mindstudio-probe 1.0.3__py3-none-any.whl → 1.0.4__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-1.0.4.dist-info}/LICENSE +201 -201
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.0.4.dist-info}/METADATA +36 -34
- mindstudio_probe-1.0.4.dist-info/RECORD +276 -0
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.0.4.dist-info}/WHEEL +1 -1
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.0.4.dist-info}/entry_points.txt +1 -0
- msprobe/README.md +101 -237
- msprobe/{config/config.json → config.json} +49 -49
- msprobe/core/advisor/advisor.py +124 -124
- msprobe/core/advisor/advisor_const.py +59 -59
- msprobe/core/advisor/advisor_result.py +58 -58
- msprobe/core/common/const.py +341 -318
- msprobe/core/common/exceptions.py +99 -99
- msprobe/core/common/{file_check.py → file_utils.py} +478 -283
- msprobe/core/common/log.py +76 -69
- msprobe/core/common/utils.py +385 -616
- msprobe/core/common_config.py +85 -71
- msprobe/core/compare/acc_compare.py +299 -298
- msprobe/core/compare/check.py +95 -95
- msprobe/core/compare/compare_cli.py +49 -49
- msprobe/core/compare/highlight.py +223 -222
- msprobe/core/compare/multiprocessing_compute.py +149 -149
- msprobe/core/compare/npy_compare.py +295 -295
- msprobe/core/compare/utils.py +430 -429
- msprobe/core/data_dump/data_collector.py +154 -144
- msprobe/core/data_dump/data_processor/base.py +314 -293
- msprobe/core/data_dump/data_processor/factory.py +59 -59
- msprobe/core/data_dump/data_processor/mindspore_processor.py +186 -198
- msprobe/core/data_dump/data_processor/pytorch_processor.py +366 -389
- msprobe/core/data_dump/json_writer.py +96 -116
- msprobe/core/data_dump/scope.py +178 -178
- msprobe/core/grad_probe/constant.py +70 -70
- msprobe/core/grad_probe/grad_compare.py +171 -175
- msprobe/core/grad_probe/utils.py +64 -52
- msprobe/docs/01.installation.md +89 -0
- msprobe/docs/02.config_introduction.md +165 -0
- msprobe/docs/03.config_examples.md +247 -0
- msprobe/docs/04.acl_config_examples.md +76 -0
- msprobe/docs/05.data_dump_PyTorch.md +198 -0
- msprobe/docs/06.data_dump_MindSpore.md +243 -0
- msprobe/docs/07.accuracy_checker_PyTorch.md +274 -0
- msprobe/docs/08.accuracy_checker_online_PyTorch.md +198 -0
- msprobe/docs/09.accuracy_checker_MindSpore.md +68 -0
- msprobe/docs/10.accuracy_compare_PyTorch.md +245 -0
- msprobe/docs/11.accuracy_compare_MindSpore.md +202 -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 +164 -0
- msprobe/{doc/grad_probe/grad_probe.md → docs/17.grad_probe.md} +207 -207
- msprobe/docs/FAQ_PyTorch.md +177 -0
- msprobe/docs/S02.report_free_benchmarking_validation_performance_baseline.md +146 -0
- msprobe/docs/img/free_benchmark_framework.png +0 -0
- msprobe/mindspore/__init__.py +1 -1
- msprobe/mindspore/api_accuracy_checker/api_accuracy_checker.py +254 -245
- msprobe/mindspore/api_accuracy_checker/api_info.py +69 -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 +34 -34
- msprobe/mindspore/common/const.py +106 -87
- msprobe/mindspore/common/log.py +37 -37
- msprobe/mindspore/common/utils.py +81 -57
- msprobe/mindspore/compare/distributed_compare.py +75 -75
- msprobe/mindspore/compare/ms_compare.py +219 -117
- msprobe/mindspore/compare/ms_graph_compare.py +348 -317
- msprobe/mindspore/compare/ms_to_pt_api.yaml +399 -399
- msprobe/mindspore/debugger/debugger_config.py +66 -74
- msprobe/mindspore/debugger/precision_debugger.py +126 -107
- msprobe/mindspore/dump/dump_tool_factory.py +35 -35
- msprobe/mindspore/dump/hook_cell/api_registry.py +118 -104
- msprobe/mindspore/dump/hook_cell/hook_cell.py +55 -53
- msprobe/mindspore/dump/hook_cell/support_wrap_ops.yaml +922 -925
- msprobe/mindspore/dump/hook_cell/wrap_api.py +113 -0
- msprobe/mindspore/dump/jit_dump.py +72 -56
- msprobe/mindspore/dump/kernel_graph_dump.py +59 -60
- msprobe/mindspore/dump/kernel_kbyk_dump.py +64 -65
- msprobe/mindspore/free_benchmark/api_pynative_self_check.py +116 -116
- msprobe/mindspore/free_benchmark/common/config.py +12 -12
- msprobe/mindspore/free_benchmark/common/handler_params.py +17 -17
- msprobe/mindspore/free_benchmark/common/utils.py +71 -71
- msprobe/mindspore/free_benchmark/data/support_wrap_ops.yaml +842 -842
