mindstudio-probe 1.1.0__py3-none-any.whl → 1.2.1__py3-none-any.whl

msprobe/core/advisor/advisor.py

@@ -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)
@@ -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()

msprobe/core/common/const.py

@@ -1,3 +1,18 @@
+# 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.
+
 import os
 import stat
 
@@ -10,9 +25,11 @@ class Const:
     """
     TOOL_NAME = "msprobe"
 
+    ipv4_pattern = "([1-9]?\d|1\d{2}|2[0-4]\d|25[0-5])(\.([1-9]?\d|1\d{2}|2[0-4]\d|25[0-5])){3}$"
     SEP = "."
     REGEX_PREFIX_MAX_LENGTH = 20
     REGEX_PREFIX_PATTERN = r"^[a-zA-Z0-9_-]+$"
+    REGEX_FORWARD_BACKWARD = r'\.(forward|backward)\.'
     FILE_PATTERN = r'^[a-zA-Z0-9_./-]+$'
     STRING_BLACKLIST = r"^[＋－＝％＠\+\-=%@]|;[＋－＝％＠\+\-=%@]"
     COMMA = ","
@@ -20,6 +37,8 @@ class Const:
     OFF = 'OFF'
     BACKWARD = 'backward'
     FORWARD = 'forward'
+    PROGRESS_TIMEOUT = 3000
+    EXCEPTION_NONE = None
     JIT = 'Jit'
     PRIMITIVE_PREFIX = 'Primitive'
     DEFAULT_LIST = []
@@ -47,6 +66,7 @@ class Const:
     ONLINE_DUMP_MODE = [ALL, LIST, AUTO, OFF]
     SUMMARY = "summary"
     MD5 = "md5"
+    VALUE = "value"
     SUMMARY_MODE = [ALL, SUMMARY, MD5]
 
     WRITE_FLAGS = os.O_WRONLY | os.O_CREAT
@@ -55,6 +75,7 @@ class Const:
 
     PKL_SUFFIX = ".pkl"
     NUMPY_SUFFIX = ".npy"
+    NUMPY_PATTERN = "*.npy"
     PT_SUFFIX = ".pt"
     ONE_GB = 1073741824  # 1 * 1024 * 1024 * 1024
     TEN_GB = 10737418240  # 10 * 1024 * 1024 * 1024
@@ -69,6 +90,8 @@ class Const:
     INPUT_KWARGS = 'input_kwargs'
     GRAD_INPUT = 'grad_input'
     GRAD_OUTPUT = 'grad_output'
+    PARAMS = 'parameters'
+    PARAMS_GRAD = 'parameters_grad'
     START = "start"
     STOP = "stop"
     ENV_ENABLE = "1"
@@ -82,6 +105,7 @@ class Const:
     GRAD_PROBE = "grad_probe"
     TASK_LIST = [TENSOR, STATISTICS, OVERFLOW_CHECK, FREE_BENCHMARK, RUN_UT, GRAD_PROBE]
     DUMP_DATA_COLLECTION_LIST = [STATISTICS, TENSOR]
+    DUMP_DATA_MODE_LIST = [ALL, INPUT, OUTPUT, FORWARD, BACKWARD]
     LEVEL_L0 = "L0"
     LEVEL_L1 = "L1"
     LEVEL_L2 = "L2"
@@ -93,6 +117,8 @@ class Const:
     DATA = "data"
     PT_FRAMEWORK = "pytorch"
     MS_FRAMEWORK = "mindspore"
+    MT_FRAMEWORK = "mindtorch"
+    UNKNOWN_FRAMEWORK = "unknown"
     DIRECTORY_LENGTH = 4096
     FILE_NAME_LENGTH = 255
     FLOAT_TYPE = [np.half, np.single, float, np.double, np.float64, np.longdouble, np.float32, np.float16]
@@ -102,7 +128,12 @@ class Const:
     NPU_LOWERCASE = 'npu'
     CPU_LOWERCASE = 'cpu'
     CUDA_LOWERCASE = 'cuda'
+    DEVICE = 'device'
     DISTRIBUTED = 'Distributed'
+    DUMP_PREFIX = ["Distributed", "Functional", "Torch", "Tensor", "Mint", "MintFunctional", "Primitive",
+                   "Aten", "VF", "NPU", "Jit"]
+    MODULE_PREFIX = ["Module", "Cell"]
+    FORWARD_NAME_SUFFIX = ".forward"
 
     # struct json param
     ORIGIN_DATA = "origin_data"
@@ -113,21 +144,28 @@ class Const:
     MODULE_WHITE_LIST = ["torch", "numpy"]
 
