mindstudio-probe 1.0.4__tar.gz → 1.1.1__tar.gz

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/PKG-INFO

@@ -1,13 +1,12 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 1.0.4
+Version: 1.1.1
 Summary: Pytorch Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
 Author-email: pmail_mindstudio@huawei.com
 License: Apache License 2.0
 Keywords: pytorch msprobe ascend
-Platform: UNKNOWN
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Education
 Classifier: Intended Audience :: Science/Research
@@ -20,6 +19,18 @@ Classifier: Topic :: Software Development :: Libraries
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.6.2
 License-File: LICENSE
+Requires-Dist: wheel
+Requires-Dist: einops
+Requires-Dist: numpy<2.0
+Requires-Dist: pandas<2.1,>=1.3.5
+Requires-Dist: pyyaml
+Requires-Dist: rich
+Requires-Dist: tqdm
+Requires-Dist: openpyxl
+Requires-Dist: pyopenssl
+Requires-Dist: twisted
+Requires-Dist: matplotlib
+Requires-Dist: tensorboard
+Requires-Dist: tabulate
 
 MindStudio-Probe is a set of tools for diagnosing and improving model accuracy on Ascend NPU, including API acc checker, ptdbg, grad tool etc.
-

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/mindstudio_probe.egg-info/PKG-INFO

@@ -1,13 +1,12 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 1.0.4
+Version: 1.1.1
 Summary: Pytorch Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
 Author-email: pmail_mindstudio@huawei.com
 License: Apache License 2.0
 Keywords: pytorch msprobe ascend
-Platform: UNKNOWN
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Education
 Classifier: Intended Audience :: Science/Research
@@ -20,6 +19,18 @@ Classifier: Topic :: Software Development :: Libraries
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.6.2
 License-File: LICENSE
+Requires-Dist: wheel
+Requires-Dist: einops
+Requires-Dist: numpy<2.0
+Requires-Dist: pandas<2.1,>=1.3.5
+Requires-Dist: pyyaml
+Requires-Dist: rich
+Requires-Dist: tqdm
+Requires-Dist: openpyxl
+Requires-Dist: pyopenssl
+Requires-Dist: twisted
+Requires-Dist: matplotlib
+Requires-Dist: tensorboard
+Requires-Dist: tabulate
 
 MindStudio-Probe is a set of tools for diagnosing and improving model accuracy on Ascend NPU, including API acc checker, ptdbg, grad tool etc.
-

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/mindstudio_probe.egg-info/SOURCES.txt

