mindstudio-probe 1.2.1__tar.gz → 1.3.0__tar.gz

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 1.2.1
+Version: 1.3.0
 Summary: Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
@@ -22,12 +22,12 @@ Requires-Python: >=3.6.2
 License-File: LICENSE
 Requires-Dist: wheel
 Requires-Dist: einops
-Requires-Dist: numpy<2.0
+Requires-Dist: numpy<2.0,>=1.23.0
 Requires-Dist: pandas<2.1,>=1.3.5
 Requires-Dist: pyyaml
 Requires-Dist: rich
 Requires-Dist: tqdm
-Requires-Dist: openpyxl
+Requires-Dist: openpyxl>=3.0.6
 Requires-Dist: pyopenssl
 Requires-Dist: twisted
 Requires-Dist: matplotlib

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/mindstudio_probe.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mindstudio-probe
-Version: 1.2.1
+Version: 1.3.0
 Summary: Ascend Probe Utils
 Home-page: https://gitee.com/ascend/mstt/tree/master/debug/accuracy_tools/msprobe
 Author: Ascend Team
@@ -22,12 +22,12 @@ Requires-Python: >=3.6.2
 License-File: LICENSE
 Requires-Dist: wheel
 Requires-Dist: einops
-Requires-Dist: numpy<2.0
+Requires-Dist: numpy<2.0,>=1.23.0
 Requires-Dist: pandas<2.1,>=1.3.5
 Requires-Dist: pyyaml
 Requires-Dist: rich
 Requires-Dist: tqdm
-Requires-Dist: openpyxl
+Requires-Dist: openpyxl>=3.0.6
 Requires-Dist: pyopenssl
 Requires-Dist: twisted
 Requires-Dist: matplotlib

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/mindstudio_probe.egg-info/SOURCES.txt

@@ -19,6 +19,7 @@ msprobe/core/advisor/advisor.py
 msprobe/core/advisor/advisor_const.py
 msprobe/core/advisor/advisor_result.py
 msprobe/core/common/const.py
+msprobe/core/common/decorator.py
 msprobe/core/common/exceptions.py
 msprobe/core/common/file_utils.py
 msprobe/core/common/inplace_op_checker.py
@@ -38,6 +39,8 @@ msprobe/core/compare/layer_mapping/layer_mapping.py
 msprobe/core/compare/layer_mapping/postprocess_pass.py
 msprobe/core/compare/merge_result/merge_result.py
 msprobe/core/compare/merge_result/merge_result_cli.py
+msprobe/core/compare/merge_result/utils.py
+msprobe/core/data_dump/api_registry.py
 msprobe/core/data_dump/data_collector.py
 msprobe/core/data_dump/json_writer.py
 msprobe/core/data_dump/scope.py
@@ -83,6 +86,10 @@ msprobe/docs/24.code_mapping_Mindspore.md
 msprobe/docs/25.tool_function_introduction.md
 msprobe/docs/26.data_dump_PyTorch_baseline.md
 msprobe/docs/27.dump_json_instruction.md
+msprobe/docs/28.debugger_save_instruction.md
+msprobe/docs/28.kernel_dump_MindSpore.md
+msprobe/docs/29.data_dump_MSAdapter.md
+msprobe/docs/30.overflow_check_MSAdapter.md
 msprobe/docs/FAQ.md
 msprobe/docs/S02.report_free_benchmarking_validation_performance_baseline.md
 msprobe/docs/accuracy_checker_MindSpore/accuracy_checker_MindSpore_baseline.md
@@ -124,12 +131,14 @@ 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/docs/img/monitor/step_count_per_record.png
 msprobe/docs/img/visualization/fuzzy_match_ms.png
 msprobe/docs/img/visualization/fuzzy_match_pt.png
 msprobe/docs/img/visualization/tensorboard_1.png
