element 0 of tensors does not require grad and does not have a grad_fn

1.检查需要训练的模型参数是否设置requires_grad_为True：

grads_lis = []
for param in model.parameters():
    # print(param.requires_grad)
    grads_lis.append(param.requires_grad)
print('-----------------------')

param_name_lis = []
for name, param in model.named_parameters():
    # print(f"Name: {name}, Parameter shape: {param.shape}")
    param_name_lis.append(name)

# set grad true
print('------set paprams grad true--------')
false_grads = [i for i, value in enumerate(grads_lis) if not value]

for i in false_grads:
    parameter_name = param_name_lis[i]

    # 通过名称获取参数
    parameter = None
    for name, param in model.named_parameters():
        if name == parameter_name:
            parameter = param
            break

    if parameter is not None:
        parameter.requires_grad = True
    else:
        print(f"Parameter with name '{parameter_name}' not found.")

# check the new model params
grads_

原文地址：https://blog.csdn.net/qq_26169815/article/details/140717301

免责声明：本站文章内容转载自网络资源，如本站内容侵犯了原著者的合法权益，可联系本站删除。更多内容请关注自学内容网（zxcms.com）！