@@ -20,6 +20,8 @@ msprobe/core/advisor/advisor_result.py
 msprobe/core/common/const.py
 msprobe/core/common/exceptions.py
 msprobe/core/common/file_utils.py
+msprobe/core/common/inplace_op_checker.py
+msprobe/core/common/inplace_ops.yaml
 msprobe/core/common/log.py
 msprobe/core/common/utils.py
 msprobe/core/compare/acc_compare.py
@@ -29,6 +31,10 @@ msprobe/core/compare/highlight.py
 msprobe/core/compare/multiprocessing_compute.py
 msprobe/core/compare/npy_compare.py
 msprobe/core/compare/utils.py
+msprobe/core/compare/layer_mapping/__init__.py
+msprobe/core/compare/layer_mapping/data_scope_parser.py
+msprobe/core/compare/layer_mapping/layer_mapping.py
+msprobe/core/compare/layer_mapping/postprocess_pass.py
 msprobe/core/data_dump/data_collector.py
 msprobe/core/data_dump/json_writer.py
 msprobe/core/data_dump/scope.py
@@ -40,10 +46,17 @@ msprobe/core/grad_probe/__init__.py
 msprobe/core/grad_probe/constant.py
 msprobe/core/grad_probe/grad_compare.py
 msprobe/core/grad_probe/utils.py
+msprobe/core/overflow_check/abnormal_scene.py
+msprobe/core/overflow_check/api_info.py
+msprobe/core/overflow_check/checker.py
+msprobe/core/overflow_check/filter.py
+msprobe/core/overflow_check/ignore_rules.yaml
+msprobe/core/overflow_check/level.py
+msprobe/core/overflow_check/utils.py
 msprobe/docs/01.installation.md
 msprobe/docs/02.config_introduction.md
 msprobe/docs/03.config_examples.md
-msprobe/docs/04.acl_config_examples.md
+msprobe/docs/04.kernel_dump_PyTorch.md
 msprobe/docs/05.data_dump_PyTorch.md
 msprobe/docs/06.data_dump_MindSpore.md
 msprobe/docs/07.accuracy_checker_PyTorch.md
@@ -55,9 +68,17 @@ msprobe/docs/12.overflow_check_PyTorch.md
 msprobe/docs/13.overflow_check_MindSpore.md
 msprobe/docs/14.data_parse_PyTorch.md
 msprobe/docs/15.free_benchmarking_PyTorch.md
+msprobe/docs/16.free_benchmarking_MindSpore.md
 msprobe/docs/17.grad_probe.md
-msprobe/docs/FAQ_PyTorch.md
+msprobe/docs/18.online_dispatch.md
+msprobe/docs/19.monitor.md
+msprobe/docs/20.monitor_performance_baseline.md
+msprobe/docs/21.visualization_PyTorch.md
+msprobe/docs/22.visualization_MindSpore.md
+msprobe/docs/23.tool_function_introduction.md
+msprobe/docs/FAQ.md
 msprobe/docs/S02.report_free_benchmarking_validation_performance_baseline.md
+msprobe/docs/data_dump_Mindspore/dynamic_graph_quick_start_example.md
 msprobe/docs/img/BLOOM-7B_1.png
 msprobe/docs/img/BLOOM-7B_2.png
 msprobe/docs/img/BLOOM-7B_3.png
@@ -77,6 +98,7 @@ msprobe/docs/img/accuracy_checking_result.png
 msprobe/docs/img/api_precision_compare_details.png
 msprobe/docs/img/api_precision_compare_result.png
 msprobe/docs/img/auto_analyze_log.png
+msprobe/docs/img/compare_result.png
 msprobe/docs/img/compare_result_pkl.png
 msprobe/docs/img/compare_result_pkl_md5.png.png
 msprobe/docs/img/cpu_info.png
@@ -88,6 +110,10 @@ msprobe/docs/img/grad_probe_image-3.png
 msprobe/docs/img/grad_probe_image-4.png
 msprobe/docs/img/grad_probe_image.png
 msprobe/docs/img/module_compare.png
+msprobe/docs/img/ms_dump.png
+msprobe/docs/img/ms_layer.png
+msprobe/docs/img/pt_dump.png
+msprobe/docs/img/monitor/cpu_info.png
 msprobe/mindspore/__init__.py
 msprobe/mindspore/cell_processor.py
 msprobe/mindspore/ms_config.py
