py +310 -295
- msprobe/core/compare/utils.py +464 -429
- msprobe/core/data_dump/data_collector.py +153 -144
- msprobe/core/data_dump/data_processor/base.py +337 -293
- msprobe/core/data_dump/data_processor/factory.py +76 -59
- msprobe/core/data_dump/data_processor/mindspore_processor.py +192 -198
- msprobe/core/data_dump/data_processor/pytorch_processor.py +383 -389
- msprobe/core/data_dump/json_writer.py +117 -116
- msprobe/core/data_dump/scope.py +194 -178
- msprobe/core/grad_probe/constant.py +74 -70
- msprobe/core/grad_probe/grad_compare.py +170 -175
- msprobe/core/grad_probe/utils.py +77 -52
- msprobe/docs/01.installation.md +99 -0
- msprobe/docs/02.config_introduction.md +137 -0
- msprobe/docs/03.config_examples.md +237 -0
- msprobe/docs/04.acl_config_examples.md +78 -0
- msprobe/docs/05.data_dump_PyTorch.md +326 -0
- msprobe/docs/06.data_dump_MindSpore.md +285 -0
- msprobe/docs/07.accuracy_checker_PyTorch.md +297 -0
- msprobe/docs/08.accuracy_checker_online_PyTorch.md +238 -0
- msprobe/docs/09.accuracy_checker_MindSpore.md +68 -0
- msprobe/docs/10.accuracy_compare_PyTorch.md +327 -0
- msprobe/docs/11.accuracy_compare_MindSpore.md +333 -0
- msprobe/docs/12.overflow_check_PyTorch.md +79 -0
- msprobe/docs/13.overflow_check_MindSpore.md +31 -0
- msprobe/{pytorch/doc/parse_tool.md → docs/14.data_parse_PyTorch.md} +283 -286
- msprobe/docs/15.free_benchmarking_PyTorch.md +170 -0
- msprobe/docs/16.free_benchmarking_MindSpore.md +140 -0
- msprobe/{doc/grad_probe/grad_probe.md → docs/17.grad_probe.md} +205 -207
- msprobe/{pytorch/doc//321/205/320/254/320/270/321/207/342/225/221/342/224/220/321/207/342/226/223/342/225/233/321/205/342/225/221/320/266/321/206/320/277/320/244/321/205/320/277/342/225/243.md → docs/18.online_dispatch.md} +89 -90
- msprobe/docs/FAQ.md +189 -0
- msprobe/docs/S02.report_free_benchmarking_validation_performance_baseline.md +146 -0
- msprobe/docs/img/free_benchmark_framework.png +0 -0
- msprobe/docs/img/ms_dump.png +0 -0
- msprobe/docs/img/ms_layer.png +0 -0
- msprobe/docs/img/pt_dump.png +0 -0
- msprobe/mindspore/__init__.py +2 -1
- msprobe/mindspore/api_accuracy_checker/api_accuracy_checker.py +278 -245
- msprobe/mindspore/api_accuracy_checker/api_info.py +76 -69
- msprobe/mindspore/api_accuracy_checker/api_runner.py +155 -151
- msprobe/mindspore/api_accuracy_checker/base_compare_algorithm.py +196 -196
- msprobe/mindspore/api_accuracy_checker/cmd_parser.py +6 -0
- msprobe/mindspore/api_accuracy_checker/compute_element.py +238 -223
- msprobe/mindspore/api_accuracy_checker/main.py +8 -15
- msprobe/mindspore/api_accuracy_checker/type_mapping.py +113 -113
- msprobe/mindspore/api_accuracy_checker/utils.py +79 -62
- msprobe/mindspore/cell_processor.py +58 -34
- msprobe/mindspore/common/const.py +108 -87
- msprobe/mindspore/common/log.py +37 -37
- msprobe/mindspore/common/utils.py +97 -57
- msprobe/mindspore/compare/distributed_compare.py +62 -75
- msprobe/mindspore/compare/layer_mapping.py +146 -0
- msprobe/mindspore/compare/modify_mapping.py +107 -0
- msprobe/mindspore/compare/ms_compare.py +357 -117
- msprobe/mindspore/compare/ms_graph_compare.py +364 -317
- msprobe/mindspore/compare/ms_to_pt_api.yaml +399 -399
- msprobe/mindspore/debugger/debugger_config.py +69 -74
- msprobe/mindspore/debugger/precision_debugger.py +150 -107
- msprobe/mindspore/dump/dump_tool_factory.py +50 -35
- msprobe/mindspore/dump/hook_cell/api_registry.py +128 -104