- msprobe/mindspore/free_benchmark/decorator/dec_forward.py +43 -42
- msprobe/mindspore/free_benchmark/decorator/decorator_factory.py +107 -107
- msprobe/mindspore/free_benchmark/handler/base_handler.py +90 -90
- msprobe/mindspore/free_benchmark/handler/check_handler.py +41 -41
- msprobe/mindspore/free_benchmark/handler/fix_handler.py +36 -36
- msprobe/mindspore/free_benchmark/handler/handler_factory.py +21 -21
- msprobe/mindspore/free_benchmark/perturbation/add_noise.py +67 -67
- msprobe/mindspore/free_benchmark/perturbation/base_perturbation.py +21 -21
- msprobe/mindspore/free_benchmark/perturbation/bit_noise.py +63 -63
- msprobe/mindspore/free_benchmark/perturbation/exchange_value.py +51 -0
- msprobe/mindspore/free_benchmark/perturbation/improve_precision.py +35 -34
- msprobe/mindspore/free_benchmark/perturbation/no_change.py +12 -12
- msprobe/mindspore/free_benchmark/perturbation/perturbation_factory.py +29 -27
- msprobe/mindspore/free_benchmark/self_check_tool_factory.py +33 -33
- msprobe/mindspore/grad_probe/global_context.py +90 -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 +44 -45
- msprobe/mindspore/overflow_check/overflow_check_tool_factory.py +34 -34
- msprobe/mindspore/runtime.py +4 -4
- msprobe/mindspore/service.py +378 -354
- msprobe/mindspore/task_handler_factory.py +24 -24
- msprobe/msprobe.py +105 -107
- msprobe/pytorch/__init__.py +3 -3
- msprobe/pytorch/api_accuracy_checker/common/config.py +53 -55
- msprobe/pytorch/api_accuracy_checker/common/utils.py +214 -165
- msprobe/pytorch/api_accuracy_checker/compare/algorithm.py +213 -213
- msprobe/pytorch/api_accuracy_checker/compare/api_precision_compare.py +606 -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 +386 -381
- msprobe/pytorch/api_accuracy_checker/compare/compare_column.py +73 -73
- msprobe/pytorch/api_accuracy_checker/compare/compare_utils.py +245 -244
- msprobe/pytorch/api_accuracy_checker/config.yaml +10 -10
- msprobe/pytorch/api_accuracy_checker/run_ut/data_generate.py +335 -332
- msprobe/pytorch/api_accuracy_checker/run_ut/multi_run_ut.py +200 -199
- msprobe/pytorch/api_accuracy_checker/run_ut/run_overflow_check.py +133 -134
- msprobe/pytorch/api_accuracy_checker/run_ut/run_ut.py +592 -581
- msprobe/pytorch/api_accuracy_checker/run_ut/run_ut_utils.py +70 -74
- msprobe/pytorch/api_accuracy_checker/run_ut/torch_ut_setting.json +7 -4
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/attl.py +197 -202
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/client.py +325 -324
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/device_dispatch.py +204 -204
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/server.py +219 -218
- msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/ssl_config.py +10 -10
- msprobe/pytorch/bench_functions/__init__.py +15 -15
- msprobe/pytorch/bench_functions/apply_adam_w.py +28 -28
- msprobe/pytorch/bench_functions/confusion_transpose.py +19 -19
- msprobe/pytorch/bench_functions/fast_gelu.py +55 -55
- msprobe/pytorch/bench_functions/layer_norm_eval.py +6 -6
- msprobe/pytorch/bench_functions/linear.py +12 -12
- msprobe/pytorch/bench_functions/matmul_backward.py +48 -48
- msprobe/pytorch/bench_functions/npu_fusion_attention.py +509 -421
- msprobe/pytorch/bench_functions/rms_norm.py +15 -15
- msprobe/pytorch/bench_functions/rotary_mul.py +52 -52
- msprobe/pytorch/bench_functions/scaled_mask_softmax.py +26 -26
- msprobe/pytorch/bench_functions/swiglu.py +55 -55
- msprobe/pytorch/common/__init__.py +2 -2
- msprobe/pytorch/common/compare_script.template +14 -14
- msprobe/pytorch/common/log.py +20 -31
- msprobe/pytorch/common/parse_json.py +39 -39
- msprobe/pytorch/common/utils.py +305 -300
- msprobe/pytorch/compare/distributed_compare.py +66 -66
- msprobe/pytorch/compare/mapping.yaml +607 -607
- msprobe/pytorch/compare/match.py +34 -33
- msprobe/pytorch/compare/pt_compare.py +50 -40
- msprobe/pytorch/debugger/debugger_config.py +95 -95
- msprobe/pytorch/debugger/precision_debugger.py +125 -125
- msprobe/pytorch/free_benchmark/__init__.py +8 -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 +37 -37
- msprobe/pytorch/free_benchmark/common/params.py +129 -129
