mindstudio-probe 1.3.0__py3-none-any.whl → 8.1.1__py3-none-any.whl

msprobe/pytorch/hook_module/support_wrap_ops.yaml

@@ -676,6 +676,7 @@ torch:
   - _batch_norm_impl_index
   - _convolution
   - _foreach_norm
+  - _fused_adamw_
   - _softmax_backward_data
   - abs
   - abs_
@@ -1258,6 +1259,7 @@ torch_npu:
   - npu_scatter_nd_update_
   - npu_scatter_nd_update
   - npu_prefetch
+  - npu_dynamic_block_quant
 
 aten:
   - signbit
@@ -2009,4 +2011,23 @@ distributed:
 
 npu_distributed:
   - isend
-  - irecv
+  - irecv
+
+mindspeed:
+  - dropout_add_layer_norm.npu_dropout_add_layer_norm
+  - npu_rotary_position_embedding.npu_rotary_position_embedding
+  - fusion_attention_v2.npu_fusion_attention
+  - npu_mm_all_reduce_add_rms_norm.npu_mm_all_reduce_add_rms_norm
+  - npu_mm_all_reduce_add_rms_norm_.npu_mm_all_reduce_add_rms_norm_
+  - gmm.npu_gmm
+  - gmm.npu_gmm_v2
+  - npu_grouped_mat_mul_all_reduce.npu_grouped_mat_mul_all_reduce
+  - ffn.npu_ffn
+  - npu_moe_token_permute.npu_moe_token_permute
+  - npu_moe_token_unpermute.npu_moe_token_unpermute
+  - npu_ring_attention_update.npu_ring_attention_update
+  - npu_matmul_add.npu_matmul_add_fp32
+  - npu_groupmatmul_add.npu_groupmatmul_add_fp32
+  - quant_gmm.npu_quant_gmm
+  - quant_gmm.npu_quant_gmm_v2
+  - npu_apply_fused_ema_adamw.npu_apply_fused_ema_adamw

msprobe/pytorch/hook_module/utils.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024-2024, Huawei Technologies Co., Ltd.
+# Copyright (c) 2024-2025, Huawei Technologies Co., Ltd.
 # All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0  (the "License");
@@ -14,7 +14,11 @@
 # limitations under the License.
 
 import os
-from msprobe.core.common.file_utils import load_yaml
+import importlib
+import inspect
+
+from msprobe.core.common.file_utils import load_yaml, check_link
+from msprobe.core.common.log import logger
 
 
 def get_ops():
@@ -26,3 +30,25 @@ def get_ops():
     wrap_torch = ops.get('torch')
     wrap_npu_ops = ops.get('torch_npu')
     return set(wrap_functional) | set(wrap_tensor) | set(wrap_torch) | set(wrap_npu_ops)
+
+
+def dynamic_import_op(package, white_list):
+    package_name = package.__name__
+    ops = {}
+    ops_dir, _ = os.path.split(package.__file__)
+    check_link(ops_dir)
+    for file_name in os.listdir(ops_dir):
+        if file_name in white_list:
+            sub_module_name = file_name[:-3]
+            module_name = f"{package_name}.{sub_module_name}"
+            try:
+                module = importlib.import_module(module_name)
+            except Exception as e:
+                logger.warning(f"import {module_name} failed!")
+                continue
+
+            func_members = inspect.getmembers(module, inspect.isfunction)
+            for func_member in func_members:
+                func_name, func = func_member[0], func_member[1]
+                ops[f"{sub_module_name}.{func_name}"] = func
+    return ops

msprobe/pytorch/monitor/csv2tb.py

@@ -28,7 +28,10 @@ from msprobe.core.common.decorator import recursion_depth_decorator
 from msprobe.pytorch.common.log import logger
 from msprobe.pytorch.monitor.utils import get_target_output_dir
 
-all_data_type_list = ["actv", "actv_grad", "exp_avg", "exp_avg_sq", "grad_unreduced", "grad_reduced", "param"]
+all_data_type_list = [
+    "actv", "actv_grad", "exp_avg", "exp_avg_sq",
+    "grad_unreduced", "grad_reduced", "param_origin", "param_updated"
+]
 CSV_FILE_SUFFIX = r"_\d+-\d+\.csv"
 MAX_PROCESS_NUM = 128
 
@@ -48,7 +51,7 @@ def parse_step_line(line, ops):
 
 def parse_step_fn(filepath):
     data = read_csv(filepath)
-    ops = [k for k in data.keys() if k in MonitorConst.OP_LIST]
+    ops = [k for k in data.keys() if k in MonitorConst.OP_LIST[:-2]]
     parse_step_result = {}
 
     for _, line in data.iterrows():
@@ -76,6 +79,7 @@ def write_step(output_dirpath, parse_step_result, rank, data_type):
             for op, value in ops.items():
                 tag = f"{vpp_name}/{op}"
                 writer.add_scalar(tag, value, step)
+    writer.flush()
 