- msprobe/mindspore/dump/hook_cell/hook_cell.py +55 -53
- msprobe/mindspore/dump/hook_cell/primitive_hooks.py +206 -0
- msprobe/mindspore/dump/hook_cell/support_wrap_ops.yaml +994 -925
- msprobe/mindspore/dump/hook_cell/wrap_api.py +121 -0
- msprobe/mindspore/dump/jit_dump.py +96 -56
- msprobe/mindspore/dump/kernel_graph_dump.py +75 -60
- msprobe/mindspore/dump/kernel_kbyk_dump.py +79 -65
- msprobe/mindspore/free_benchmark/api_pynative_self_check.py +131 -116
- msprobe/mindspore/free_benchmark/common/config.py +27 -12
- msprobe/mindspore/free_benchmark/common/handler_params.py +32 -17
- msprobe/mindspore/free_benchmark/common/utils.py +85 -71
- msprobe/mindspore/free_benchmark/data/support_wrap_ops.yaml +842 -842
- msprobe/mindspore/free_benchmark/decorator/dec_forward.py +57 -42
- msprobe/mindspore/free_benchmark/decorator/decorator_factory.py +122 -107
- msprobe/mindspore/free_benchmark/handler/base_handler.py +105 -90
- msprobe/mindspore/free_benchmark/handler/check_handler.py +56 -41
- msprobe/mindspore/free_benchmark/handler/fix_handler.py +51 -36
- msprobe/mindspore/free_benchmark/handler/handler_factory.py +36 -21
- msprobe/mindspore/free_benchmark/perturbation/add_noise.py +82 -67
- msprobe/mindspore/free_benchmark/perturbation/base_perturbation.py +36 -21
- msprobe/mindspore/free_benchmark/perturbation/bit_noise.py +78 -63
- msprobe/mindspore/free_benchmark/perturbation/exchange_value.py +77 -0
- msprobe/mindspore/free_benchmark/perturbation/improve_precision.py +49 -34
- msprobe/mindspore/free_benchmark/perturbation/no_change.py +27 -12
- msprobe/mindspore/free_benchmark/perturbation/perturbation_factory.py +44 -27
- msprobe/mindspore/free_benchmark/self_check_tool_factory.py +48 -33
- msprobe/mindspore/grad_probe/global_context.py +100 -91
- msprobe/mindspore/grad_probe/grad_analyzer.py +231 -231
- msprobe/mindspore/grad_probe/grad_monitor.py +27 -27
- msprobe/mindspore/grad_probe/grad_stat_csv.py +131 -131
- msprobe/mindspore/grad_probe/hook.py +94 -92
- msprobe/mindspore/grad_probe/utils.py +29 -28
- msprobe/mindspore/ms_config.py +128 -126
- msprobe/mindspore/overflow_check/kernel_graph_overflow_check.py +60 -45
- msprobe/mindspore/overflow_check/overflow_check_tool_factory.py +49 -34
- msprobe/mindspore/runtime.py +4 -4
- msprobe/mindspore/service.py +297 -354
- msprobe/mindspore/task_handler_factory.py +24 -24
- msprobe/msprobe.py +105 -107
- msprobe/pytorch/__init__.py +23 -4
- msprobe/pytorch/api_accuracy_checker/common/config.py +70 -55
- msprobe/pytorch/api_accuracy_checker/common/utils.py +246 -165
- msprobe/pytorch/api_accuracy_checker/compare/algorithm.py +230 -213
- msprobe/pytorch/api_accuracy_checker/compare/api_precision_compare.py +632 -581
- msprobe/pytorch/api_accuracy_checker/compare/api_precision_standard.yaml +132 -132
- msprobe/pytorch/api_accuracy_checker/compare/api_precision_threshold.yaml +390 -390
- msprobe/pytorch/api_accuracy_checker/compare/compare.py +416 -381
- msprobe/pytorch/api_accuracy_checker/compare/compare_column.py +90 -73
- msprobe/pytorch/api_accuracy_checker/compare/compare_utils.py +265 -244
- msprobe/pytorch/api_accuracy_checker/config.yaml +10 -10
- msprobe/pytorch/api_accuracy_checker/run_ut/data_generate.py +370 -332
- msprobe/pytorch/api_accuracy_checker/run_ut/multi_run_ut.py +221 -199
- msprobe/pytorch/api_accuracy_checker/run_ut/run_overflow_check.py +150 -134
- msprobe/pytorch/api_accuracy_checker/run_ut/run_ut.py +518 -581
