zx214

Results 8 comments of zx214

dbnet在Ascend910B3上训练速度非常慢

感谢，我来试一下

dbnet在Ascend910B3上训练速度非常慢

哈喽，安装官网安装了mindspore2.2.12和CANN7.0.0.beta1，测试源代码模型编译时间没问题，根据任务需求添加了summary的记录和修改数据读取部分，模型部分没有动；但是模型编译时间还是很长，请问模型编译速度慢可能因为什么影响呢

dbnet在Ascend910B3上训练速度非常慢

好的，我定位一下，再向您请教个问题在2卡训练中，输出这样的日志，怎样原因导致的，如何解决呢 ![image](https://github.com/mindspore-lab/mindocr/assets/55902473/52de9c5e-b7f0-4fe4-b218-08099e1b15db)

When I used svtr_tiny_ch.yaml to configure training, there was a loss of nan

My yaml config is as follows： system: mode: 0 # 0 for graph mode, 1 for pynative mode in MindSpore distribute: False amp_level: O0 amp_level_infer: O0 # running inference in...

When I used svtr_tiny_ch.yaml to configure training, there was a loss of nan

In my config，I haved set the amp_level and amp_level_infer to O0。How can I solve this problem?

When I used svtr_tiny_ch.yaml to configure training, there was a loss of nan

There is no problem training with English datasets and svtr_tiny.yaml.

When I used svtr_tiny_ch.yaml to configure training, there was a loss of nan

I used RCTW dataset on svtr_tiny_ch with svtr_tiny_ch-2ee6ade4.ckpt pretrained model.

When I used svtr_tiny_ch.yaml to configure training, there was a loss of nan

CUDA Version: 11.4 and mindspore2.2.11