Fix optimizer list parameters input bug

[wyg1997]

Optimizer支持3种传参方式：

1. Iterator: flow.optim.SGD(module.parameters(), lr=0.1)
2. List[Dict]: flow.optim.SGD([{"params": module1.parameters()}, {"params": module2.parameters()}])
3. List[Parameter or Tensor]: flow.optim.SGD([module.weight, module.bias])

现在在ParamGroup构造函数中统一参数为dict

[tea321000]

我切分支测过test_graph_optimizer和课件了 可以收敛

batch_size, num_epochs, lr = 1, 100, 0.003
net, loss = linear, squared_loss
#划分数据集
dataset = flow.utils.data.TensorDataset(train_features, train_labels)
train_iter = flow.utils.data.DataLoader(dataset, batch_size, shuffle=True)
#训练模型
def fit_and_plot(lambd):
    w, b = init_params()
    train_ls, test_ls = [], []
    for _ in range(num_epochs):
        for X, y in train_iter:
            # 添加了L2范数惩罚项
            l = loss(net(X, w, b), y) + lambd * l2_penalty(w)
            l = l.sum()
            if w.grad is not None:
                w.grad.data.zeros_()
                b.grad.data.zeros_()
            l.backward()
            SGD([w, b], lr)
        train_ls.append(loss(net(train_features, w, b), train_labels).mean().numpy())
        test_ls.append(loss(net(test_features, w, b), test_labels).mean().numpy())
    Draw_Loss_Curve(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
                 range(1, num_epochs + 1), test_ls, ['train', 'test'])
    print('L2 norm of w:', w.norm().numpy())

[github-actions]

CI failed, removing label automerge

[chengtbf]

mark。暂时移除 automerge，两个小时以后加回来。

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 148.1ms (= 7405.2ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.1ms (= 6403.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.16 (= 148.1ms / 128.1ms)

PyTorch resnet50 time: 83.5ms (= 4174.2ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.5ms (= 3724.7ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 83.5ms / 74.5ms)

PyTorch resnet50 time: 58.7ms (= 2933.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.7ms (= 2385.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.23 (= 58.7ms / 47.7ms)

PyTorch resnet50 time: 50.0ms (= 2500.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 45.2ms (= 2260.3ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 50.0ms / 45.2ms)

PyTorch resnet50 time: 44.4ms (= 2217.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.2ms (= 2358.3ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.94 (= 44.4ms / 47.2ms)