功能简介

TorchAir python层日志开启主要通过python中logger.setLevel( )函数实现，方便进行python层功能调试和问题定位。

支持的日志级别如下：

logging.DEBUG：日志级别DEBUG。
logging.INFO：日志级别INFO。
logging.WARNING：日志级别WARNING。
logging.ERROR：日志级别ERROR。

logger.setLevel的默认值为“logging.ERROR”，更详细的介绍请参见python官网logging模块。

使用方法

import logging
from torchair import logger
logger.setLevel(logging.DEBUG)

python侧debug日志样例如下

[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: arg1_1
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, 1] npu:Tensor(arg1_1:0, dtype=DT_FLOAT, size=[s0, 1]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: arg2_1
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(arg2_1:0, dtype=DT_FLOAT, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: arg3_1
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output Pack(meta:FakeTensor(dtype=torch.int32, size=[s0, s0] npu:Tensor(arg3_1:0, dtype=DT_INT32, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: aten.mul.Tensor
[DEBUG] TORCHAIR 2024-04-03 09:37:40 input 0: Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(arg2_1:0, dtype=DT_FLOAT, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 input 1: Pack(meta:FakeTensor(dtype=torch.int32, size=[s0, s0] npu:Tensor(arg3_1:0, dtype=DT_INT32, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(Cast_1:0, dtype=DT_FLOAT, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: aten.add.Tensor
[DEBUG] TORCHAIR 2024-04-03 09:37:40 input 0: Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, 1] npu:Tensor(arg1_1:0, dtype=DT_FLOAT, size=[s0, 1]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 input 1: Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(Cast_1:0, dtype=DT_FLOAT, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(Add:0, dtype=DT_FLOAT, size=[s0, s0]))
[DEBUG] TORCHAIR 2024-04-03 09:37:40 -------------------
[DEBUG] TORCHAIR 2024-04-03 09:37:40 target: output
[DEBUG] TORCHAIR 2024-04-03 09:37:40 input 0: (Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(Add:0, dtype=DT_FLOAT, size=[s0, s0])),)
[DEBUG] TORCHAIR 2024-04-03 09:37:40 output (Pack(meta:FakeTensor(dtype=torch.float32, size=[s0, s0] npu:Tensor(Add:0, dtype=DT_FLOAT, size=[s0, s0])),)
[DEBUG] TORCHAIR 2024-04-03 09:37:41 runtime inputs
[DEBUG] TORCHAIR 2024-04-03 09:37:41   input 0: <class 'int'>(4)