- msprobe/pytorch/free_benchmark/common/utils.py +102 -102
- msprobe/pytorch/free_benchmark/compare/grad_saver.py +179 -179
- msprobe/pytorch/free_benchmark/compare/single_benchmark.py +104 -104
- msprobe/pytorch/free_benchmark/main.py +105 -105
- msprobe/pytorch/free_benchmark/perturbed_layers/base_layer.py +13 -13
- msprobe/pytorch/free_benchmark/perturbed_layers/layer_factory.py +41 -41
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/add_noise.py +90 -90
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/bit_noise.py +104 -104
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/change_value.py +63 -63
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/improve_precision.py +68 -68
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/no_change.py +28 -28
- msprobe/pytorch/free_benchmark/perturbed_layers/npu/npu_base_layser.py +45 -45
- msprobe/pytorch/free_benchmark/perturbed_layers/run_cpu.py +19 -19
- msprobe/pytorch/free_benchmark/result_handlers/base_handler.py +217 -217
- msprobe/pytorch/free_benchmark/result_handlers/check_handler.py +39 -39
- msprobe/pytorch/free_benchmark/result_handlers/fix_handler.py +23 -23
- msprobe/pytorch/free_benchmark/result_handlers/handler_factory.py +30 -30
- msprobe/pytorch/free_benchmark/result_handlers/preheat_handler.py +170 -170
- msprobe/pytorch/function_factory.py +76 -75
- msprobe/pytorch/functional/dump_module.py +39 -39
- msprobe/pytorch/grad_probe/grad_monitor.py +91 -90
- msprobe/pytorch/grad_probe/grad_stat_csv.py +128 -128
- msprobe/pytorch/hook_module/api_registry.py +161 -161
- msprobe/pytorch/hook_module/hook_module.py +120 -120
- msprobe/pytorch/hook_module/support_wrap_ops.yaml +1879 -1877
- msprobe/pytorch/hook_module/utils.py +30 -29
- msprobe/pytorch/hook_module/wrap_aten.py +110 -110
- msprobe/pytorch/hook_module/wrap_distributed.py +78 -78
- msprobe/pytorch/hook_module/wrap_functional.py +105 -105
- msprobe/pytorch/hook_module/wrap_npu_custom.py +93 -84
- msprobe/pytorch/hook_module/wrap_tensor.py +71 -71
- msprobe/pytorch/hook_module/wrap_torch.py +86 -86
- msprobe/pytorch/hook_module/wrap_vf.py +62 -62
- msprobe/pytorch/module_processer.py +138 -138
- msprobe/pytorch/online_dispatch/__init__.py +20 -20
- msprobe/pytorch/online_dispatch/compare.py +236 -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 +49 -49
- msprobe/pytorch/online_dispatch/utils.py +130 -146
- msprobe/pytorch/parse.py +4 -4
- msprobe/pytorch/parse_tool/cli.py +32 -32
- msprobe/pytorch/parse_tool/lib/compare.py +260 -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 +158 -158
- msprobe/pytorch/parse_tool/lib/utils.py +316 -321
- msprobe/pytorch/parse_tool/lib/visualization.py +85 -91
- msprobe/pytorch/pt_config.py +188 -187
- msprobe/pytorch/service.py +246 -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/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//321/205/320/254/320/270/321/207/342/225/221/342/224/220/321/207/342/226/223/342/225/233/321/205/342/225/221/320/266/321/206/320/277/320/244/321/205/320/277/342/225/243.md +0 -90
- 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
- {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.0.4.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
|
@@ -1,218 +1,219 @@
|
|
|
1
|
-
import os.path
|
|
2
|
-
import struct
|
|
3
|
-
import hashlib
|
|
4
|
-
import time
|
|
5
|
-
import io
|
|
6
|
-
from threading import Thread
|
|
7
|
-
|
|
8
|
-
from
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
from msprobe.pytorch.
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
self.
|
|
18
|
-
self.
|
|
19
|
-
self.
|
|
20
|
-
self.
|
|
21
|
-
self.
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
server_context_.
|
|
37
|
-
server_context_.
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
self.reactor_thread.
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
self.
|
|
68
|
-
self.
|
|
69
|
-
self.
|
|
70
|
-
self.
|
|
71
|
-
self.
|
|
72
|
-
self.
|
|
73
|
-
self.
|
|
74
|
-
self.
|
|
75
|
-
self.
|
|
76
|
-
self.
|
|
77
|
-
self.
|
|
78
|
-
self.
|
|
79
|
-
self.
|
|
80
|
-
|
|
81
|
-
|
|
82
|
-
|
|
83
|
-
self.
|
|
84
|
-
self.
|
|
85
|
-
self.
|
|
86
|
-
self.
|
|
87
|
-
self.
|
|
88
|
-
self.factory.