- msprobe/pytorch/api_accuracy_checker/run_ut/run_ut_utils.py +213 -74
- msprobe/pytorch/api_accuracy_checker/run_ut/torch_ut_setting.json +7 -4
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/attl.py +218 -202
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/client.py +370 -324
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/device_dispatch.py +227 -204
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/dump_dispatch.py +110 -0
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/server.py +244 -218
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/torch_ops_config.yaml +63 -0
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/utils.py +44 -0
- msprobe/pytorch/bench_functions/__init__.py +30 -15
- msprobe/pytorch/bench_functions/apply_adam_w.py +43 -28
- msprobe/pytorch/bench_functions/confusion_transpose.py +34 -19
- msprobe/pytorch/bench_functions/fast_gelu.py +70 -55
- msprobe/pytorch/bench_functions/layer_norm_eval.py +21 -6
- msprobe/pytorch/bench_functions/linear.py +27 -12
- msprobe/pytorch/bench_functions/matmul_backward.py +63 -48
- msprobe/pytorch/bench_functions/npu_fusion_attention.py +538 -421
- msprobe/pytorch/bench_functions/rms_norm.py +30 -15
- msprobe/pytorch/bench_functions/rotary_mul.py +71 -52
- msprobe/pytorch/bench_functions/scaled_mask_softmax.py +41 -26
- msprobe/pytorch/bench_functions/swiglu.py +70 -55
- msprobe/pytorch/common/__init__.py +17 -2
- msprobe/pytorch/common/compare_script.template +14 -14
- msprobe/pytorch/common/log.py +33 -32
- msprobe/pytorch/common/parse_json.py +54 -39
- msprobe/pytorch/common/utils.py +310 -300
- msprobe/pytorch/compare/distributed_compare.py +66 -66
- msprobe/pytorch/compare/mapping.yaml +607 -607
- msprobe/pytorch/compare/match.py +49 -33
- msprobe/pytorch/compare/pt_compare.py +82 -40
- msprobe/pytorch/debugger/debugger_config.py +108 -95
- msprobe/pytorch/debugger/precision_debugger.py +173 -125
- msprobe/pytorch/free_benchmark/__init__.py +23 -8
- msprobe/pytorch/free_benchmark/common/constant.py +70 -70
- msprobe/pytorch/free_benchmark/common/counter.py +71 -71
- msprobe/pytorch/free_benchmark/common/enums.py +65 -37
- msprobe/pytorch/free_benchmark/common/params.py +144 -129
- msprobe/pytorch/free_benchmark/common/utils.py +118 -102
- msprobe/pytorch/free_benchmark/compare/grad_saver.py +200 -179
- msprobe/pytorch/free_benchmark/compare/single_benchmark.py +119 -104
- msprobe/pytorch/free_benchmark/main.py +120 -105
- msprobe/pytorch/free_benchmark/perturbed_layers/base_layer.py +28 -13
- msprobe/pytorch/free_benchmark/perturbed_layers/layer_factory.py +56 -41
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/add_noise.py +105 -90
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/bit_noise.py +119 -104
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/change_value.py +87 -63
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/improve_precision.py +83 -68
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/no_change.py +43 -28
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/npu_base_layser.py +60 -45
- msprobe/pytorch/free_benchmark/perturbed_layers/run_cpu.py +34 -19
- msprobe/pytorch/free_benchmark/result_handlers/base_handler.py +256 -217
- msprobe/pytorch/free_benchmark/result_handlers/check_handler.py +54 -39
- msprobe/pytorch/free_benchmark/result_handlers/fix_handler.py +38 -23
- msprobe/pytorch/free_benchmark/result_handlers/handler_factory.py +45 -30