 
 @recursion_depth_decorator("update_dict", max_depth=50)

msprobe/pytorch/monitor/data_writers.py

@@ -0,0 +1,259 @@
+# 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 itertools
+import os
+from collections import defaultdict
+from dataclasses import dataclass
+
+import pandas as pd
+import torch
+from torch.utils.tensorboard import SummaryWriter
+
+from msprobe.core.common.const import FileCheckConst, MonitorConst
+from msprobe.core.common.file_utils import change_mode, create_directory, write_df_to_csv
+from msprobe.core.monitor.anomaly_processor import AnomalyDataFactory, AnomalyTurbulence, AnomalyScanner
+from msprobe.pytorch.common.log import logger
+
+
+class BCOLORS:
+    HEADER = '\033[95m'
+    OKBLUE = '\033[94m'
+    OKCYAN = '\033[96m'
+    OKGREEN = '\033[92m'
+    WARNING = '\033[93m'
+    FAIL = '\033[91m'
+    ENDC = '\033[0m'
+    BOLD = '\033[1m'
+    UNDERLINE = '\033[4m'
+
+
+@dataclass
+class WriterInput:
+    path: str
+    ad_rules: list
+    job_id: str
+    anomaly_factory: AnomalyDataFactory = None
+    ndigits: int = 6
+    step_count_per_record: int = 1
+
+
+class BaseWriterWithAD:
+    def __init__(self, writer_input: WriterInput):
+        self.tag2scalars = {}
+        self.ad_rules = writer_input.ad_rules
+        self.job_id = writer_input.job_id
+        self.anomaly_factory = writer_input.anomaly_factory
+        self.anomalies = []
+        self.ndigits = writer_input.ndigits
+        self.beta = 0.99
+
+    @staticmethod
+    def stack_tensors(tensor_list):
+        """
+        Torch not support stack cpu and xpu tensors. Group the tensors into cpu_group and xpu_group,
+        stack them separately, migrate xpu_group to cpu, and then restore in the order of input.
+
+        :param tensor_list: [tensor(-1.6165), tensor(-1.0985), tensor(-1.7777), tensor(-1.8408, device='npu:0')]
+        :return: result: list of float
+        """
+        cpu_tensors = []
+        xpu_tensors = []
+
+        for tensor in tensor_list:
+            if isinstance(tensor, torch.Tensor) and tensor.device.type != 'cpu':
+                # 将device上的tensor先stack后to cpu
+                xpu_tensors.append(tensor)
+            else:
+                cpu_tensors.append(tensor)
+
+        xpu_stack = torch.stack(xpu_tensors).cpu() if xpu_tensors else torch.tensor([])
+
+        # 按照输入的顺序恢复
+        result = []
+        cpu_tensors_idx, xpu_tensors_idx = 0, 0
+        for tensor in tensor_list:
+            if isinstance(tensor, torch.Tensor) and tensor.device.type != 'cpu':
+                result.append(xpu_stack[xpu_tensors_idx])
+                xpu_tensors_idx += 1
+            else:
+                result.append(cpu_tensors[cpu_tensors_idx])
+                cpu_tensors_idx += 1
+
+        return result
+
+    def get_anomalies(self):
+        """返回已检测到的异常列表
+        """
+        return self.anomalies
+
+    def clear_anomalies(self):
+        self.anomalies.clear()
+
+    def add_scalar(self, tag, scalar_value, global_step=None):
+        """If an anomaly is detected, the anomaly information is recorded and added to self.anomalies.
+        Args:
+            tag (tuple): tuple of tag_name and tag like ('0:1.post_attention_norm.weight/rank0/pre_grad', 'min').
+            scalar_value (float): scalar_value.
+            global_step (int): global_step.
+        Returns:
+            None
+        """
+        if not self.ad_rules or tag[-1] in ["shape", "dtype"]:
+            return
+        if isinstance(scalar_value, torch.Tensor):
+            scalar_value = scalar_value.item()
+        avg = self._update_tag2scalars(tag, scalar_value)
+        detected, rule_name = self._ad(scalar_value, history=avg)
+        if detected:
+            if rule_name == AnomalyTurbulence.name and tag[-1] not in ["norm", "mean"]:
+                return
+            exception_message = (f"Rule {rule_name} reports anomaly signal in {tag} at step {global_step}, "
+                                 f"current value {scalar_value}, history mean {avg}.")
+            logger.info(f"{BCOLORS.WARNING}> {exception_message}{BCOLORS.ENDC}")
+            # append to self.anomalies for dump
+            if self.anomaly_factory:
+                self.anomalies.append(self.anomaly_factory.create(tag, exception_message, global_step))
+
+    def write_metrics(self, ops, metric_value, step, prefix=''):
+        if not metric_value:
+            return
+        tensors = []
+        tags = list(itertools.product(metric_value.keys(), ops))
+        for op2tensor in metric_value.values():
+            tensors.extend(op2tensor.values())
+        if not tensors:
+            return
+
+        n_slices = len(tensors) // MonitorConst.SLICE_SIZE