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
99
|
-
|
|
100
|
-
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
self.
|
|
104
|
-
self.
|
|
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
|
-
|
|
133
|
-
self.
|
|
134
|
-
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
|
|
141
|
-
|
|
142
|
-
|
|
143
|
-
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
|
|
148
|
-
|
|
149
|
-
|
|
150
|
-
|
|
151
|
-
self.
|
|
152
|
-
|
|
153
|
-
|
|
154
|
-
|
|
155
|
-
|
|
156
|
-
self.
|
|
157
|
-
self.
|
|
158
|
-
self.
|
|
159
|
-
self.
|
|
160
|
-
self.
|
|
161
|
-
|
|
162
|
-
|
|
163
|
-
|
|
164
|
-
self.buffer.
|
|
165
|
-
self.buffer.
|
|
166
|
-
|
|
167
|
-
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
self.
|
|
171
|
-
self.
|
|
172
|
-
self.
|
|
173
|
-
self.
|
|
174
|
-
self.
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
|
|
178
|
-
|
|
179
|
-
|
|
180
|
-
|
|
181
|
-
and self.
|
|
182
|
-
and
|
|
183
|
-
|
|
184
|
-
|
|
185
|
-
self.
|
|
186
|
-
|
|
187
|
-
|
|
188
|
-
|
|
189
|
-
|
|
190
|
-
|
|
191
|
-
|
|
192
|
-
|
|
193
|
-
|
|
194
|
-
self.
|
|
195
|
-
self.buffer.
|
|
196
|
-
self.
|
|
197
|
-
|
|
198
|
-
|
|
199
|
-
|
|
200
|
-
|
|
201
|
-
|
|
202
|
-
|
|
203
|
-
|
|
204
|
-
|
|
205
|
-
|
|
206
|
-
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
213
|
-
|
|
214
|
-
|
|
215
|
-
self.
|
|
216
|
-
|
|
217
|
-
|
|
218
|
-
|
|
1
|
+
import os.path
|
|
2
|
+
import struct
|
|
3
|
+
import hashlib
|
|
4
|
+
import time
|
|
5
|
+
import io
|
|
6
|
+
from threading import Thread
|
|
7
|
+
|
|
8
|
+
from twisted.internet import reactor, protocol, endpoints
|
|
9
|
+
|
|
10
|
+
from msprobe.pytorch.common.utils import logger
|
|
11
|
+
from msprobe.pytorch.api_accuracy_checker.tensor_transport_layer.ssl_config import cipher_list
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
class TCPServer:
|
|
15
|
+
def __init__(self, port, shared_queue, check_sum=False, tls_path=None) -> None:
|
|
16
|
+
self.port = port
|
|
17
|
+
self.shared_queue = shared_queue
|
|
18
|
+
self.check_sum = check_sum
|
|
19
|
+
self.tls_path = tls_path
|
|
20
|
+
self.factory = MessageServerFactory()
|
|
21
|
+
self.reactor_thread = None
|
|
22
|
+
|
|
23
|
+
@staticmethod
|
|
24
|
+
def run_reactor():
|
|
25
|
+
reactor.run(installSignalHandlers=False)
|
|
26
|
+
|
|
27
|
+
def start(self):
|
|
28
|
+
self.factory.protocol = self.build_protocol
|
|
29
|
+
|
|
30
|
+
if self.tls_path:
|
|
31
|
+
from OpenSSL import SSL
|
|
32
|
+
from twisted.internet import ssl
|
|
33
|
+
server_key = os.path.join(self.tls_path, "server.key")
|
|
34
|
+
server_crt = os.path.join(self.tls_path, "server.crt")
|
|
35
|
+
server_context_factory = ssl.DefaultOpenSSLContextFactory(server_key, server_crt, SSL.TLSv1_2_METHOD)
|
|
36
|
+
server_context_ = server_context_factory.getContext()
|
|
37
|
+
server_context_.set_cipher_list(cipher_list)
|
|
38
|
+
server_context_.set_options(SSL.OP_NO_RENEGOTIATION)
|
|
39
|
+
endpoint = endpoints.SSL4ServerEndpoint(reactor, self.port, server_context_factory)
|
|
40
|
+
else:
|
|
41
|
+
endpoint = endpoints.TCP4ServerEndpoint(reactor, self.port)