- msprobe/pytorch/free_benchmark/result_handlers/preheat_handler.py +185 -170
- msprobe/pytorch/function_factory.py +91 -75
- msprobe/pytorch/functional/module_dump.py +84 -0
- msprobe/pytorch/grad_probe/grad_monitor.py +91 -90
- msprobe/pytorch/grad_probe/grad_stat_csv.py +128 -128
- msprobe/pytorch/hook_module/__init__.py +16 -1
- msprobe/pytorch/hook_module/api_registry.py +166 -161
- msprobe/pytorch/hook_module/hook_module.py +118 -120
- msprobe/pytorch/hook_module/support_wrap_ops.yaml +1879 -1877
- msprobe/pytorch/hook_module/utils.py +28 -29
- msprobe/pytorch/hook_module/wrap_aten.py +111 -110
- msprobe/pytorch/hook_module/wrap_distributed.py +77 -78
- msprobe/pytorch/hook_module/wrap_functional.py +104 -105
- msprobe/pytorch/hook_module/wrap_npu_custom.py +85 -84
- msprobe/pytorch/hook_module/wrap_tensor.py +69 -71
- msprobe/pytorch/hook_module/wrap_torch.py +84 -86
- msprobe/pytorch/hook_module/wrap_vf.py +60 -62
- msprobe/pytorch/module_processer.py +153 -138
- msprobe/pytorch/online_dispatch/__init__.py +20 -20
- msprobe/pytorch/online_dispatch/compare.py +235 -236
- msprobe/pytorch/online_dispatch/dispatch.py +271 -271
- msprobe/pytorch/online_dispatch/dump_compare.py +155 -156
- msprobe/pytorch/online_dispatch/single_compare.py +391 -391
- msprobe/pytorch/online_dispatch/torch_ops_config.yaml +57 -49
- msprobe/pytorch/online_dispatch/utils.py +127 -146
- msprobe/pytorch/parse.py +19 -4
- msprobe/pytorch/parse_tool/cli.py +31 -32
- msprobe/pytorch/parse_tool/lib/compare.py +259 -271
- msprobe/pytorch/parse_tool/lib/config.py +52 -52
- msprobe/pytorch/parse_tool/lib/file_desc.py +31 -31
- msprobe/pytorch/parse_tool/lib/interactive_cli.py +102 -102
- msprobe/pytorch/parse_tool/lib/parse_exception.py +54 -54
- msprobe/pytorch/parse_tool/lib/parse_tool.py +161 -158
- msprobe/pytorch/parse_tool/lib/utils.py +320 -321
- msprobe/pytorch/parse_tool/lib/visualization.py +85 -91
- msprobe/pytorch/pt_config.py +317 -187
- msprobe/pytorch/service.py +311 -252
- mindstudio_probe-1.0.3.dist-info/RECORD +0 -272
- msprobe/config/README.md +0 -539
- msprobe/mindspore/doc/compare.md +0 -58
- msprobe/mindspore/doc/dump.md +0 -217
- msprobe/mindspore/dump/hook_cell/wrap_functional.py +0 -91
- msprobe/mindspore/dump/hook_cell/wrap_tensor.py +0 -63
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/ssl_config.py +0 -10
- msprobe/pytorch/doc/FAQ.md +0 -193
- msprobe/pytorch/doc/api_accuracy_checker.md +0 -313
- msprobe/pytorch/doc/api_accuracy_checker_online.md +0 -187
- msprobe/pytorch/doc/dump.md +0 -260
- msprobe/pytorch/doc/msprobe/321/207/342/226/223/342/225/233/321/205/342/225/221/320/266/321/205/342/225/226/320/265/321/205/320/225/342/225/226/321/206/320/245/342/226/221/321/206/320/235/320/276dump/321/206/320/260/320/227/321/205/320/227/320/226/321/206/320/220/320/267/321/210/320/223/342/225/234/321/205/320/257/342/225/221/321/207/342/225/221/342/224/220/321/206/320/232/320/265/321/205/320/241/320/232.md +0 -182
- msprobe/pytorch/doc/ptdbg_ascend_compare.md +0 -240
- msprobe/pytorch/doc/ptdbg_ascend_overview.md +0 -68
- msprobe/pytorch/doc/ptdbg_ascend_quickstart.md +0 -381
- msprobe/pytorch/doc/run_overflow_check.md +0 -25
- msprobe/pytorch/doc//321/206/320/247/320/260/321/206/320/260/320/227/321/206/320/255/320/226/321/205/342/225/226/320/265/321/205/320/225/342/225/226/321/205/320/254/342/225/221/321/206/320/251/320/277/321/211/320/272/320/234/321/210/320/277/320/221/321/205/320/242/320/234/321/206/320/220/320/267/321/210/320/223/342/225/234/321/205/320/257/342/225/221/321/207/342/225/221/342/224/220/321/206/320/232/320/265/321/205/320/241/320/232.md +0 -151