+        with torch.no_grad():
+            for i in range(n_slices + 1):
+                begin = i * MonitorConst.SLICE_SIZE
+                end = (i + 1) * MonitorConst.SLICE_SIZE
+                if begin == len(tensors):
+                    continue
+                metric_list = self.stack_tensors(tensors[begin:end])
+                for tag, metric in zip(tags[begin:end], metric_list):
+                    self.add_scalar(tag, metric, step)
+
+    def _ad(self, scalar_value, history):
+        return AnomalyScanner.scan(self.ad_rules, history, cur=scalar_value)
+
+    def _update_tag2scalars(self, tag, scalar_value):
+        """Update the average and count of a scalar value associated with a tag.
+
+        This method is used to maintain a running average of scalar values for each tag.
+
+
+        Args:
+            tag (str): The tag identifier.
+            scalar_value (float): The scalar value to be added.
+
+        Returns:
+            float: The average value before update.
+        """
+        abs_scalar_value = abs(scalar_value)
+        if tag not in self.tag2scalars:
+            self.tag2scalars[tag] = {'avg': abs_scalar_value, 'count': 0}
+        avg = self.tag2scalars[tag]['avg']
+        self.tag2scalars[tag]['avg'] = self.beta * avg + (1 - self.beta) * abs_scalar_value
+        self.tag2scalars[tag]['count'] += 1
+        return avg
+
+
+class CSVWriterWithAD(BaseWriterWithAD):
+    def __init__(self, writer_input: WriterInput):
+        super().__init__(writer_input)
+
+        path = writer_input.path
+        self.log_dir = path
+        create_directory(path)
+        change_mode(path, FileCheckConst.DATA_DIR_AUTHORITY)
+        self.context_dict = defaultdict(list)
+        self.header = []
+        self.step_count_per_record = writer_input.step_count_per_record
+
+    def get_step_interval(self, step):
+        count = step // self.step_count_per_record
+        return count * self.step_count_per_record, (count + 1) * self.step_count_per_record - 1
+
+    def write_csv(self, prefix, step):
+        """
+        Args:
+            prefix[str]: prefix of output csv file e.g. grad_unreduced
+            step[int]
+        """
+        if len(self.context_dict) == 0:
+            return
+
+        ster_start, step_end = self.get_step_interval(step)
+        filepath = os.path.join(self.log_dir, f'{prefix}_{ster_start}-{step_end}.csv')
+        if not os.path.exists(filepath):
+            data_frame = pd.DataFrame(columns=self.header)
+            write_df_to_csv(data_frame, filepath)
+
+        new_data = []
+        for name, metric_value in self.context_dict.items():
+            new_line = name.split(MonitorConst.NAME_SEP) + metric_value
+            new_line.insert(2, step)
+            new_data.append(new_line)
+        new_data = pd.DataFrame(new_data).round(self.ndigits).fillna("nan")
+        write_df_to_csv(new_data, filepath, mode='a+', header=False)
+        self.context_dict = defaultdict(list)
+
+    def add_scalar(self, tag, scalar_value, global_step):
+        """
+        ('0:1.post_attention_norm.weight/rank0/pre_grad', 'min')
+        """
+        super().add_scalar(tag, scalar_value, global_step)
+
+        name = tag[0].split('/')[0]
+        if isinstance(scalar_value, torch.Tensor):
+            value = scalar_value.item()
+        elif isinstance(scalar_value, torch.Size):
+            value = list(scalar_value)
+        else:
+            value = scalar_value
+        self.context_dict[name].append(value)
+
+    def write_metrics(self, ops, metric_value, step, prefix='', **kwargs):
+        super().write_metrics(ops, metric_value, step, prefix='')
+
+        if prefix in [MonitorConst.ACTV, MonitorConst.ACTVGRAD] or kwargs.get("use_micro_step"):
+            self.header = MonitorConst.CSV_HEADER_MICRO_STEP + ops
+        else:
+            self.header = MonitorConst.CSV_HEADER + ops
+        self.write_csv(prefix, step)
+
+    def close(self):
+        pass
+
+
+class SummaryWriterWithAD(SummaryWriter, BaseWriterWithAD):
+    def __init__(self, writer_input: WriterInput):
+
+        path = writer_input.path
+        if not os.path.exists(path):
+            create_directory(path)
+        try:
+            super(SummaryWriter, self).__init__(writer_input)
+            super().__init__(path)
+        except Exception as e:
+            logger.error(f'error when init summary writer at {path}: {e}')
+            raise ValueError("Init summary writer error.") from e
+
+    def add_scalar(self, tag, scalar_value, global_step):
+        super(SummaryWriter, self).add_scalar(tag, scalar_value, global_step)
+        tag = f'{tag[0]}_{tag[1]}'
+        super().add_scalar(tag, scalar_value, global_step)