@@ -99,9 +125,13 @@ msprobe/mindspore/api_accuracy_checker/api_accuracy_checker.py
 msprobe/mindspore/api_accuracy_checker/api_info.py
 msprobe/mindspore/api_accuracy_checker/api_runner.py
 msprobe/mindspore/api_accuracy_checker/base_compare_algorithm.py
+msprobe/mindspore/api_accuracy_checker/checker_support_api.yaml
 msprobe/mindspore/api_accuracy_checker/cmd_parser.py
 msprobe/mindspore/api_accuracy_checker/compute_element.py
+msprobe/mindspore/api_accuracy_checker/data_manager.py
 msprobe/mindspore/api_accuracy_checker/main.py
+msprobe/mindspore/api_accuracy_checker/multi_api_accuracy_checker.py
+msprobe/mindspore/api_accuracy_checker/multi_data_manager.py
 msprobe/mindspore/api_accuracy_checker/type_mapping.py
 msprobe/mindspore/api_accuracy_checker/utils.py
 msprobe/mindspore/common/const.py
@@ -121,8 +151,11 @@ msprobe/mindspore/dump/kernel_graph_dump.py
 msprobe/mindspore/dump/kernel_kbyk_dump.py
 msprobe/mindspore/dump/hook_cell/api_registry.py
 msprobe/mindspore/dump/hook_cell/hook_cell.py
+msprobe/mindspore/dump/hook_cell/primitive_hooks.py
 msprobe/mindspore/dump/hook_cell/support_wrap_ops.yaml
 msprobe/mindspore/dump/hook_cell/wrap_api.py
+msprobe/mindspore/dym_loader/hook_dynamic_loader.cc
+msprobe/mindspore/dym_loader/hook_dynamic_loader.h
 msprobe/mindspore/free_benchmark/__init__.py
 msprobe/mindspore/free_benchmark/api_pynative_self_check.py
 msprobe/mindspore/free_benchmark/self_check_tool_factory.py
@@ -131,9 +164,6 @@ msprobe/mindspore/free_benchmark/common/config.py
 msprobe/mindspore/free_benchmark/common/handler_params.py
 msprobe/mindspore/free_benchmark/common/utils.py
 msprobe/mindspore/free_benchmark/data/support_wrap_ops.yaml
-msprobe/mindspore/free_benchmark/decorator/__init__.py
-msprobe/mindspore/free_benchmark/decorator/dec_forward.py
-msprobe/mindspore/free_benchmark/decorator/decorator_factory.py
 msprobe/mindspore/free_benchmark/handler/__init__.py
 msprobe/mindspore/free_benchmark/handler/base_handler.py
 msprobe/mindspore/free_benchmark/handler/check_handler.py
@@ -177,6 +207,9 @@ msprobe/pytorch/api_accuracy_checker/compare/api_precision_threshold.yaml
 msprobe/pytorch/api_accuracy_checker/compare/compare.py
 msprobe/pytorch/api_accuracy_checker/compare/compare_column.py
 msprobe/pytorch/api_accuracy_checker/compare/compare_utils.py
+msprobe/pytorch/api_accuracy_checker/generate_op_script/config_op.json
+msprobe/pytorch/api_accuracy_checker/generate_op_script/op_generator.py
+msprobe/pytorch/api_accuracy_checker/generate_op_script/operator_replication.template
 msprobe/pytorch/api_accuracy_checker/run_ut/.keep
 msprobe/pytorch/api_accuracy_checker/run_ut/__init__.py
 msprobe/pytorch/api_accuracy_checker/run_ut/data_generate.py
@@ -189,8 +222,10 @@ msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/__init__.py
 msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/attl.py
 msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/client.py
 msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/device_dispatch.py
+msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/dump_dispatch.py
 msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/server.py
-msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/ssl_config.py
+msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/torch_ops_config.yaml
+msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/utils.py
 msprobe/pytorch/bench_functions/__init__.py
 msprobe/pytorch/bench_functions/apply_adam_w.py
 msprobe/pytorch/bench_functions/confusion_transpose.py
@@ -215,6 +250,7 @@ msprobe/pytorch/compare/pt_compare.py
 msprobe/pytorch/debugger/__init__.py
 msprobe/pytorch/debugger/debugger_config.py
 msprobe/pytorch/debugger/precision_debugger.py
+msprobe/pytorch/dump/kernel_dump/kernel_config.py
 msprobe/pytorch/free_benchmark/__init__.py
 msprobe/pytorch/free_benchmark/main.py
 msprobe/pytorch/free_benchmark/common/__init__.py
@@ -243,8 +279,7 @@ msprobe/pytorch/free_benchmark/result_handlers/fix_handler.py
 msprobe/pytorch/free_benchmark/result_handlers/handler_factory.py
 msprobe/pytorch/free_benchmark/result_handlers/preheat_handler.py
 msprobe/pytorch/functional/__init__.py
-msprobe/pytorch/functional/data_processor.py
-msprobe/pytorch/functional/dump_module.py
+msprobe/pytorch/functional/module_dump.py
 msprobe/pytorch/grad_probe/__init__.py
 msprobe/pytorch/grad_probe/grad_monitor.py
 msprobe/pytorch/grad_probe/grad_stat_csv.py
@@ -260,6 +295,22 @@ msprobe/pytorch/hook_module/wrap_npu_custom.py
 msprobe/pytorch/hook_module/wrap_tensor.py
 msprobe/pytorch/hook_module/wrap_torch.py
 msprobe/pytorch/hook_module/wrap_vf.py
+msprobe/pytorch/monitor/__init__.py
+msprobe/pytorch/monitor/anomaly_analyse.py
+msprobe/pytorch/monitor/anomaly_detect.py
+msprobe/pytorch/monitor/features.py
+msprobe/pytorch/monitor/module_hook.py
+msprobe/pytorch/monitor/module_metric.py
+msprobe/pytorch/monitor/module_spec_verifier.py
+msprobe/pytorch/monitor/optimizer_collect.py
+msprobe/pytorch/monitor/utils.py
+msprobe/pytorch/monitor/visualizer.py
+msprobe/pytorch/monitor/distributed/__init__.py
+msprobe/pytorch/monitor/distributed/distributed_ops.yaml
+msprobe/pytorch/monitor/distributed/stack_blacklist.yaml
+msprobe/pytorch/monitor/distributed/wrap_distributed.py
+msprobe/pytorch/monitor/unittest/__init__.py
+msprobe/pytorch/monitor/unittest/test_monitor.py
 msprobe/pytorch/online_dispatch/__init__.py
 msprobe/pytorch/online_dispatch/compare.py
 msprobe/pytorch/online_dispatch/dispatch.py
@@ -277,4 +328,18 @@ msprobe/pytorch/parse_tool/lib/interactive_cli.py
 msprobe/pytorch/parse_tool/lib/parse_exception.py
 msprobe/pytorch/parse_tool/lib/parse_tool.py
 msprobe/pytorch/parse_tool/lib/utils.py
-msprobe/pytorch/parse_tool/lib/visualization.py
+msprobe/pytorch/parse_tool/lib/visualization.py
+msprobe/visualization/__init__.py
+msprobe/visualization/graph_service.py
+msprobe/visualization/utils.py
+msprobe/visualization/builder/__init__.py
+msprobe/visualization/builder/graph_builder.py
+msprobe/visualization/builder/msprobe_adapter.py
+msprobe/visualization/compare/__init__.py
+msprobe/visualization/compare/graph_comparator.py
+msprobe/visualization/compare/mode_adapter.py
+msprobe/visualization/graph/__init__.py
+msprobe/visualization/graph/base_node.py
+msprobe/visualization/graph/graph.py
+msprobe/visualization/graph/node_colors.py
+msprobe/visualization/graph/node_op.py