|
|
42
|
+
endpoint.listen(self.factory)
|
|
43
|
+
self.reactor_thread = Thread(target=self.run_reactor, daemon=True)
|
|
44
|
+
self.reactor_thread.start()
|
|
45
|
+
|
|
46
|
+
def is_running(self):
|
|
47
|
+
return not self.factory.is_all_connection_closed()
|
|
48
|
+
|
|
49
|
+
def stop(self):
|
|
50
|
+
self.factory.doStop()
|
|
51
|
+
reactor.callFromThread(reactor.sigInt, 2)
|
|
52
|
+
self.reactor_thread.join()
|
|
53
|
+
|
|
54
|
+
def build_protocol(self):
|
|
55
|
+
return ServerProtocol(self.shared_queue, self.check_sum)
|
|
56
|
+
|
|
57
|
+
|
|
58
|
+
class ServerProtocol(protocol.Protocol):
|
|
59
|
+
ACK_SUCCESS = b"OK___"
|
|
60
|
+
ACK_ERROR = b"ERROR"
|
|
61
|
+
ACK_BUSY = b"BUSY_"
|
|
62
|
+
ACK_STOP = b"STOP_"
|
|
63
|
+
ACK_STOP_CONFIRM = b"OVER_"
|
|
64
|
+
ACK_KILL_PROCESS = b"KILL_"
|
|
65
|
+
|
|
66
|
+
def __init__(self, shared_queue, check_sum=False):
|
|
67
|
+
self.start_time = None
|
|
68
|
+
self.buffer = io.BytesIO()
|
|
69
|
+
self.consumer_queue = shared_queue
|
|
70
|
+
self.check_sum = check_sum
|
|
71
|
+
self.length_width = 8
|
|
72
|
+
self.md5_width = 32
|
|
73
|
+
self.obj_length = None
|
|
74
|
+
self.tell = 0
|
|
75
|
+
self.obj_md5 = None
|
|
76
|
+
self.obj_body = None
|
|
77
|
+
self.sequence_number = -1
|
|
78
|
+
self.rank = -1
|
|
79
|
+
self.step = -1
|
|
80
|
+
self.sequence_number_dict = dict()
|
|
81
|
+
|
|
82
|
+
def connectionMade(self):
|
|
83
|
+
self.buffer = io.BytesIO()
|
|
84
|
+
self.obj_length = None
|
|
85
|
+
self.tell = 0
|
|
86
|
+
self.obj_md5 = None
|
|
87
|
+
self.obj_body = None
|
|
88
|
+
self.factory.transport_dict[self.transport] = 1
|
|
89
|
+
self.factory.transport_list.append(self.transport)
|
|
90
|
+
logger.info(f"Connected to {self.transport.getPeer()} successfully.")
|
|
91
|
+
|
|
92
|
+
def connectionLost(self, reason):
|
|
93
|
+
self.factory.transport_dict.pop(self.transport, None)
|
|
94
|
+
if len(self.factory.transport_dict) == 0:
|
|
95
|
+
self.consumer_queue.put(self.ACK_KILL_PROCESS)
|
|
96
|
+
|
|
97
|
+
logger.info(f"Lost connection with {self.transport.getPeer()}. Reason is: {reason} 与客户端 断开连接, "
|
|
98
|
+
f"current connection number is: {len(self.factory.transport_dict)}")
|
|
99
|
+
|
|
100
|
+
def send_ack(self, ack_info):
|
|
101
|
+
ack_message = b"".join([
|
|
102
|
+
ack_info,
|
|
103
|
+
self.sequence_number.to_bytes(8, byteorder='big'),
|
|
104
|
+
self.rank.to_bytes(8, byteorder='big'),
|
|
105
|
+
self.step.to_bytes(8, byteorder='big')
|
|
106
|
+
])
|
|
107
|
+
self.transport.write(ack_message)
|
|
108
|
+
|
|
109
|
+
def post_process(self):
|
|
110
|
+
send_busy_ack = False
|
|
111
|
+
while self.consumer_queue.full():
|
|
112
|
+
if not send_busy_ack:
|
|
113
|
+
self.send_ack(self.ACK_BUSY)
|
|
114
|
+
logger.debug("sending BUSY ACK")
|
|
115
|
+
send_busy_ack = True
|
|
116
|
+
time.sleep(0.1)
|
|
117
|
+
|
|
118
|
+
obj_key = str(self.sequence_number) + "_" + str(self.rank) + "_" + str(self.step)
|
|
119
|
+
|
|
120
|
+
recv_md5 = hashlib.md5(self.obj_body).hexdigest()
|
|
121
|
+
if self.check_sum and recv_md5 != self.obj_md5:
|
|
122
|
+
# when needs check md5 and check no pass, indicates received data error, send b"ERROR" to client.
|
|
123
|
+
logger.debug(f"Error:接收数据有问题,流水号{self.sequence_number}, expected {self.obj_md5}, but get {recv_md5}")
|
|
124
|
+
self.send_ack(self.ACK_ERROR)
|
|
125
|
+
else:
|
|
126
|
+
if self.obj_body == self.ACK_STOP:
|
|
127
|
+
self.handle_with_stop()
|
|
128
|
+
else:
|
|
129
|
+
self.send_ack(self.ACK_SUCCESS)
|
|
130
|
+
if obj_key in self.sequence_number_dict:
|
|
131
|
+
logger.debug(f"这是一次异常的重传,可以忽略。 {obj_key}, {self.sequence_number_dict}")
|
|
132
|
+
else:
|
|
133
|
+
self.sequence_number_dict[obj_key] = self.obj_md5
|
|
134
|
+
self.consumer_queue.put(self.obj_body, block=True)
|
|
135
|
+
|
|
136
|
+
self.reset_env()
|
|
137
|
+
finish_time = time.time()
|
|
138
|
+
logger.debug(f"finish_time: {finish_time - self.start_time}")
|
|
139
|
+
|
|
140
|
+
def handle_with_stop(self):
|
|
141
|
+
logger.debug(f"接收到停止传输信号 TCP{self.transport.getPeer()}")
|
|
142
|
+
self.send_ack(self.ACK_STOP_CONFIRM)
|
|
143
|
+
if len(self.factory.transport_dict) == 0:
|
|
144
|
+
_rank, _step, _sequence_number = 0, 0, 100000000
|
|
145
|
+
ack_kill = self.ACK_KILL_PROCESS + \
|
|
146
|
+
_sequence_number.to_bytes(8, byteorder='big') + \
|
|
147
|
+
_rank.to_bytes(8, byteorder='big') + \
|
|
148
|
+
_step.to_bytes(8, byteorder='big')
|
|
149
|
+
for trans in self.factory.transport_list:
|
|
150
|
+
trans.write(ack_kill)
|
|
151
|
+
logger.debug(f"发送KILL信息给{self.transport.getPeer()}")
|
|
152
|
+
self.consumer_queue.put(self.ACK_KILL_PROCESS)
|
|
153
|
+
time.sleep(2)
|
|
154
|
+
|
|
155
|
+
def reset_env(self):
|
|
156
|
+
self.obj_length = None
|
|
157
|
+
self.sequence_number = -1
|
|
158
|
+
self.rank = -1
|
|
159
|
+
self.step = -1
|
|
160
|
+
self.obj_md5 = None
|
|
161
|
+
self.obj_body = None
|
|
162
|
+
|
|
163
|
+
def dataReceived(self, data):
|
|
164
|
+
self.buffer.seek(0, 2)
|
|
165
|
+
self.buffer.write(data)
|
|
166
|
+
self.buffer.seek(self.tell)
|
|
167
|
+
|
|
168
|
+
# The first data packet is packet header, it contains obj_length, sequence_number, rank, step
|
|
169
|
+
if self.obj_length is None and len(self.buffer.getvalue()) >= self.length_width * 4:
|
|
170
|
+
self.start_time = time.time()
|
|
171
|
+
self.obj_length = struct.unpack('!Q', self.buffer.read(self.length_width))[0]
|
|
172
|
+
self.sequence_number = struct.unpack('!Q', self.buffer.read(self.length_width))[0]
|
|
173
|
+
self.rank = struct.unpack('!Q', self.buffer.read(self.length_width))[0]
|
|
174
|
+
self.step = struct.unpack('!Q', self.buffer.read(self.length_width))[0]
|
|
175
|
+
self.tell += self.length_width * 4
|
|
176
|
+
logger.debug(
|
|
177
|
+
f"流水号: {self.sequence_number}; RANK: {self.rank}; STEP: {self.step}; Length: {self.obj_length}")
|
|
178
|
+
|
|
179
|
+
# If needs check md5 but not parse md5 yet, read 32b md5 values
|
|
180
|
+
check_sum_and_md5 = (self.check_sum
|
|
181
|
+
and self.obj_length is not None
|
|
182
|
+
and self.obj_md5 is None
|
|
183
|
+
and len(self.buffer.getvalue()) - self.tell >= self.md5_width)
|
|
184
|
+
if check_sum_and_md5:
|
|
185
|
+
self.obj_md5 = self.buffer.read(self.md5_width).decode()
|
|
186
|
+
self.tell += self.md5_width
|
|
187
|
+
logger.debug(f"MD5: {self.obj_md5}")
|
|
188
|
+
|
|
189
|
+
current_length = len(self.buffer.getvalue()) - self.tell
|
|
190
|
+
if self.obj_length is not None and 0 < self.obj_length <= current_length:
|
|
191
|
+
# Current api data receive finished
|
|
192
|
+
self.obj_body = self.buffer.read(self.obj_length)
|
|
193
|
+
|
|
194
|
+
self.tell += self.obj_length
|
|
195
|
+
self.buffer = io.BytesIO(self.buffer.getvalue()[self.tell:])
|
|
196
|
+
self.buffer.seek(0)
|
|
197
|
+
self.tell = 0
|
|
198
|
+
recv_data_time = time.time()
|
|
199
|
+
logger.debug(f"self.sequence_number {self.sequence_number} "
|
|
200
|
+
f"recv_data_time {recv_data_time - self.start_time}")
|
|
201
|
+
|
|
202
|
+
if self.obj_body == self.ACK_STOP:
|
|
203
|
+
# Indicates the current TCP link receives a STOP signal and remove from the transport_dict
|
|
204
|
+
_transport = self.factory.transport_dict.pop(self.transport, None)
|
|
205
|
+
logger.debug(f"接收到b'STOP_' self.sequence_number {self.sequence_number} ")
|
|
206
|
+
self.post_process()
|
|
207
|
+
|
|
208
|
+
|
|
209
|
+
class MessageServerFactory(protocol.ServerFactory):
|
|
210
|
+
def __init__(self) -> None:
|
|
211
|
+
"""
|
|
212
|
+
transport_dict: links that have not completed data transmission.