     FUNC_SKIP_LIST = ["construct", "__call__"]
-
-    FILE_SKIP_LIST = ["site-packages/mindspore", "package/mindspore", "msprobe", "site-packages/torch", "package/torch"]
+    FILE_SKIP_LIST = ["msprobe", "MindSpeed"]
+    DATA_TYPE_SKIP_LIST = ["Primitive", "Jit"]
 
     STACK_FILE_INDEX = 0
-
     STACK_FUNC_INDEX = 2
-
     STACK_FUNC_ELE_INDEX = 1
 
-    CONSTRUCT_NAME_INDEX = -3
-
-    NAME_FIRST_POSSIBLE_INDEX = -4
-
-    NAME_SECOND_POSSIBLE_INDEX = -5
-
+    SCOPE_ID_INDEX = -1
+    SCOPE_DIRECTION_INDEX = -2
+    TYPE_NAME_INDEX = -3
+    PARAMS_GRAD_TYPE_NAME_INDEX = -2
+    LAYER_NAME_INDEX = -4
+    PARAMS_GRAD_NAME_INDEX = -3
+    API_TYPE_INDEX = 0
+    LEFT_MOVE_INDEX = -1
+    RIGHT_MOVE_INDEX = 1
+    LAST_INDEX = -1
+
+    TOP_LAYER = "TopLayer"
+    CELL = "Cell"
+    MODULE = "Module"
+    FRAME_FILE_LIST = ["site-packages/torch", "package/torch", "site-packages/mindspore", "package/mindspore"]
     INPLACE_LIST = [
         "broadcast", "all_reduce", "reduce", "all_gather", "gather", "scatter", "reduce_scatter",
         "_reduce_scatter_base", "_all_gather_base", "send", "recv", "irecv", "isend", "all_to_all_single", "all_to_all",
@@ -136,22 +174,29 @@ class Const:
 
     CONVERT = {
         "int32_to_int64": ["torch.int32", "torch.int64"],
+        "int64_to_fp32": ["torch.int64", "torch.float32"]
     }
 
     CONVERT_API = {
-        "int32_to_int64": ["cross_entropy"]
+        "int32_to_int64": ["cross_entropy"],
+        "int64_to_fp32": ["histc"]
     }
 
+    FA_SPECIAL_SPARSE_MODE = [2, 3, 4]
+
     FILL_CHAR_NUMS = 50
     TOOL_ENDS_SUCCESSFULLY = f"{TOOL_NAME} ends successfully."
     WITHOUT_CALL_STACK = "The call stack retrieval failed."
-    
+
     STEP = "step"
     RANK = "rank"
     HYPHEN = "-"
-    STEP_RANK_MAXIMUM_RANGE = [int(0), int(1e6)]
+    STEP_RANK_MINIMUM_VALUE = 0
+    STEP_RANK_MAXIMUM_VALUE = int(1e6)
 
     # data type const
+    TORCH_INT_DTYPE = ["torch.int8", "torch.int32", "torch.int64"]
+    TORCH_FLOAT_DTYPE = ["torch.bfloat16", "torch.float16", "torch.float32", "torch.float64"]
     FLOAT16 = "Float16"
     FLOAT32 = "Float32"
     BFLOAT16 = "BFloat16"
@@ -159,6 +204,30 @@ class Const:
     TORCH_FLOAT32 = "torch.float32"
     TORCH_BFLOAT16 = "torch.bfloat16"
 
+    DTYPE = 'dtype'
+    SHAPE = 'shape'
+    MAX = 'Max'
+    MIN = 'Min'
+    MEAN = 'Mean'
+    NORM = 'Norm'
+
+    CODE_STACK = 'Code Stack'
+    OP_NAME = 'Op Name'
+    SCOPE_NAME = 'Scope Name'
+    CODE_STACKS = 'Code Stacks'
+    FILE_PATH = 'File Path'
+    NEW_LINE = '\n'
+    CSV_NEWLINE_SEPARATOR = ',\n'
+    # 分隔符常量
+    SCOPE_SEPARATOR = "/"
+    REPLACEMENT_CHARACTER = "_"
+
+    OPTIMIZER = "optimizer"
+    CLIP_GRAD = "clip_grad"
+    END_PREFIX = "end_"
+
+    TENSOR_STAT_LEN = 2
+
 