- msprobe/pytorch/functional/data_processor.py +0 -0
- msprobe/pytorch/functional/dump_module.py +0 -39
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/top_level.txt +0 -0
- /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_1.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_2.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_3.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_4.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_1.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_2.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_3.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_4.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_5.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_6.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_7.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/GPT-3_8.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/YOLOV5S_1.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/YOLOV5S_2.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/accuracy_checking_details.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/accuracy_checking_result.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/api_precision_compare_details.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/api_precision_compare_result.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/auto_analyze_log.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/compare_result_pkl.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/compare_result_pkl_md5.png.png +0 -0
- /msprobe/{pytorch/doc → docs}/img/cpu_info.png +0 -0
- /msprobe/{config → docs}/img/free_benchmark.png +0 -0
- /msprobe/{doc/grad_probe/img/image-1.png → docs/img/grad_probe_image-1.png} +0 -0
- /msprobe/{doc/grad_probe/img/image-2.png → docs/img/grad_probe_image-2.png} +0 -0
- /msprobe/{doc/grad_probe/img/image-3.png → docs/img/grad_probe_image-3.png} +0 -0
- /msprobe/{doc/grad_probe/img/image-4.png → docs/img/grad_probe_image-4.png} +0 -0
- /msprobe/{doc/grad_probe/img/image.png → docs/img/grad_probe_image.png} +0 -0
- /msprobe/{pytorch/doc → docs}/img/module_compare.png +0 -0
msprobe/README.md
CHANGED
|
@@ -1,237 +1,131 @@
|
|
|
1
|
-
#
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
|
|
71
|
-
|
|
72
|
-
|
|
73
|
-
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
82
|
-
|
|
83
|
-
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
99
|
-
|
|
100
|
-
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
|
|
104
|
-
|
|
105
|
-
|
|
106
|
-
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
117
|
-
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
127
|
-
|
|
128
|
-
|
|
129
|
-
|
|
130
|
-
|
|
131
|
-
|
|
132
|
-
关键字段含义:
|
|
133
|
-
|
|
134
|
-
- Name:工具名称。
|
|
135
|
-
- Version:工具版本号。
|
|
136
|
-
- Summary:工具概述。
|
|
137
|
-
- Location:工具安装路径。
|
|
138
|
-
- Requires:工具依赖。
|
|
139
|
-
|
|
140
|
-
## 工具使用
|
|
141
|
-
|
|
142
|
-
安装msprobe工具后,可以按照如下思路选择合适的子工具进行精度调试:
|
|
143
|
-
|
|
144
|
-
1. 判断框架场景。
|
|
145
|
-
|
|
146
|
-
当前支持PyTorch和MindSpore场景。
|
|
147
|
-
|
|
148
|
-
2. 执行数据采集。
|
|
149
|
-
|
|
150
|
-
工具通过在训练脚本中添加PrecisionDebugger接口的方式对API执行精度数据dump操作。
|
|
151
|
-
|
|
152
|
-
PyTorch场景:详见[PyTorch_精度数据采集](./pytorch/doc/dump.md)。
|
|
153
|
-
|
|
154
|
-