 msprobe/docs/img/visualization/tensorboard_2.png
 msprobe/docs/img/visualization/vis_browser_1.png
 msprobe/docs/img/visualization/vis_browser_2.png
+msprobe/docs/img/visualization/vis_match_info.png
 msprobe/docs/img/visualization/vis_precision_info.png
 msprobe/docs/img/visualization/vis_search_info.png
 msprobe/docs/img/visualization/vis_show_info.png
@@ -164,8 +173,11 @@ 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/torch_mindtorch_importer.py
 msprobe/mindspore/api_accuracy_checker/type_mapping.py
 msprobe/mindspore/api_accuracy_checker/utils.py
+msprobe/mindspore/api_accuracy_checker/bench_functions/flash_attention_score.py
+msprobe/mindspore/api_accuracy_checker/bench_functions/fusion_operator.py
 msprobe/mindspore/code_mapping/__init__.py
 msprobe/mindspore/code_mapping/bind.py
 msprobe/mindspore/code_mapping/cmd_parser.py
@@ -188,11 +200,10 @@ msprobe/mindspore/dump/dump_tool_factory.py
 msprobe/mindspore/dump/jit_dump.py
 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/api_register.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/dump/kernel_dump/kernel_config.py
 msprobe/mindspore/dym_loader/hook_dynamic_loader.cc
 msprobe/mindspore/dym_loader/hook_dynamic_loader.h
@@ -225,12 +236,20 @@ msprobe/mindspore/grad_probe/hook.py
 msprobe/mindspore/grad_probe/utils.py
 msprobe/mindspore/mindtorch/__init__.py
 msprobe/mindspore/mindtorch/mindtorch_adaptor.py
+msprobe/mindspore/monitor/anomaly_detect.py
+msprobe/mindspore/monitor/features.py
+msprobe/mindspore/monitor/module_hook.py
+msprobe/mindspore/monitor/module_spec_verifier.py
+msprobe/mindspore/monitor/utils.py
+msprobe/mindspore/monitor/distributed/__init__.py
+msprobe/mindspore/monitor/distributed/distributed_ops.yaml
+msprobe/mindspore/monitor/distributed/stack_blacklist.yaml
+msprobe/mindspore/monitor/distributed/wrap_distributed.py
 msprobe/mindspore/overflow_check/__init__.py
 msprobe/mindspore/overflow_check/kernel_graph_overflow_check.py
 msprobe/mindspore/overflow_check/overflow_check_tool_factory.py
 msprobe/pytorch/__init__.py
 msprobe/pytorch/function_factory.py
-msprobe/pytorch/parse.py
 msprobe/pytorch/pt_config.py
 msprobe/pytorch/service.py
 msprobe/pytorch/api_accuracy_checker/.keep
@@ -278,16 +297,21 @@ msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/server.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.py
 msprobe/pytorch/bench_functions/apply_adam_w.py
 msprobe/pytorch/bench_functions/confusion_transpose.py
 msprobe/pytorch/bench_functions/fast_gelu.py
+msprobe/pytorch/bench_functions/group_norm_silu.py
 msprobe/pytorch/bench_functions/layer_norm_eval.py
 msprobe/pytorch/bench_functions/linear.py
 msprobe/pytorch/bench_functions/matmul_backward.py
+msprobe/pytorch/bench_functions/mish.py
+msprobe/pytorch/bench_functions/moe_gating_top_k_softmax.py
 msprobe/pytorch/bench_functions/npu_fusion_attention.py
 msprobe/pytorch/bench_functions/rms_norm.py