 class CompareConst:
     """
@@ -201,10 +270,62 @@ class CompareConst:
     RESULT = "Result"
     MAGNITUDE = 0.5
     OP_NAME = "op_name"
+    STRUCT = "struct"
     INPUT_STRUCT = "input_struct"
+    KWARGS_STRUCT = "kwargs_struct"
     OUTPUT_STRUCT = "output_struct"
+    PARAMS_STRUCT = "params_struct"
+    PARAMS_GRAD_STRUCT = "params_grad_struct"
     SUMMARY = "summary"
+    COMPARE_RESULT = "compare_result"
+    COMPARE_MESSAGE = "compare_message"
     MAX_EXCEL_LENGTH = 1048576
+    YES = "Yes"
+    NO = "No"
+    STATISTICS_INDICATOR_NUM = 4
+    EPSILON = 1e-10
+    COMPARE_ENDS_SUCCESSFULLY = "msprobe compare ends successfully."
+    DEFAULT_RATIO_VALUE = 10000
+    THOUSANDTH_PASS_VALUE = 0.999
+    ZERO_SHAPE = '(0,)'
+
+    BENCHMARK_COMPARE_ALGORITHM_NAME = "标杆比对法"
+    ULP_COMPARE_ALGORITHM_NAME = "ULP误差比对法"
+    BINARY_CONSISTENCY_ALGORITHM_NAME = "二进制一致法"
+    ABSOLUTE_THRESHOLD_ALGORITHM_NAME = "绝对阈值法"
+    THOUSANDTH_STANDARD_ALGORITHM_NAME = "双千指标法"
+    ACCUMULATIVE_ERROR_COMPARE_ALGORITHM_NAME = "累积误差比对法"
+
+    ABSOLUTE_THRESHOLD = 'absolute_threshold'
+    BINARY_CONSISTENCY = 'binary_consistency'
+    ULP_COMPARE = 'ulp_compare'
+    THOUSANDTH_STANDARD = 'thousandth_threshold'
+    BENCHMARK = 'benchmark'
+    ACCUMULATIVE_ERROR_COMPARE = 'accumulative_error_compare'
+
+    SMALL_VALUE_ERR_RATIO = "small_value_err_ratio"
+    RMSE_RATIO = "rmse_ratio"
+    MAX_REL_ERR_RATIO = "max_rel_err_ratio"
+    MEAN_REL_ERR_RATIO = "mean_rel_err_ratio"
+    EB_RATIO = "eb_ratio"
+
+    SMALL_VALUE = "small_value"
+    RMSE = "rmse"
+    MAX_REL_ERR = "max_rel_err"
+    MEAN_REL_ERR = "mean_rel_err"
+    EB = "eb"
+
+    SMALL_VALUE_ERR_STATUS = "small_value_err_status"
+    RMSE_STATUS = "rmse_status"
+    MAX_REL_ERR_STATUS = "max_rel_err_status"
+    MEAN_REL_ERR_STATUS = "mean_rel_err_status"
+    EB_STATUS = "eb_status"
+
+    MEAN_ULP_ERR = "mean_ulp_err"
+    ULP_ERR_PROPORTION = "ulp_err_proportion"
+    ULP_ERR_PROPORTION_RATIO = "ulp_err_proportion_ratio"
+
+    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,
@@ -222,6 +343,57 @@ class CompareConst:
         NPU_NAME, BENCH_NAME, NPU_DTYPE, BENCH_DTYPE, NPU_SHAPE, BENCH_SHAPE, NPU_MD5, BENCH_MD5, RESULT
     ]
 