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/mindstudio_probe.egg-info/entry_points.txt

@@ -1,3 +1,2 @@
 [console_scripts]
 msprobe = msprobe.msprobe:main
-

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/mindstudio_probe.egg-info/requires.txt

@@ -9,3 +9,5 @@ openpyxl
 pyopenssl
 twisted
 matplotlib
+tensorboard
+tabulate

mindstudio_probe-1.1.1/msprobe/README.md

@@ -0,0 +1,167 @@
+# 📖 msprobe 使用手册
+
+![version](https://img.shields.io/badge/version-1.0.4-blueviolet)
+![python](https://img.shields.io/badge/python-3.8|3.9|3.10-blue)
+![platform](https://img.shields.io/badge/platform-Linux-yellow)
+
+**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。
+
+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。
+
+```python
+from msprobe.pytorch import PrecisionDebugger
+
+debugger = PrecisionDebugger(config_path='./config.json')
+...
+debugger.start() # 一般在训练循环开头启动工具
+... # 循环体
+debugger.stop() # 一般在训练循环末尾结束工具
+debugger.step() # 在训练循环的最后需要重置工具，非循环场景不需要
+```
+
+此外，根据以下规则，可以通过环境变量设置日志级别。
+- MSPROBE_LOG_LEVEL=4，不打印任何日志；
+- MSPROBE_LOG_LEVEL=3，仅打印 ERROR；
+- MSPROBE_LOG_LEVEL=2，仅打印 WARNING、ERROR；
+- MSPROBE_LOG_LEVEL=1，仅打印 INFO、WARNING、ERROR（默认配置）；
+- MSPROBE_LOG_LEVEL=0，打印 DEBUG、INFO、WARNING、ERROR。
+
+例如在 shell 脚本：
+
+```shell
+export MSPROBE_LOG_LEVEL={x}
+```
+**config.json** 的配置要求和各功能具体的使用指导详见后续章节。
+
+## 环境和依赖
+
+- 硬件环境请参见《[昇腾产品形态说明](https://gitee.com/link?target=https%3A%2F%2Fwww.hiascend.com%2Fdocument%2Fdetail%2Fzh%2Fcanncommercial%2F80RC22%2Fquickstart%2Fquickstart%2Fquickstart_18_0002.html)》。
+- 软件环境请参见《[CANN 软件安装指南](https://gitee.com/link?target=https%3A%2F%2Fwww.hiascend.com%2Fdocument%2Fdetail%2Fzh%2Fcanncommercial%2F80RC22%2Fsoftwareinst%2Finstg%2Finstg_0000.html%3FMode%3DPmIns%26OS%3DUbuntu%26Software%3DcannToolKit)》安装昇腾设备开发或运行环境，即toolkit软件包。
+
+以上环境依赖请根据实际环境选择适配的版本。
+
+## 版本配套说明
+
+- msprobe支持AscendPyTorch 1.11.0或更高版本，支持的PyTorch和CANN以及PyTorch和python软件版本配套关系请参见《[Ascend Extension for PyTorch插件](https://gitee.com/ascend/pytorch)》。
+- msprobe支持MindSpore 2.4.0或更高版本，支持的MindSpore和CANN以及MindSpore和python软件版本配套关系请参见《[MindSpore版本发布列表](https://www.mindspore.cn/versions)》。
+- msprobe支持的固件驱动版本与配套CANN软件支持的固件驱动版本相同，开发者可通过“[昇腾社区-固件与驱动](https://gitee.com/link?target=https%3A%2F%2Fwww.hiascend.com%2Fhardware%2Ffirmware-drivers%2Fcommunity%3Fproduct%3D2%26model%3D28%26cann%3D8.0.RC3.alpha003%26driver%3D1.0.25.alpha)”页面根据产品型号与CANN软件版本获取配套的固件与驱动。
+
+
+## 🚨 工具限制与注意事项
+
+**1. Pytorch 框架下，工具暂不支持 Fully Sharded Data Parallel(FSDP)。**
+
+## ⚙️ [安装](./docs/01.installation.md)
+
+## 🛠️ config.json [介绍](./docs/02.config_introduction.md) 和 [示例](./docs/03.config_examples.md)
+
+## 🧰 主要功能
+
+### 0 用前必看
+
+使用工具前，建议先浏览[**工具功能模块简介、适用场景和当前版本局限性**](./docs/23.tool_function_introduction.md)，了解功能特性。
+
+### 1 数据采集
+
+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。
+
+[PyTorch 场景的数据采集](./docs/05.data_dump_PyTorch.md)
+
+[MindSpore 场景的数据采集](./docs/06.data_dump_MindSpore.md)
+
+### 2 精度预检
+
+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。
+
+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)
+
+MindSpore 动态图场景的[离线预检](./docs/09.accuracy_checker_MindSpore.md)
+
+### 3 精度比对
+
+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。
+
+[PyTorch 场景的精度比对](./docs/10.accuracy_compare_PyTorch.md)
+
+[MindSpore 场景的精度比对](./docs/11.accuracy_compare_MindSpore.md)
+
+### 4 溢出检测与解析
+
+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。
+
+[PyTorch 场景的溢出检测与解析](./docs/12.overflow_check_PyTorch.md)
+
+[MindSpore 场景的溢出检测与解析](./docs/13.overflow_check_MindSpore.md)
+
+### 5 数据解析
+
+该功能用于比对前后两次 NPU ACL 层级 dump 数据的一致性。
+
+[PyTorch 场景的数据解析](./docs/14.data_parse_PyTorch.md)
+
+### 6 无标杆比对
+
+[PyTorch 场景的无标杆比对](./docs/15.free_benchmarking_PyTorch.md)
+
+[MindSpore 场景的无标杆比对](./docs/16.free_benchmarking_MindSpore.md)
+
+### 7 梯度状态监测
+
+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。
+
+[兼容 PyTorch 和 MindSpore 框架的梯度监测](./docs/17.grad_probe.md)
+
+### 8 在线精度比对
+
+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。
+
+[PyTorch 场景的在线精度比对](./docs/18.online_dispatch.md)
+
+### 9 训练状态监控
+
+该功能收集和聚合模型训练过程中的网络层，优化器， 通信算子的中间值，帮助诊断模型训练过程中计算， 通信，优化器各部分出现的异常情况。
+
+[PyTorch 场景的训练状态监控](./docs/19.monitor.md)
+
+### 10 分级可视化构图比对
+
+该功能将msprobe工具dump的精度数据进行解析，还原模型图结构，实现模型各个层级的精度数据比对，方便用户理解模型结构、分析精度问题。
+
+[PyTorch 场景的分级可视化构图比对](./docs/21.visualization_PyTorch.md)
+
+[MindSpore 场景的分级可视化构图比对](./docs/22.visualization_MindSpore.md)
+
+## 🌟 新版本特性
+
+若查看历史版本特性，请参见[安装](./docs/01.installation.md)。
+
+【数据采集】
+- 支持 config.json 中的 step 传入范围；
+- 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：
+    ```bash
+    ****************************************
+    *      msprobe ends successfully.      *
+    ****************************************
+    ```
+    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。
+
+【精度预检】
+- 在 PyTorch 场景，支持部分 NPU 融合算子预检。
+
+【精度比对】
+- 解决了使用 MindSpore 需要安装 PyTorch 的问题。
+
+【无标杆比对】
+- 补充在 PyTorch 场景的性能基线报告；
+- 支持 MindSpore 场景的 change_value 扰动模式。
+
+## 📑 补充材料
+
+[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)
+
+## ❗ 免责声明
+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。
+
+## ❓ FAQ
+
+[FAQ for PyTorch](./docs/FAQ.md)