 msprobe/pytorch/bench_functions/rotary_mul.py
 msprobe/pytorch/bench_functions/scaled_mask_softmax.py
+msprobe/pytorch/bench_functions/sort_v2.py
 msprobe/pytorch/bench_functions/swiglu.py
 msprobe/pytorch/common/__init__.py
 msprobe/pytorch/common/compare_script.template
@@ -336,18 +360,12 @@ msprobe/pytorch/grad_probe/__init__.py
 msprobe/pytorch/grad_probe/grad_monitor.py
 msprobe/pytorch/grad_probe/grad_stat_csv.py
 msprobe/pytorch/hook_module/__init__.py
-msprobe/pytorch/hook_module/api_registry.py
+msprobe/pytorch/hook_module/api_register.py
 msprobe/pytorch/hook_module/hook_module.py
 msprobe/pytorch/hook_module/register_optimizer_hook.py
 msprobe/pytorch/hook_module/support_wrap_ops.yaml
 msprobe/pytorch/hook_module/utils.py
 msprobe/pytorch/hook_module/wrap_aten.py
-msprobe/pytorch/hook_module/wrap_distributed.py
-msprobe/pytorch/hook_module/wrap_functional.py
-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

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/mindstudio_probe.egg-info/requires.txt

@@ -1,11 +1,11 @@
 wheel
 einops
-numpy<2.0
+numpy<2.0,>=1.23.0
 pandas<2.1,>=1.3.5
 pyyaml
 rich
 tqdm
-openpyxl
+openpyxl>=3.0.6
 pyopenssl
 twisted
 matplotlib

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/msprobe/README.md

@@ -15,7 +15,7 @@ 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.stop() # 一般在训练循环末尾结束工具。必须调用，否则可能导致精度数据落盘不全
 debugger.step() # 在训练循环的最后需要重置工具，非循环场景不需要
 ```
 
@@ -44,6 +44,7 @@ export MSPROBE_LOG_LEVEL={x}
 
 - 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MSAdapter 2.1.0。
 - 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。
 
 
@@ -69,35 +70,37 @@ export MSPROBE_LOG_LEVEL={x}
 
 ### 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。
+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 "statistics" 或 "tensor" task。
 
 [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)
 
+[MSAdapter 场景的数据采集](./docs/29.data_dump_MSAdapter.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。
+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 "run_ut" task。
 
 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
+### 3 分级可视化构图比对
 
-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。
+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 场景的精度比对](./docs/10.accuracy_compare_PyTorch.md)
+[PyTorch 场景的分级可视化构图比对](./docs/21.visualization_PyTorch.md)
 
-[MindSpore 场景的精度比对](./docs/11.accuracy_compare_MindSpore.md)
+[MindSpore 场景的分级可视化构图比对](./docs/22.visualization_MindSpore.md)
 
-### 4 溢出检测与解析
+### 4 精度比对
 
-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。
+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 场景的溢出检测与解析](./docs/12.overflow_check_PyTorch.md)
+[PyTorch 场景的精度比对](./docs/10.accuracy_compare_PyTorch.md)
 
-[MindSpore 场景的溢出检测与解析](./docs/13.overflow_check_MindSpore.md)
+[MindSpore 场景的精度比对](./docs/11.accuracy_compare_MindSpore.md)
 
 ### 5 数据解析
 
@@ -127,28 +130,30 @@ MindSpore 动态图场景的[离线预检](./docs/09.accuracy_checker_MindSpore.