MindSpore场景:详见[MindSpore_精度数据采集](./mindspore/doc/dump.md)。
|
|
155
|
-
|
|
156
|
-
3. 执行精度预检。
|
|
157
|
-
|
|
158
|
-
在昇腾NPU上扫描用户训练模型中所有API,进行API复现,给出精度情况的诊断和分析。
|
|
159
|
-
|
|
160
|
-
PyTorch场景:详见[PyTorch_精度预检工具](./pytorch/doc/api_accuracy_checker.md)。
|
|
161
|
-
|
|
162
|
-
MindSpore场景:暂不支持。
|
|
163
|
-
|
|
164
|
-
4. 执行精度比对。
|
|
165
|
-
|
|
166
|
-
进行PyTorch整网API粒度的数据dump、精度比对和溢出检测,从而定位训练场景下的精度问题。
|
|
167
|
-
|
|
168
|
-
PyTorch场景:详见[PyTorch_精度比对工具](./pytorch/doc/ptdbg_ascend_compare.md)。
|
|
169
|
-
|
|
170
|
-
MindSpore场景:详见[MindSpore_精度比对工具](./mindspore/doc/compare.md)。
|
|
171
|
-
|
|
172
|
-
5. 执行溢出解析。
|
|
173
|
-
|
|
174
|
-
溢出解析是在执行精度数据dump时,配置了溢出检测dump,那么对于输入正常但输出存在溢出的API,可以判断是否为正常溢出。
|
|
175
|
-
|
|
176
|
-
PyTorch场景:详见[PyTorch_溢出解析工具](./pytorch/doc/run_overflow_check.md)。
|
|
177
|
-
|
|
178
|
-
MindSpore场景:暂不支持。
|
|
179
|
-
|
|
180
|
-
6. 执行数据解析。
|
|
181
|
-
|
|
182
|
-
用于比对前后两次NPU ACL层级dump数据的一致性。
|
|
183
|
-
|
|
184
|
-
PyTorch场景:详见[PyTorch_数据解析工具](./pytorch/doc/parse_tool.md)。
|
|
185
|
-
|
|
186
|
-
MindSpore场景:暂不支持。
|
|
187
|
-
|
|
188
|
-
6. 执行梯度采集和比对。
|
|
189
|
-
|
|
190
|
-
用于采集梯度数据并进行梯度相似度比对。可以精准定位问题出现的step。
|
|
191
|
-
|
|
192
|
-
详见[梯度状态监测工具](./doc/grad_probe/grad_probe.md)。
|
|
193
|
-
|
|
194
|
-
|
|
195
|
-
|
|
196
|
-
上述流程中的工具均为msprobe工具的子工具,使用相同的命令行,格式如下:
|
|
197
|
-
|
|
198
|
-
精度预检工具
|
|
199
|
-
|
|
200
|
-
```bash
|
|
201
|
-
msprobe -f <framework> run_ut [-h]
|
|
202
|
-
```
|
|
203
|
-
|
|
204
|
-
```bash
|
|
205
|
-
msprobe -f <framework> multi_run_ut [-h]
|
|
206
|
-
```
|
|
207
|
-
|
|
208
|
-
```bash
|
|
209
|
-
msprobe -f <framework> api_precision_compare [-h]
|
|
210
|
-
```
|
|
211
|
-
|
|
212
|
-
精度比对工具
|
|
213
|
-
|
|
214
|
-
```bash
|
|
215
|
-
msprobe -f <framework> compare [-h]
|
|
216
|
-
```
|
|
217
|
-
|
|
218
|
-
溢出解析工具
|
|
219
|
-
|
|
220
|
-
```bash
|
|
221
|
-
msprobe -f <framework> run_overflow_check [-h]
|
|
222
|
-
```
|
|
223
|
-
|
|
224
|
-
数据解析工具
|
|
225
|
-
|
|
226
|
-
```bash
|
|
227
|
-
msprobe -f <framework> parse [-h]
|
|
228
|
-
```
|
|
229
|
-
|
|
230
|
-
| 参数 | 说明 |
|
|
231
|
-
| ---- | ------------------------------------------------------------ |
|
|
232
|
-
| -f | 框架,请按所使用框架配置,当前支持配置为:pytorch、mindspore。 |
|
|
233
|
-
| -h | 帮助信息。 |
|
|
234
|
-
|
|
235
|
-
## 贡献
|
|
236
|
-
|
|
237
|
-
push代码前,请务必保证已经完成了基础功能测试和网络测试。
|
|
1
|
+
# 📖 msprobe 使用手册
|
|
2
|
+
|
|
3
|
+

|
|
4
|
+

|
|
5
|
+

|
|
6
|
+
|
|
7
|
+
**msprobe** 是 MindStudio Training Tools 工具链下精度调试部分的工具包。主要包括精度预检、溢出检测和精度比对等功能,目前适配 [PyTorch](https://pytorch.org/) 和 [MindSpore](https://www.mindspore.cn/) 框架。这些子工具侧重不同的训练场景,可以定位模型训练中的精度问题。
|
|
8
|
+
|
|
9
|
+
为方便使用,本工具提供了统一、简易的程序接口:**PrecisionDebugger**。以 PyTorch 框架为例,通过以下示例模板和 **config.json** 可以轻松使用各种功能。
|
|
10
|
+
|
|
11
|
+
```python
|
|
12
|
+
from msprobe.pytorch import PrecisionDebugger
|
|
13
|
+
|
|
14
|
+
debugger = PrecisionDebugger(config_path='./config.json')
|
|
15
|
+
...
|
|
16
|
+
debugger.start() # 一般在训练循环开头启动工具
|
|
17
|
+
... # 循环体
|
|
18
|
+
debugger.stop() # 一般在训练循环末尾结束工具
|
|
19
|
+
debugger.step() # 在训练循环的最后需要重置工具,非循环场景不需要
|
|
20
|
+
```
|
|
21
|
+
|
|
22
|
+
此外,根据以下规则,可以通过环境变量设置日志级别。
|
|
23
|
+
- MSPROBE_LOG_LEVEL=4,不打印任何日志;
|
|
24
|
+
- MSPROBE_LOG_LEVEL=3,仅打印 ERROR;
|
|
25
|
+
- MSPROBE_LOG_LEVEL=2,仅打印 WARNING、ERROR;
|
|
26
|
+
- MSPROBE_LOG_LEVEL=1,仅打印 INFO、WARNING、ERROR(默认配置);
|
|
27
|
+
- MSPROBE_LOG_LEVEL=0,打印 DEBUG、INFO、WARNING、ERROR。
|
|
28
|
+
|
|
29
|
+
例如在 shell 脚本:
|
|
30
|
