Ascend / pytorch

环境：

Version=23.0.0
ascendhal_version=7.35.19
aicpu_version=1.0
tdt_version=1.0
log_version=1.0
prof_version=2.0
dvppkernels_version=1.1
tsfw_version=1.0
Innerversion=V100R001C15SPC002B224
compatible_version=[V100R001C30],[V100R001C13],[V100R001C15]
compatible_version_fw=[7.0.0,7.1.99]
package_version=23.0.0

torch==2.1.0
torch-npu==2.1.0.post20240112

现象描述

8卡910b训练 sdxl（参考 https://gitee.com/ascend/ModelZoo-PyTorch/tree/master/PyTorch/built-in/diffusion/sd-scripts-xl）
按照 https://www.hiascend.com/document/detail/zh/CANNCommunityEdition/700alpha002/ptmoddevg/ptmigr/AImpug_0066.html#ZH-CN_TOPIC_0000001711588536__section1022713933315 的方式做 profile，得到 trace_view.json
无法解析 trace_view.json

可能和以下拼 json 的代码有关：

pytorch/torch_npu/profiler/analysis/prof_common_func/file_manager.py

Lines 93 to 100 in 6e76afc

    
           def append_trace_json_by_path(cls, output_path: str, data: list, new_name: str) -> None: 
        
               PathManager.check_directory_path_writeable(output_path) 
        
               try: 
        
                   with open(output_path, "a") as file: 
        
                       data = json.dumps(data, ensure_ascii=False) 
        
                       data = f",{data[1:]}" 
        
                       file.write(data) 
        
               except Exception as err:

建议按照 https://gitee.com/ascend/pytorch/tree/v2.1.0-5.0.0/ 配套表中(昇腾辅助软件)的版本使用cann和torch_npu

	def append_trace_json_by_path(cls, output_path: str, data: list, new_name: str) -> None:
	PathManager.check_directory_path_writeable(output_path)
	try:
	with open(output_path, "a") as file:
	data = json.dumps(data, ensure_ascii=False)
	data = f",{data[1:]}"
	file.write(data)
	except Exception as err:

Profile 拿到的 trace_view.json 不合法

环境：

现象描述