mindstudio_probe-1.1.1/msprobe/__init__.py

@@ -0,0 +1,16 @@
+# Copyright (c) 2024-2024, Huawei Technologies Co., Ltd.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from msprobe.core.grad_probe.grad_compare import GradComparator

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/msprobe/config.json

@@ -4,15 +4,11 @@
     "rank": [],  
     "step": [],  
     "level": "L1", 
-    "seed": 1234,
-    "is_deterministic": false, 
     "enable_dataloader": false,
-    "acl_config": "",
     "tensor": {
         "scope": [], 
         "list":[], 
-        "data_mode": ["all"], 
-        "backward_input": [],
+        "data_mode": ["all"],
         "file_format": "npy" 
     },
     "statistics": {

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/msprobe/core/advisor/advisor.py

@@ -1,8 +1,7 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-"""
-# Copyright (C) 2022-2024. Huawei Technologies Co., Ltd. All rights reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
+# Copyright (c) 2022-2024, Huawei Technologies Co., Ltd.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0  (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
@@ -13,7 +12,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""
 
 import os
 
@@ -24,15 +22,17 @@ from msprobe.core.common.utils import CompareException
 from msprobe.core.common.file_utils import FileChecker
 from msprobe.core.common.const import Const, CompareConst, FileCheckConst
 
+
 class Advisor:
     """
     Class for generate advisor
     """
 
-    def __init__(self, input_data, out_path=""):
+    def __init__(self, input_data, out_path="", suffix=""):
         self.input_data = input_data
         self.out_path = os.path.realpath(out_path)
         self.file_type = None
+        self.suffix = suffix
 
     @staticmethod
     def deterministic_advisor(message, node_name):
@@ -62,7 +62,12 @@ class Advisor:
                             .format(item[CompareConst.NPU_NAME]))
 
     def gen_advisor_result(self, pd_data):
-        first_failing_data = pd_data.iloc[0]
+        try:
+            first_failing_data = pd_data.iloc[0]
+        except IndexError as e:
+            err_msg = "index out of bounds error occurs, pd_data is empty, please check!"
+            logger.error(err_msg)
+            raise CompareException(CompareException.INDEX_OUT_OF_BOUNDS_ERROR) from e
         node_name = first_failing_data[CompareConst.NPU_NAME]
         index = first_failing_data['index']
         message = self.gen_advisor_message(node_name)
@@ -87,7 +92,7 @@ class Advisor:
         return message
 
     def analysis(self):
-        self._check_path_vaild()
+        self._check_path_valid()
         analyze_data = self._parse_input_data()
         logger.info("Start analyzing the comparison result: %s" % self.file_type)
         self.analyze_unmatched(analyze_data)
@@ -103,7 +108,7 @@ class Advisor:
         else:
             result = self.gen_advisor_result(failing_data)
         message_list = result.print_advisor_log()
-        result.gen_summary_file(self.out_path, message_list)
+        result.gen_summary_file(self.out_path, message_list, suffix=self.suffix)
 
     def _parse_input_data(self):
         data_columns = self.input_data.columns.values
@@ -119,6 +124,6 @@ class Advisor:
         df = self.input_data.reset_index()
         return df
 