+
|
|
31
|
+
```shell
|
|
32
|
+
export MSPROBE_LOG_LEVEL={x}
|
|
33
|
+
```
|
|
34
|
+
**config.json** 的配置要求和各功能具体的使用指导详见后续章节。
|
|
35
|
+
|
|
36
|
+
## ⚙️ [安装](./docs/01.installation.md)
|
|
37
|
+
|
|
38
|
+
## 🛠️ config.json [介绍](./docs/02.config_introduction.md) 和 [示例](./docs/03.config_examples.md)
|
|
39
|
+
|
|
40
|
+
## 🧰 主要功能
|
|
41
|
+
|
|
42
|
+
### 1 数据采集
|
|
43
|
+
|
|
44
|
+
msprobe 通过在训练脚本中添加 PrecisionDebugger 接口的方式对 API 执行精度数据 dump 操作,对应 config.json 中的 task 为 statistics 或 tensor。
|
|
45
|
+
|
|
46
|
+
[PyTorch 场景的数据采集](./docs/05.data_dump_PyTorch.md)
|
|
47
|
+
|
|
48
|
+
[MindSpore 场景的数据采集](./docs/06.data_dump_MindSpore.md)
|
|
49
|
+
|
|
50
|
+
### 2 精度预检
|
|
51
|
+
|
|
52
|
+
精度预检旨在昇腾 NPU 上扫描训练模型中的所有 API 进行 API 复现,给出精度情况的诊断和分析。对应 config.json 中的 task 为 run_ut。
|
|
53
|
+
|
|
54
|
+
PyTorch 场景的[离线预检](./docs/07.accuracy_checker_PyTorch.md)和[在线预检](./docs/08.accuracy_checker_online_PyTorch.md)
|
|
55
|
+
|
|
56
|
+
MindSpore 动态图场景的[离线预检](./docs/09.accuracy_checker_MindSpore.md)
|
|
57
|
+
|
|
58
|
+
### 3 精度比对
|
|
59
|
+
|
|
60
|
+
该功能进行 PyTorch 整网 API 粒度的数据 dump、精度比对,进而定位训练场景下的精度问题。
|
|
61
|
+
|
|
62
|
+
[PyTorch 场景的精度比对](./docs/10.accuracy_compare_PyTorch.md)
|
|
63
|
+
|
|
64
|
+
[MindSpore 场景的精度比对](./docs/11.accuracy_compare_MindSpore.md)
|
|
65
|
+
|
|
66
|
+
### 4 溢出检测与解析
|
|
67
|
+
|
|
68
|
+
溢出检测与解析是在执行精度数据 dump 时,判断是否存在输入正常但输出存在溢出的 API,从而判断是否为正常溢出。对应 config.json 中的 overflow_check。
|
|
69
|
+
|
|
70
|
+
[PyTorch 场景的溢出检测与解析](./docs/12.overflow_check_PyTorch.md)
|
|
71
|
+
|
|
72
|
+
[MindSpore 场景的溢出检测与解析](./docs/13.overflow_check_MindSpore.md)
|
|
73
|
+
|
|
74
|
+
### 5 数据解析
|
|
75
|
+
|
|
76
|
+
该功能用于比对前后两次 NPU ACL 层级 dump 数据的一致性。
|
|
77
|
+
|
|
78
|
+
[PyTorch 场景的数据解析](./docs/14.data_parse_PyTorch.md)
|
|
79
|
+
|
|
80
|
+
### 6 无标杆比对
|
|
81
|
+
|
|
82
|
+
[PyTorch 场景的无标杆比对](./docs/15.free_benchmarking_PyTorch.md)
|
|
83
|
+
|
|
84
|
+
[MindSpore 场景的无标杆比对](./docs/16.free_benchmarking_MindSpore.md)
|
|
85
|
+
|
|
86
|
+
### 7 梯度状态监测
|
|
87
|
+
|
|
88
|
+
本功能用于采集梯度数据并进行梯度相似度比对,可以精准定位出现问题的 step。
|
|
89
|
+
|
|
90
|
+
[兼容 PyTorch 和 MindSpore 框架的梯度监测](./docs/17.grad_probe.md)
|
|
91
|
+
|
|
92
|
+
### 8 在线精度比对
|
|
93
|
+
|
|
94
|
+
在线精度比对是实现在PyTorch训练过程中直接完成精度比对并输出比对结果的功能,是NPU与CPU之间的精度比对。
|
|
95
|
+
|
|
96
|
+
[PyTorch 场景的在线精度比对](./docs/18.online_dispatch.md)
|
|
97
|
+
|
|
98
|
+
## 🌟 新版本特性
|
|
99
|
+
|
|
100
|
+
若查看历史版本特性,请点击[安装](./docs/01.installation.md)。
|
|
101
|
+
|
|
102
|
+
【数据采集】
|
|
103
|
+
- 支持 config.json 中的 step 传入范围;
|
|
104
|
+
- 优化了指定 step 的机制,指定 step 结束后工具不再采集数据,但训练会继续运行。工具结束运行后,日志提示信息如下:
|
|
105
|
+
```bash
|
|
106
|
+
****************************************
|
|
107
|
+
* msprobe ends successfully. *
|
|
108
|
+
****************************************
|
|
109
|
+
```
|
|
110
|
+
注:在多卡场景,每张卡进程训练到指定 step 之后都会打印一次上述信息。
|
|
111
|
+
|
|
112
|
+
【精度预检】
|
|
113
|
+
- 在 PyTorch 场景,支持部分 NPU 融合算子预检。
|
|
114
|
+
|
|
115
|
+
【精度比对】
|
|
116
|
+
- 解决了使用 MindSpore 需要安装 PyTorch 的问题。
|
|
117
|
+
|
|
118
|
+
【无标杆比对】
|
|
119
|
+
- 补充在 PyTorch 场景的性能基线报告;
|
|
120
|
+
- 支持 MindSpore 场景的 change_value 扰动模式。
|
|
121
|
+
|
|
122
|
+
## 📑 补充材料
|
|
123
|
+
|
|
124
|
+
[无标杆比对功能在 PyTorch 场景的性能基线报告](./docs/S02.report_free_benchmarking_validation_performance_baseline.md)
|
|
125
|
+
|
|
126
|
+
## ❗ 免责声明
|
|
127
|
+
本工具建议执行用户与安装用户保持一致,如果您要使用 root 执行,请自行关注 root 高权限触及的安全风险。
|
|
128
|
+
|
|
129
|
+
## ❓ FAQ
|
|
130
|
+
|
|
131
|
+
[FAQ for PyTorch](./docs/FAQ_PyTorch.md)