+    COMPARE_RESULT_HEADER_STACK = COMPARE_RESULT_HEADER + [STACK]
+
+    SUMMARY_COMPARE_RESULT_HEADER_STACK = SUMMARY_COMPARE_RESULT_HEADER + [STACK]
+
+    MD5_COMPARE_RESULT_HEADER_STACK = MD5_COMPARE_RESULT_HEADER + [STACK]
+
+    HEAD_OF_COMPARE_MODE = {
+        Const.ALL: COMPARE_RESULT_HEADER,
+        Const.SUMMARY: SUMMARY_COMPARE_RESULT_HEADER,
+        Const.MD5: MD5_COMPARE_RESULT_HEADER
+    }
+
+    ALL_COMPARE_INDEX = [COSINE, 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]
+    ]
+
+    # read_op
+    IO_NAME_MAPPING = {
+        Const.INPUT_ARGS: '.input',
+        Const.INPUT_KWARGS: '.input',
+        Const.INPUT: '.input',
+        Const.OUTPUT: '.output',
+        Const.PARAMS: '.parameters'
+    }
+
+    # state to struct mapping
+    STATE_TO_STRUCT_MAPPING = {
+        Const.INPUT: INPUT_STRUCT,
+        Const.KWARGS: INPUT_STRUCT,
+        Const.OUTPUT: OUTPUT_STRUCT,
+        Const.PARAMS: PARAMS_STRUCT,
+        Const.PARAMS_GRAD: PARAMS_GRAD_STRUCT
+    }
+
+    STRUCT_COMPARE_KEY = [
+        INPUT_STRUCT,
+        OUTPUT_STRUCT,
+        PARAMS_STRUCT,
+        PARAMS_GRAD_STRUCT
+    ]
+
     # compare standard
     HUNDRED_RATIO_THRESHOLD = 0.01
     THOUSAND_RATIO_THRESHOLD = 0.001
@@ -241,6 +413,8 @@ class CompareConst:
     PASS = 'pass'
     WARNING = 'Warning'
     ERROR = 'error'
+    TRUE = 'TRUE'
+    FALSE = 'FALSE'
     SKIP = 'SKIP'
     N_A = 'N/A'
     INF = 'inf'
@@ -298,6 +472,15 @@ class CompareConst:
         MAX_DIFF: None, MIN_DIFF: None, MEAN_DIFF: None, NORM_DIFF: None, MAX_RELATIVE_ERR: None,
         MIN_RELATIVE_ERR: None, MEAN_RELATIVE_ERR: None, NORM_RELATIVE_ERR: None
     }
+    INPUT_PATTERN = Const.SEP + Const.INPUT + Const.SEP
+    KWARGS_PATTERN = Const.SEP + Const.KWARGS + Const.SEP
+    OUTPUT_PATTERN = Const.SEP + Const.OUTPUT + Const.SEP
+    PARAMS_PATTERN = Const.SEP + Const.PARAMS + Const.SEP
+    PARAMS_GRAD_PATTERN = Const.SEP + Const.PARAMS_GRAD + Const.SEP
+    COMPARE_KEY = 'compare_key'
+    COMPARE_SHAPE = 'compare_shape'
+    INTERNAL_API_MAPPING_FILE = 'ms_to_pt_api.yaml'
+    UNREADABLE = 'unreadable data'
 
 
 class FileCheckConst:
@@ -316,13 +499,17 @@ class FileCheckConst:
     JSON_SUFFIX = ".json"
     PT_SUFFIX = ".pt"
     CSV_SUFFIX = ".csv"
+    XLSX_SUFFIX = ".xlsx"
     YAML_SUFFIX = ".yaml"
+    IR_SUFFIX = ".ir"
     MAX_PKL_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     MAX_NUMPY_SIZE = 10737418240  # 10 * 1024 * 1024 * 1024
     MAX_JSON_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     MAX_PT_SIZE = 10737418240  # 10 * 1024 * 1024 * 1024
     MAX_CSV_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
-    MAX_YAML_SIZE = 1048576  # 1 * 1024 * 1024
+    MAX_XLSX_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
+    MAX_YAML_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
+    MAX_IR_SIZE = 1073741824  # 1 * 1024 * 1024 * 1024
     COMMOM_FILE_SIZE = 1048576  # 1 * 1024 * 1024
     DIR = "dir"
     FILE = "file"
@@ -334,7 +521,9 @@ class FileCheckConst:
         JSON_SUFFIX: MAX_JSON_SIZE,
         PT_SUFFIX: MAX_PT_SIZE,
         CSV_SUFFIX: MAX_CSV_SIZE,
-        YAML_SUFFIX: MAX_YAML_SIZE
+        XLSX_SUFFIX: MAX_XLSX_SIZE,
+        YAML_SUFFIX: MAX_YAML_SIZE,
+        IR_SUFFIX: MAX_IR_SIZE
     }
     CSV_BLACK_LIST = r'^[＋－＝％＠\+\-=%@]|;[＋－＝％＠\+\-=%@]'
 
@@ -351,6 +540,9 @@ class MsCompareConst:
     # api_info field
     MINT = "Mint"
     MINT_FUNCTIONAL = "MintFunctional"
+    TENSOR_API = "Tensor"
+
+    API_NAME_STR_LENGTH = 4
 
     TASK_FIELD = "task"
     STATISTICS_TASK = "statistics"
@@ -358,6 +550,10 @@ class MsCompareConst:
     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"
@@ -374,6 +570,11 @@ class MsCompareConst:
 