 
 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)
+[兼容 PyTorch 和 MindSpore 框架的训练状态监控](./docs/19.monitor.md)
 
-### 10 分级可视化构图比对
+### 10 单算子API自动生成脚本
 
-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。
+该功能将msprobe工具dump的精度数据进行解析，自动生成单API脚本，用于复现整网中出现的算子问题，降低用户复现问题的成本，供开发分析算子问题。
 
-[PyTorch 场景的分级可视化构图比对](./docs/21.visualization_PyTorch.md)
+[PyTorch 单算子API自动生成脚本](./docs/23.generate_operator_PyTorch.md)
 
-[MindSpore 场景的分级可视化构图比对](./docs/22.visualization_MindSpore.md)
+### 11 数码关联
 
+该功能只支持 MindSpore 静态图场景，用于将IR图与dump数据进行关联，获取dump数据和代码调用栈的关联关系。
 
-### 11 单算子API自动生成脚本
+[MindSpore 场景的数码关联](./docs/24.code_mapping_Mindspore.md)
 
-该功能将msprobe工具dump的精度数据进行解析，自动生成单API脚本，用于复现整网中出现的算子问题，降低用户复现问题的成本，供开发分析算子问题。
+### 12 溢出检测与解析
 
-[PyTorch 单算子API自动生成脚本](./docs/23.generate_operator_PyTorch.md)
+溢出检测用于采集溢出 API 或 模块的精度数据，而溢出解析则是通过对溢出数据的分析，进一步判断是否为正常溢出。对应 config.json 中的 "overflow_check" task。 
+推荐直接使用[数据采集](#1-数据采集)功能采集统计量信息,检测溢出问题。
 
-### 12 数码关联
+[PyTorch 场景的溢出检测与解析](./docs/12.overflow_check_PyTorch.md)
 
-该功能只支持 MindSpore 静态图场景，用于将IR图与dump数据进行关联，获取dump数据和代码调用栈的关联关系。
+[MindSpore 场景的溢出检测](./docs/13.overflow_check_MindSpore.md)
 
-[MindSpore 场景的数码关联](./docs/24.code_mapping_Mindspore.md)
+[MSAdapter 场景的溢出检测](./docs/30.overflow_check_MSAdapter.md)
 
 ## 📑 补充材料
 

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/msprobe/core/common/const.py

@@ -51,7 +51,10 @@ class Const:
     FOUR_SEGMENT = 4
     SIX_SEGMENT = 6
     SEVEN_SEGMENT = 7
+
     MAX_DEPTH = 10
+    CPU_QUARTER = 4
+    DUMP_MAX_DEPTH = 50
 
     # dump mode
     ALL = "all"
@@ -103,14 +106,16 @@ class Const:
     FREE_BENCHMARK = "free_benchmark"
     RUN_UT = "run_ut"
     GRAD_PROBE = "grad_probe"
-    TASK_LIST = [TENSOR, STATISTICS, OVERFLOW_CHECK, FREE_BENCHMARK, RUN_UT, GRAD_PROBE]
-    DUMP_DATA_COLLECTION_LIST = [STATISTICS, TENSOR]
+    STRUCTURE = "structure"
+    TASK_LIST = [TENSOR, STATISTICS, OVERFLOW_CHECK, FREE_BENCHMARK, RUN_UT, GRAD_PROBE, STRUCTURE]
+    DUMP_DATA_COLLECTION_LIST = [STATISTICS, TENSOR, STRUCTURE]
     DUMP_DATA_MODE_LIST = [ALL, INPUT, OUTPUT, FORWARD, BACKWARD]
     LEVEL_L0 = "L0"
     LEVEL_L1 = "L1"
     LEVEL_L2 = "L2"
     LEVEL_MIX = "mix"
-    LEVEL_LIST = [LEVEL_L0, LEVEL_L1, LEVEL_L2, LEVEL_MIX]
+    LEVEL_DEBUG = "debug"
+    LEVEL_LIST = [LEVEL_L0, LEVEL_L1, LEVEL_L2, LEVEL_MIX, LEVEL_DEBUG]
     ATTR_NAME_PREFIX = "wrap_"
     ATTR_NAME_PREFIX_LEN = len(ATTR_NAME_PREFIX)
     KERNEL_DUMP = "kernel_dump"
@@ -228,6 +233,92 @@ class Const:
 
     TENSOR_STAT_LEN = 2
 
+    SUPPORT_API_FILE_NAME = "support_wrap_ops.yaml"
+