|
msprobe/__init__.py
CHANGED
|
@@ -1 +1,16 @@
|
|
|
1
|
-
|
|
1
|
+
# Copyright (c) 2024-2024, Huawei Technologies Co., Ltd.
|
|
2
|
+
# All rights reserved.
|
|
3
|
+
#
|
|
4
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
5
|
+
# you may not use this file except in compliance with the License.
|
|
6
|
+
# You may obtain a copy of the License at
|
|
7
|
+
#
|
|
8
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
9
|
+
#
|
|
10
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
11
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
12
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
13
|
+
# See the License for the specific language governing permissions and
|
|
14
|
+
# limitations under the License.
|
|
15
|
+
|
|
16
|
+
from msprobe.core.grad_probe.grad_compare import GradComparator
|
|
@@ -1,50 +1,48 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task": "statistics",
|
|
3
|
-
"dump_path": "./dump_path",
|
|
4
|
-
"rank": [],
|
|
5
|
-
"step": [],
|
|
6
|
-
"level": "L1",
|
|
7
|
-
"
|
|
8
|
-
"
|
|
9
|
-
"
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
"
|
|
13
|
-
"
|
|
14
|
-
"
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
"
|
|
20
|
-
"
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
"
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
"
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
"
|
|
40
|
-
"
|
|
41
|
-
"
|
|
42
|
-
"
|
|
43
|
-
"
|
|
44
|
-
"
|
|
45
|
-
"
|
|
46
|
-
"
|
|
47
|
-
|
|
48
|
-
"max_sample": 20
|
|
49
|
-
}
|
|
1
|
+
{
|
|
2
|
+
"task": "statistics",
|
|
3
|
+
"dump_path": "./dump_path",
|
|
4
|
+
"rank": [],
|
|
5
|
+
"step": [],
|
|
6
|
+
"level": "L1",
|
|
7
|
+
"enable_dataloader": false,
|
|
8
|
+
"acl_config": "",
|
|
9
|
+
"tensor": {
|
|
10
|
+
"scope": [],
|
|
11
|
+
"list":[],
|
|
12
|
+
"data_mode": ["all"],
|
|
13
|
+
"backward_input": [],
|
|
14
|
+
"file_format": "npy"
|
|
15
|
+
},
|
|
16
|
+
"statistics": {
|
|
17
|
+
"scope": [],
|
|
18
|
+
"list":[],
|
|
19
|
+
"data_mode": ["all"],
|
|
20
|
+
"summary_mode": "statistics"
|
|
21
|
+
},
|
|
22
|
+
"overflow_check": {
|
|
23
|
+
"overflow_nums": 1,
|
|
24
|
+
"check_mode":"all"
|
|
25
|
+
},
|
|
26
|
+
"run_ut": {
|
|
27
|
+
"white_list": [],
|
|
28
|
+
"black_list": [],
|
|
29
|
+
"error_data_path": "./"
|
|
30
|
+
},
|
|
31
|
+
"grad_probe": {
|
|
32
|
+
"grad_level": "L1",
|
|
33
|
+
"param_list": [],
|
|
34
|
+
"bounds": [-1, 0, 1]
|
|
35
|
+
},
|
|
36
|
+
"free_benchmark": {
|
|
37
|
+
"scope": [],
|
|
38
|
+
"list": [],
|
|
39
|
+
"fuzz_device": "npu",
|
|
40
|
+
"pert_mode": "improve_precision",
|
|
41
|
+
"handler_type": "check",
|
|
42
|
+
"fuzz_level": "L1",
|
|
43
|
+
"fuzz_stage": "forward",
|
|
44
|
+
"if_preheat": false,
|
|
45
|
+
"preheat_step": 15,
|
|
46
|
+
"max_sample": 20
|
|
47
|
+
}
|
|
50
48
|
}
|