|
|
213
|
+
transport_list: Records all TCP links. Appends TCP link to the transport list when a new TCP link is established.
|
|
214
|
+
"""
|
|
215
|
+
self.transport_dict = {}
|
|
216
|
+
self.transport_list = []
|
|
217
|
+
|
|
218
|
+
def is_all_connection_closed(self):
|
|
219
|
+
return len(self.transport_dict) == 0
|
|
@@ -1,10 +1,10 @@
|
|
|
1
|
-
cipher_list = ":".join([
|
|
2
|
-
'ECDHE-ECDSA-AES128-GCM-SHA256',
|
|
3
|
-
'ECDHE-RSA-AES128-GCM-SHA256',
|
|
4
|
-
'ECDHE-ECDSA-AES256-GCM-SHA384',
|
|
5
|
-
'ECDHE-RSA-AES256-GCM-SHA384',
|
|
6
|
-
'ECDHE-ECDSA-CHACHA20-POLY1305',
|
|
7
|
-
'ECDHE-RSA-CHACHA20-POLY1305',
|
|
8
|
-
'DHE-RSA-AES128-GCM-SHA256',
|
|
9
|
-
'DHE-RSA-AES256-GCM-SHA384'
|
|
10
|
-
]).encode()
|
|
1
|
+
cipher_list = ":".join([
|
|
2
|
+
'ECDHE-ECDSA-AES128-GCM-SHA256',
|
|
3
|
+
'ECDHE-RSA-AES128-GCM-SHA256',
|
|
4
|
+
'ECDHE-ECDSA-AES256-GCM-SHA384',
|
|
5
|
+
'ECDHE-RSA-AES256-GCM-SHA384',
|
|
6
|
+
'ECDHE-ECDSA-CHACHA20-POLY1305',
|
|
7
|
+
'ECDHE-RSA-CHACHA20-POLY1305',
|
|
8
|
+
'DHE-RSA-AES128-GCM-SHA256',
|
|
9
|
+
'DHE-RSA-AES256-GCM-SHA384'
|
|
10
|
+
]).encode()
|
|
@@ -1,15 +1,15 @@
|
|
|
1
|
-
import os
|
|
2
|
-
from pkgutil import iter_modules
|
|
3
|
-
from importlib import import_module
|
|
4
|
-
|
|
5
|
-
"""
|
|
6
|
-
gpu and cpu not implement benchmark function, supplementary benchmarking function implementation
|
|
7
|
-
"""
|
|
8
|
-
|
|
9
|
-
package_path = os.path.dirname(os.path.realpath(__file__))
|
|
10
|
-
for _, module_name, _ in iter_modules([package_path]):
|
|
11
|
-
module = import_module(f"{__name__}.{module_name}")
|
|
12
|
-
for attr_name in dir(module):
|
|
13
|
-
attr = getattr(module, attr_name)
|
|
14
|
-
if callable(attr) and "npu_custom" not in attr_name:
|
|
15
|
-
globals()[attr_name] = attr
|
|
1
|
+
import os
|
|
2
|
+
from pkgutil import iter_modules
|
|
3
|
+
from importlib import import_module
|
|
4
|
+
|
|
5
|
+
"""
|
|
6
|
+
gpu and cpu not implement benchmark function, supplementary benchmarking function implementation
|
|
7
|
+
"""
|
|
8
|
+
|
|
9
|
+
package_path = os.path.dirname(os.path.realpath(__file__))
|
|
10
|
+
for _, module_name, _ in iter_modules([package_path]):
|
|
11
|
+
module = import_module(f"{__name__}.{module_name}")
|
|
12
|
+
for attr_name in dir(module):
|
|
13
|
+
attr = getattr(module, attr_name)
|
|
14
|
+
if callable(attr) and "npu_custom" not in attr_name:
|
|
15
|
+
globals()[attr_name] = attr
|
|
@@ -1,28 +1,28 @@
|
|
|
1
|
-
import torch
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
def npu_apply_adam_w(beta1_power, beta2_power, lr, weight_decay,
|
|
5
|
-
beta1, beta2, eps, grad, max_grad_norm, amsgrad, maximize, out):
|
|
6
|
-
var, m, v = out
|
|
7
|
-
if amsgrad:
|
|
8
|
-
max_grad_norm = (torch.rand(var.shape) * 10.0 - 5.0).to(var.dtype)
|
|
9
|
-
beta1_power_out = beta1_power * beta1
|
|
10
|
-
beta2_power_out = beta2_power * beta2
|
|
11
|
-
var_t = var * (1 + (-lr * weight_decay))
|
|
12
|
-
gt = -grad if maximize else grad
|
|
13
|
-
m_out = m * beta1 - (beta1 + (-1)) * gt
|
|
14
|
-
v_out = v * beta2 - (beta2 + (-1)) * gt * gt
|
|
15
|
-
|
|
16
|
-
if amsgrad:
|
|
17
|
-
max_grad_norm_out = torch.max(max_grad_norm, v_out)
|
|
18
|
-
if (1 - beta2_power_out) == 0:
|
|
19
|
-
beta2_power_out -= eps