+    PT_API_TYPE_FUNCTIONAL = "functional"
+    PT_API_TYPE_TENSOR = "tensor"
+    PT_API_TYPE_TORCH = "torch"
+    PT_API_TYPE_VF = "_VF"
+    PT_API_TYPE_NPU = "torch_npu"
+    PT_API_TYPE_ATEN = "aten"
+    PT_API_TYPE_DIST = "distributed"
+    PT_API_TYPE_NPU_DIST = "npu_distributed"
+
+    MS_API_TYPE_OPS = "ops"
+    MS_API_TYPE_TENSOR = "tensor"
+    MS_API_TYPE_STUB_TENSOR = "stubtensor"
+    MS_API_TYPE_MINT = "mint.ops"
+    MS_API_TYPE_MINT_FUNC = "mint.nn.functional"
+    MS_API_TYPE_COM = "communication.comm_func"
+
+    FUNCTIONAL_API_TYPE_PREFIX = "Functional"
+    TENSOR_API_TYPE_PREFIX = "Tensor"
+    DIST_API_TYPE_PREFIX = "Distributed"
+
+    TORCH_API_TYPE_PREFIX = "Torch"
+    NPU_API_TYPE_PREFIX = "NPU"
+    ATEN_API_TYPE_PREFIX = "Aten"
+    VF_API_TYPE_PREFIX = "VF"
+
+    MINT_API_TYPE_PREFIX = "Mint"
+    MINT_FUNC_API_TYPE_PREFIX = "MintFunctional"
+
+    SUPPORT_API_DICT_KEY_MAP = {
+        PT_FRAMEWORK: {
+            PT_API_TYPE_FUNCTIONAL: PT_API_TYPE_FUNCTIONAL,
+            PT_API_TYPE_TENSOR: PT_API_TYPE_TENSOR,
+            PT_API_TYPE_TORCH: PT_API_TYPE_TORCH,
+            PT_API_TYPE_VF: PT_API_TYPE_VF,
+            PT_API_TYPE_NPU: PT_API_TYPE_NPU,
+            PT_API_TYPE_ATEN: PT_API_TYPE_ATEN,
+            PT_API_TYPE_DIST: PT_API_TYPE_DIST,
+            PT_API_TYPE_NPU_DIST: PT_API_TYPE_NPU_DIST
+        },
+        MS_FRAMEWORK: {
+            MS_API_TYPE_OPS: MS_API_TYPE_OPS,
+            MS_API_TYPE_TENSOR: MS_API_TYPE_TENSOR,
+            MS_API_TYPE_STUB_TENSOR: MS_API_TYPE_TENSOR,
+            MS_API_TYPE_MINT: MS_API_TYPE_MINT,
+            MS_API_TYPE_MINT_FUNC: MS_API_TYPE_MINT_FUNC,
+            MS_API_TYPE_COM: MS_API_TYPE_COM
+        },
+        MT_FRAMEWORK: {
+            PT_API_TYPE_FUNCTIONAL: PT_API_TYPE_FUNCTIONAL,
+            PT_API_TYPE_TENSOR: PT_API_TYPE_TENSOR,
+            PT_API_TYPE_TORCH: PT_API_TYPE_TORCH,
+            PT_API_TYPE_NPU: PT_API_TYPE_NPU,
+            PT_API_TYPE_DIST: PT_API_TYPE_DIST
+        }
+    }
+
+    API_DATA_PREFIX = {
+        PT_FRAMEWORK: {
+            PT_API_TYPE_FUNCTIONAL: FUNCTIONAL_API_TYPE_PREFIX,
+            PT_API_TYPE_TENSOR: TENSOR_API_TYPE_PREFIX,
+            PT_API_TYPE_TORCH: TORCH_API_TYPE_PREFIX,
+            PT_API_TYPE_VF: VF_API_TYPE_PREFIX,
+            PT_API_TYPE_NPU: NPU_API_TYPE_PREFIX,
+            PT_API_TYPE_ATEN: ATEN_API_TYPE_PREFIX,
+            PT_API_TYPE_DIST: DIST_API_TYPE_PREFIX,
+            PT_API_TYPE_NPU_DIST: DIST_API_TYPE_PREFIX
+        },
+        MS_FRAMEWORK: {
+            MS_API_TYPE_OPS: FUNCTIONAL_API_TYPE_PREFIX,
