Captain_THU
Captain_THU
 Why does the eta time gradually increase with training?
 将linux 主机生成的install文件放到板子上,发现初始化模型就失败了,这是为什么 onnx模型可以使用Python demo 可以正常推理

my question is above like the title
merge lora 
resume training need what config setting?
我使用自己数据集进行训练,在验证集上map很低,我看你提供的example中,似乎随着训练 map不升反降
Could you provide the detail of install on the cuda>11 ? cuda9.0 is too old . I attempted many times but failed.