|
|
20
|
-
denom = torch.sqrt(torch.div(max_grad_norm_out, (1 - beta2_power_out))) + eps
|
|
21
|
-
else:
|
|
22
|
-
vraintain = torch.div(v_out, (1 - beta2_power_out))
|
|
23
|
-
denom = torch.sqrt(vraintain) + eps
|
|
24
|
-
|
|
25
|
-
if (1 - beta1_power_out) == 0:
|
|
26
|
-
beta1_power_out -= eps
|
|
27
|
-
var_out = var_t + torch.div(-lr * m_out, (1 - beta1_power_out)).div(denom)
|
|
28
|
-
return var_out
|
|
1
|
+
import torch
|
|
2
|
+
|
|
3
|
+
|
|
4
|
+
def npu_apply_adam_w(beta1_power, beta2_power, lr, weight_decay,
|
|
5
|
+
beta1, beta2, eps, grad, max_grad_norm, amsgrad, maximize, out):
|
|
6
|
+
var, m, v = out
|
|
7
|
+
if amsgrad:
|
|
8
|
+
max_grad_norm = (torch.rand(var.shape) * 10.0 - 5.0).to(var.dtype)
|
|
9
|
+
beta1_power_out = beta1_power * beta1
|
|
10
|
+
beta2_power_out = beta2_power * beta2
|
|
11
|
+
var_t = var * (1 + (-lr * weight_decay))
|
|
12
|
+
gt = -grad if maximize else grad
|
|
13
|
+
m_out = m * beta1 - (beta1 + (-1)) * gt
|
|
14
|
+
v_out = v * beta2 - (beta2 + (-1)) * gt * gt
|
|
15
|
+
|
|
16
|
+
if amsgrad:
|
|
17
|
+
max_grad_norm_out = torch.max(max_grad_norm, v_out)
|
|
18
|
+
if (1 - beta2_power_out) == 0:
|
|
19
|
+
beta2_power_out -= eps
|
|
20
|
+
denom = torch.sqrt(torch.div(max_grad_norm_out, (1 - beta2_power_out))) + eps
|
|
21
|
+
else:
|
|
22
|
+
vraintain = torch.div(v_out, (1 - beta2_power_out))
|
|
23
|
+
denom = torch.sqrt(vraintain) + eps
|
|
24
|
+
|
|
25
|
+
if (1 - beta1_power_out) == 0:
|
|
26
|
+
beta1_power_out -= eps
|
|
27
|
+
var_out = var_t + torch.div(-lr * m_out, (1 - beta1_power_out)).div(denom)
|
|
28
|
+
return var_out, m_out, v_out
|
|
@@ -1,19 +1,19 @@
|
|
|
1
|
-
def npu_confusion_transpose(data, perm, shape, transpose_first):
|
|
2
|
-
if transpose_first:
|
|
3
|
-
output = data.permute(*perm).contiguous().view(shape)
|
|
4
|
-
else:
|
|
5
|
-
output = data.view(shape).permute(*perm)
|
|
6
|
-
return output
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
def npu_confusion_transpose_backward(grad, perm, shape, transpose_first):
|
|
10
|
-
shape_cal = shape if transpose_first else [shape[perm_dim] for perm_dim in perm]
|
|
11
|
-
perm_cal = [0] * len(perm)
|
|
12
|
-
for i, perm_dim in enumerate(perm):
|
|
13
|
-
perm_cal[perm_dim] = i
|
|
14
|
-
|
|
15
|
-
if transpose_first:
|
|
16
|
-
result = grad.permute(*perm_cal).reshape(shape_cal)
|
|
17
|
-
else:
|
|
18
|
-
result = grad.reshape(shape_cal).permute(*perm_cal)
|
|
19
|
-
return result.cpu()
|
|
1
|
+
def npu_confusion_transpose(data, perm, shape, transpose_first):
|
|
2
|
+
if transpose_first:
|
|
3
|
+
output = data.permute(*perm).contiguous().view(shape)
|
|
4
|
+
else:
|
|
5
|
+
output = data.view(shape).permute(*perm)
|
|
6
|
+
return output
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
def npu_confusion_transpose_backward(grad, perm, shape, transpose_first):
|
|
10
|
+
shape_cal = shape if transpose_first else [shape[perm_dim] for perm_dim in perm]
|
|
11
|
+
perm_cal = [0] * len(perm)
|
|
12
|
+
for i, perm_dim in enumerate(perm):
|
|
13
|
+
perm_cal[perm_dim] = i
|
|
14
|
+
|
|
15
|
+
if transpose_first:
|
|
16
|
+
result = grad.permute(*perm_cal).reshape(shape_cal)
|
|
17
|
+
else:
|
|
18
|
+
result = grad.reshape(shape_cal).permute(*perm_cal)
|
|
19
|
+
return result.cpu()
|