+            MS_API_TYPE_TENSOR: TENSOR_API_TYPE_PREFIX,
+            MS_API_TYPE_STUB_TENSOR: TENSOR_API_TYPE_PREFIX,
+            MS_API_TYPE_MINT: MINT_API_TYPE_PREFIX,
+            MS_API_TYPE_MINT_FUNC: MINT_FUNC_API_TYPE_PREFIX,
+            MS_API_TYPE_COM: DIST_API_TYPE_PREFIX
+        },
+        MT_FRAMEWORK: {
+            PT_API_TYPE_FUNCTIONAL: FUNCTIONAL_API_TYPE_PREFIX,
+            PT_API_TYPE_TENSOR: TENSOR_API_TYPE_PREFIX,
+            PT_API_TYPE_TORCH: TORCH_API_TYPE_PREFIX,
+            PT_API_TYPE_NPU: NPU_API_TYPE_PREFIX,
+            PT_API_TYPE_DIST: DIST_API_TYPE_PREFIX
+        }
+    }
+
 
 class CompareConst:
     """
@@ -254,6 +345,7 @@ class CompareConst:
     MEAN_DIFF = "Mean diff"
     NORM_DIFF = "L2norm diff"
     COSINE = "Cosine"
+    EUC_DIST = "EucDist"
     MAX_ABS_ERR = "MaxAbsErr"
     MAX_RELATIVE_ERR = "MaxRelativeErr"
     MIN_RELATIVE_ERR = "MinRelativeErr"
@@ -328,8 +420,8 @@ class CompareConst:
     ULP_ERR_STATUS = "ulp_err_status"
 
     COMPARE_RESULT_HEADER = [
-        NPU_NAME, BENCH_NAME, NPU_DTYPE, BENCH_DTYPE, NPU_SHAPE, BENCH_SHAPE, COSINE, MAX_ABS_ERR, MAX_RELATIVE_ERR,
-        ONE_THOUSANDTH_ERR_RATIO, FIVE_THOUSANDTHS_ERR_RATIO,
+        NPU_NAME, BENCH_NAME, NPU_DTYPE, BENCH_DTYPE, NPU_SHAPE, BENCH_SHAPE, COSINE, EUC_DIST,
+        MAX_ABS_ERR, MAX_RELATIVE_ERR, ONE_THOUSANDTH_ERR_RATIO, FIVE_THOUSANDTHS_ERR_RATIO,
         NPU_MAX, NPU_MIN, NPU_MEAN, NPU_NORM, BENCH_MAX, BENCH_MIN, BENCH_MEAN, BENCH_NORM, ACCURACY, ERROR_MESSAGE
     ]
 
@@ -355,18 +447,16 @@ class CompareConst:
         Const.MD5: MD5_COMPARE_RESULT_HEADER
     }
 
-    ALL_COMPARE_INDEX = [COSINE, MAX_ABS_ERR, MAX_RELATIVE_ERR, ONE_THOUSANDTH_ERR_RATIO, FIVE_THOUSANDTHS_ERR_RATIO]
+    ALL_COMPARE_INDEX = [COSINE, EUC_DIST, MAX_ABS_ERR, MAX_RELATIVE_ERR, ONE_THOUSANDTH_ERR_RATIO,
+                         FIVE_THOUSANDTHS_ERR_RATIO]
     SUMMARY_COMPARE_INDEX = [MAX_DIFF, MIN_DIFF, MEAN_DIFF, NORM_DIFF,
                              MAX_RELATIVE_ERR, MIN_RELATIVE_ERR, MEAN_RELATIVE_ERR, NORM_RELATIVE_ERR]
 
     # dtype match
-    MS_TYPE = [
-        [Const.FLOAT16, Const.FLOAT32], [Const.FLOAT32, Const.FLOAT16],
-        [Const.FLOAT16, Const.BFLOAT16], [Const.BFLOAT16, Const.FLOAT16]
-    ]
-    TORCH_TYPE = [
-        [Const.TORCH_FLOAT16, Const.TORCH_FLOAT32], [Const.TORCH_FLOAT32, Const.TORCH_FLOAT16],
-        [Const.TORCH_FLOAT16, Const.TORCH_BFLOAT16], [Const.TORCH_BFLOAT16, Const.TORCH_FLOAT16]
+
+    DTYPE_MATCH_GROUPS = [
+        {Const.FLOAT16, Const.FLOAT32, Const.BFLOAT16},