-    def _check_path_vaild(self):
+    def _check_path_valid(self):
         out_path_checker = FileChecker(self.out_path, FileCheckConst.DIR, FileCheckConst.WRITE_ABLE)
         out_path_checker.common_check()

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/msprobe/core/advisor/advisor_const.py

@@ -1,8 +1,7 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-"""
-# Copyright (C) 2022-2024. Huawei Technologies Co., Ltd. All rights reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
+# Copyright (c) 2022-2024, Huawei Technologies Co., Ltd.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0  (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
@@ -13,7 +12,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""
 
 
 class AdvisorConst:
@@ -35,7 +33,8 @@ class AdvisorConst:
     # advisor message
     NO_ERR_SUGGEST = "All data in comparison result meets the accuracy requirements."
     FORWARD_INPUT_SUGGEST = "1. Analyze the model to view the input source.\n" \
-                            "2. Check whether an inplace API causes the output result to overwrite the input result. That is, the fault is actually caused by a computation error.\n" \
+                            "2. Check whether an inplace API causes the output result to overwrite the input result. "\
+                            "That is, the fault is actually caused by a computation error.\n" \
                             "3. The fault may be caused by memory corruption and further analysis is required."
     FORWARD_OUTPUT_SUGGEST = "This is a forward API computation error. Check the computation implementation."
     BACKWARD_INPUT_SUGGEST = "Check whether the forward computation result is affected."

{mindstudio-probe-1.0.4 → mindstudio_probe-1.1.1}/msprobe/core/advisor/advisor_result.py

@@ -1,8 +1,7 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-"""
-# Copyright (C) 2022-2024. Huawei Technologies Co., Ltd. All rights reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
+# Copyright (c) 2022-2024, Huawei Technologies Co., Ltd.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0  (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
@@ -13,13 +12,12 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""
 import os
 import time
 
 from msprobe.core.advisor.advisor_const import AdvisorConst
 from msprobe.core.common.log import logger
-from msprobe.core.common.const import Const, FileCheckConst
+from msprobe.core.common.const import FileCheckConst
 from msprobe.core.common.file_utils import change_mode, FileOpen
 
 
@@ -34,8 +32,8 @@ class AdvisorResult:
         self.advisor_message = message
 
     @staticmethod
-    def gen_summary_file(out_path, message_list):
-        file_name = 'advisor_{}.txt'.format(time.strftime("%Y%m%d%H%M%S", time.localtime(time.time())))
+    def gen_summary_file(out_path, message_list, suffix):
+        file_name = 'advisor{}_{}.txt'.format(suffix, time.strftime("%Y%m%d%H%M%S", time.localtime(time.time())))
         result_file = os.path.join(out_path, file_name)
         try:
             with FileOpen(result_file, 'w+') as output_file:
@@ -50,9 +48,11 @@ class AdvisorResult:
 
     def print_advisor_log(self):
         logger.info("The summary of the expert advice is as follows: ")
-        message_list = [AdvisorConst.LINE + AdvisorConst.COLON + str(self.line),
-                        AdvisorConst.SUSPECT_NODES + AdvisorConst.COLON + self.suspect_node,
-                        AdvisorConst.ADVISOR_SUGGEST + AdvisorConst.COLON + self.advisor_message]
+        message_list = [
+            AdvisorConst.LINE + AdvisorConst.COLON + str(self.line),
+            AdvisorConst.SUSPECT_NODES + AdvisorConst.COLON + self.suspect_node,
+            AdvisorConst.ADVISOR_SUGGEST + AdvisorConst.COLON + self.advisor_message
+        ]
         for message in message_list:
             logger.info(message)
         return message_list