MindSpore开启profile，使用并行策略报错ValueError: When dıstrıbuted loads are slıced we1ghts, sınk mode must be

1 系统环境

硬件环境(Ascend/GPU/CPU): Ascend/GPU/CPU

MindSpore版本: 2.1

执行模式（PyNative/ Graph）: 不限

2 报错信息

2.1 问题描述

使用离线切分方法切好模型后，加载切分后的模型进行训练，出现以下报错：

Traceback (most recent call last):   File "wizardcoder/ run_wizardcoder.py", line 148, in <module>     device_id=args.device_id)   File "wizardcoder/ run_wizardcoder.py", line 90, in main     task.finetune(finetune_checkpoint=config. load_checkpoint, auto_trans_ckpt=config.auto_trans_ckpt, resume=resume)     File "/home/wizardcoder/1_wizardcoder-mindformers-916/mindformers/trainer/trainer.py", Tine 522, in finetune     is full config=True. **kwargs)   File "/home/wizardcoder/1 wizardcoder-mindformers-916/mindforme rs/trainer/caus al_language_modeling/caus al_language_modeling.py", line 106, in train     **kwargs) File "/home/wizardcode r/1_wizardcoder-mindformers -916/mindformers/trainer/base_t rainer.py", line 616, in training_ process transform and load checkpoint(c onfig, model, network, dataset)  File "/home/wizardcoder/1 wizardcodermindformers -916/mindformers/trainer/utils .py", line 300, in transform and_load checkpoint build model(config, model. datas et, do eval=do eval, do predict-do predict)   File "/home/wizardcoder/1 wizardcoder-mindformers-916/mindformers/trainer/utils.py". line 317, in build model  raise ValueError( "when distributed loads are s liced weights. sink mode must be set True

ValueError: When dıstrıbuted loads are slıced we1ghts, sınk mode must be set True.复制

3 根因分析

分析配置文件可知，此时开启了profile功能，当前不支持两者同时使用。

4 解决方案

不同时使用这两个功能。