+        {Const.TORCH_FLOAT16, Const.TORCH_FLOAT32, Const.TORCH_BFLOAT16}
     ]
 
     # read_op
@@ -465,7 +555,7 @@ class CompareConst:
         BENCH_MEAN: None, BENCH_NORM: None, ACCURACY: '', ERROR_MESSAGE: ''
     }
     MS_GRAPH_NPY = {
-        COSINE: None, MAX_ABS_ERR: None, MAX_RELATIVE_ERR: None, ONE_THOUSANDTH_ERR_RATIO: None,
+        COSINE: None, EUC_DIST: None, MAX_ABS_ERR: None, MAX_RELATIVE_ERR: None, ONE_THOUSANDTH_ERR_RATIO: None,
         FIVE_THOUSANDTHS_ERR_RATIO: None
     }
     MS_GRAPH_STATISTIC = {
@@ -536,46 +626,6 @@ class OverflowConst:
     OVERFLOW_DEBUG_MODE = 1
 
 
-class MsCompareConst:
-    # api_info field
-    MINT = "Mint"
-    MINT_FUNCTIONAL = "MintFunctional"
-    TENSOR_API = "Tensor"
-
-    API_NAME_STR_LENGTH = 4
-
-    TASK_FIELD = "task"
-    STATISTICS_TASK = "statistics"
-    TENSOR_TASK = "tensor"
-    DUMP_DATA_DIR_FIELD = "dump_data_dir"
-    DATA_FIELD = "data"
-
-    # supported api yaml
-    SUPPORTED_API_LIST_FILE = "checker_support_api.yaml"
-    SUPPORTED_TENSOR_LIST_KEY = "tensor"
-
-    # detail_csv
-    DETAIL_CSV_API_NAME = "API Name"
-    DETAIL_CSV_BENCH_DTYPE = "Bench Dtype"
-    DETAIL_CSV_TESTED_DTYPE = "Tested Dtype"
-    DETAIL_CSV_SHAPE = "Shape"
-    DETAIL_CSV_PASS_STATUS = "Status"
-    DETAIL_CSV_MESSAGE = "Message"
-    DETAIL_CSV_FILE_NAME = "accuracy_checking_details"
-
-    # result_csv
-    RESULT_CSV_FORWARD_TEST_SUCCESS = "Forward Test Success"
-    RESULT_CSV_BACKWARD_TEST_SUCCESS = "Backward Test Success"
-    RESULT_CSV_FILE_NAME = "accuracy_checking_result"
-
-    EPSILON = 1e-8
-
-    class ProcessStatus:
-        SUCCESS = "success"
-        API_NOT_FOUND = "api_not_found"
-        EXCEPTION_SKIP = "exception_skip"
-
-
 class MsgConst:
     """
     Class for log messages const
@@ -612,6 +662,15 @@ class MonitorConst:
     """
     Class for monitor const
     """
+
+    # monitor config set default values
+    DEFAULT_GRAD_ACC_STEPS = 1
+    DEFAULT_START_ITERATION = 0
+    DEFAULT_START_STEP = 0
+    DEFAULT_MAX_COLLECT_TIMES = 1e8
+    DEFAULT_MIN_COLLECT_TIMES = 0
+    DEFAULT_STEP_INTERVAL = 1
+
     OP_LIST = ["norm", "min", "max", "zeros", "nans", "id", "mean"]
     MONITOR_OUTPUT_DIR = "MONITOR_OUTPUT_DIR"
     DEFAULT_MONITOR_OUTPUT_DIR = "./monitor_output"
@@ -623,29 +682,39 @@ class MonitorConst:
         "DeepSpeedZeroOptimizer_Stage1_or_2",
         "DeepSpeedZeroOptimizer_Stage3"
     )
+    DEEPSPEED_ZERO_OPT_FILTER = "DeepSpeedZeroOptimizer"
     RULE_NAME = ['AnomalyTurbulence']
 
     SLICE_SIZE = 20480
+    # used for name
     DOT = "."