     EPSILON = 1e-8
 
+    class ProcessStatus:
+        SUCCESS = "success"
+        API_NOT_FOUND = "api_not_found"
+        EXCEPTION_SKIP = "exception_skip"
+
 
 class MsgConst:
     """
@@ -382,15 +583,20 @@ class MsgConst:
     MSPROBE_LOG_LEVEL = "MSPROBE_LOG_LEVEL"
     LOG_LEVEL_ENUM = ["0", "1", "2", "3", "4"]
     LOG_LEVEL = ["DEBUG", "INFO", "WARNING", "ERROR"]
+
     class LogLevel:
         class DEBUG:
             value = 0
+
         class INFO:
             value = 1
+
         class WARNING:
             value = 2
+
         class ERROR:
             value = 3
+
     SPECIAL_CHAR = ["\n", "\r", "\u007F", "\b", "\f", "\t", "\u000B", "%08", "%0a", "%0b", "%0c", "%0d", "%7f"]
 
     NOT_CREATED_INSTANCE = "PrecisionDebugger instance is not created."
@@ -400,3 +606,46 @@ class GraphMode:
     NPY_MODE = "NPY_MODE"
     STATISTIC_MODE = "STATISTIC_MODE"
     ERROR_MODE = "ERROR_MODE"
+
+
+class MonitorConst:
+    """
+    Class for monitor const
+    """
+    OP_LIST = ["norm", "min", "max", "zeros", "nans", "id", "mean"]
+    MONITOR_OUTPUT_DIR = "MONITOR_OUTPUT_DIR"
+    DEFAULT_MONITOR_OUTPUT_DIR = "./monitor_output"
+    DATABASE = "database"
+    EMAIL = "email"
+    OPT_TY = ['Megatron_DistributedOptimizer', 'Megatron_Float16OptimizerWithFloat16Params']
+    DEEPSPEED_OPT_TY = (
+        "DeepSpeedZeroOptimizer_Stage0",
+        "DeepSpeedZeroOptimizer_Stage1_or_2",
+        "DeepSpeedZeroOptimizer_Stage3"
+    )
+    RULE_NAME = ['AnomalyTurbulence']
+
+    SLICE_SIZE = 20480
+    DOT = "."
+    VPP_SEP = ":"
+    ACTV_IN = "input"
+    ACTV_OUT = "output"
+    ACTVGRAD_IN = "input_grad"
+    ACTVGRAD_OUT = "output_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"
+
+    ANOMALY_JSON = "anomaly.json"
+    ANALYSE_JSON = "anomaly_analyse.json"
+    TENSORBOARD = "tensorboard"
+    CSV = "csv"
+    API = "api"
+    OPS_START_INDEX = 3
+    HEADER_NAME_INDEX = 1

msprobe/core/common/exceptions.py

@@ -1,3 +1,18 @@
+# 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.
+
 class CodedException(Exception):
     def __init__(self, code, error_info=''):
         super().__init__()
@@ -11,10 +26,14 @@ class CodedException(Exception):
 class MsprobeException(CodedException):
     INVALID_PARAM_ERROR = 0
     OVERFLOW_NUMS_ERROR = 1
+    RECURSION_LIMIT_ERROR = 2
+    INTERFACE_USAGE_ERROR = 3
 
     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："
+        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: "
     }
 
 
@@ -41,7 +60,7 @@ class ParseJsonException(CodedException):
     InvalidDumpJson = 1
     err_strs = {
         UnexpectedNameStruct: "[msprobe] Unexpected name in json: ",
-        InvalidDumpJson: "[msprobe] json格式不正确: ",
+        InvalidDumpJson: "[msprobe] Invalid dump.json format: ",
     }
 
 
@@ -73,9 +92,13 @@ class StepException(CodedException):
 class FreeBenchmarkException(CodedException):
     UnsupportedType = 0
     InvalidGrad = 1
+    InvalidPerturbedOutput = 2
+    OutputIndexError = 3
     err_strs = {
         UnsupportedType: "[msprobe] Free benchmark get unsupported type: ",
         InvalidGrad: "[msprobe] Free benchmark gradient invalid: ",
+        InvalidPerturbedOutput: "[msprobe] Free benchmark invalid perturbed output: ",
+        OutputIndexError: "[msprobe] Free benchmark output index out of bounds: ",
     }
 
 
@@ -87,6 +110,7 @@ class DistributedNotInitializedError(Exception):
     def __str__(self):
         return self.msg
 
+
 class ApiAccuracyCheckerException(CodedException):
     ParseJsonFailed = 0
     UnsupportType = 1
@@ -97,4 +121,4 @@ class ApiAccuracyCheckerException(CodedException):
         UnsupportType: "[msprobe] Api Accuracy Checker get unsupported type: ",
         WrongValue: "[msprobe] Api Accuracy Checker get wrong value: ",
         ApiWrong: "[msprobe] Api Accuracy Checker something wrong with api: ",
-    }
+    }