-    VPP_SEP = ":"
+    NAME_SEP = ":"
+    INPUT_GRAD = "input_grad"
+    OUTPUT_GRAD = "output_grad"
     ACTV_IN = "input"
     ACTV_OUT = "output"
     ACTVGRAD_IN = "input_grad"
     ACTVGRAD_OUT = "output_grad"
+    # used for tasks
+    ACTV = "actv"
+    ACTVGRAD = "actv_grad"
     POST_GRAD = "post_grad"
     PRE_GRAD = "pre_grad"
     ACC_GRAD = "acc_grad"
     PREFIX_POST = "post"
     PREFIX_PRE = "pre"
-    OUTPUT_DIR_PATTERN = r"([\w-]{0,20})-rank(\d{1,5})-"
-
     EXP_AVG = "exp_avg"
-    EFXP_AVG_SQ = "efxp_avg_sq"
+    EXP_AVG_SQ = "exp_avg_sq"
+    PARAM = "param"
 
+    CSV_HEADER = ["vpp_stage", "name", "step"]
+    CSV_HEADER_XY = ["vpp_stage", "name", "step", "micro_step"]
+    OUTPUT_DIR_PATTERN = r"([\w-]{0,20})-rank(\d{1,5})-"
     ANOMALY_JSON = "anomaly.json"
     ANALYSE_JSON = "anomaly_analyse.json"
     TENSORBOARD = "tensorboard"
     CSV = "csv"
     API = "api"
-    OPS_START_INDEX = 3
-    HEADER_NAME_INDEX = 1
+    HEADER_NAME = 'name'
+
+    MAX_NDIGITS = 20

mindstudio_probe-1.3.0/msprobe/core/common/decorator.py

@@ -0,0 +1,50 @@
+# Copyright (c) 2024-2025, 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 collections import defaultdict
+from functools import wraps
+
+from msprobe.core.common.const import Const
+from msprobe.core.common.exceptions import MsprobeException
+from msprobe.core.common.log import logger
+
+# 记录工具函数递归的深度
+recursion_depth = defaultdict(int)
+
+
+def recursion_depth_decorator(func_info, max_depth=Const.MAX_DEPTH):
+    """装饰一个函数，当函数递归调用超过限制时，抛出异常并打印函数信息。"""
+    def decorator(func):
+        @wraps(func)
+        def wrapper(*args, **kwargs):
+            func_id = id(func)
+            recursion_depth[func_id] += 1
+            if recursion_depth[func_id] > max_depth:
+                msg = f"call {func_info} exceeds the recursion limit."
+                logger.error_log_with_exp(
+                    msg,
+                    MsprobeException(
+                        MsprobeException.RECURSION_LIMIT_ERROR, msg
+                    ),
+                )
+            try:
+                result = func(*args, **kwargs)
+            finally:
+                recursion_depth[func_id] -= 1
+            return result
+
+        return wrapper
+
+    return decorator

{mindstudio_probe-1.2.1 → mindstudio_probe-1.3.0}/msprobe/core/common/exceptions.py

@@ -28,12 +28,14 @@ class MsprobeException(CodedException):
     OVERFLOW_NUMS_ERROR = 1
     RECURSION_LIMIT_ERROR = 2
     INTERFACE_USAGE_ERROR = 3
+    UNSUPPORTED_TYPE_ERROR = 4
 
     err_strs = {
         INVALID_PARAM_ERROR: "[msprobe] 无效参数：",
         OVERFLOW_NUMS_ERROR: "[msprobe] 超过预设溢出次数 当前溢出次数：",
         RECURSION_LIMIT_ERROR: "[msprobe] 递归调用超过限制：",
-        INTERFACE_USAGE_ERROR: "[msprobe] Invalid interface usage: "
+        INTERFACE_USAGE_ERROR: "[msprobe] Invalid interface usage: ",
+        UNSUPPORTED_TYPE_ERROR: "[msprobe] Unsupported type: "
     }