真香！Vision Transformer 快速实现 Mnist 识别

#残差模块，放在每个前馈网络和注意力之后
class Residual(nn.Module):
    def __init__(self, fn):
        super().__init__()
        self.fn = fn
    def forward(self, x, **kwargs):
        return self.fn(x, **kwargs) + x
#layernorm归一化,放在多头注意力层和激活函数层。用绝对位置编码的BERT，layernorm用来自身通道归一化
class PreNorm(nn.Module):
    def __init__(self, dim, fn):
        super().__init__()
        self.norm = nn.LayerNorm(dim)
        self.fn = fn
    def forward(self, x, **kwargs):
        return self.fn(self.norm(x), **kwargs)
#放置多头注意力后，因为在于多头注意力使用的矩阵乘法为线性变换，后面跟上由全连接网络构成的FeedForward增加非线性结构
class FeedForward(nn.Module):
    def __init__(self, dim, hidden_dim):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(dim, hidden_dim),
            nn.GELU(),
            nn.Linear(hidden_dim, dim)
        )
    def forward(self, x):
        return self.net(x)
#多头注意力层，多个自注意力连起来。使用qkv计算
class Attention(nn.Module):
    def __init__(self, dim, heads=8):
        super().__init__()
        self.heads = heads
        self.scale = dim ** -0.5
        self.to_qkv = nn.Linear(dim, dim * 3, bias=False)
        self.to_out = nn.Linear(dim, dim)
    def forward(self, x, mask = None):
        b, n, _, h = *x.shape, self.heads
        qkv = self.to_qkv(x)
        q, k, v = rearrange(qkv, 'b n (qkv h d) -> qkv b h n d', qkv=3, h=h)
        dots = torch.einsum('bhid,bhjd->bhij', q, k) * self.scale
        if mask is not None:
            mask = F.pad(mask.flatten(1), (1, 0), value = True)
            assert mask.shape[-1] == dots.shape[-1], 'mask has incorrect dimensions'
            mask = mask[:, None, :] * mask[:, :, None]
            dots.masked_fill_(~mask, float('-inf'))
            del mask
        attn = dots.softmax(dim=-1)
        out = torch.einsum('bhij,bhjd->bhid', attn, v)
        out = rearrange(out, 'b h n d -> b n (h d)')
        out =  self.to_out(out)
        return out

class ViT(nn.Module):
    def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels=3):
        super().__init__()
        assert image_size % patch_size == 0, 'image dimensions must be divisible by the patch size'
        num_patches = (image_size // patch_size) ** 2
        patch_dim = channels * patch_size ** 2
        self.patch_size = patch_size
        self.pos_embedding = nn.Parameter(torch.randn(1, num_patches + 1, dim))
        self.patch_to_embedding = nn.Linear(patch_dim, dim)
        self.cls_token = nn.Parameter(torch.randn(1, 1, dim))
        self.transformer = Transformer(dim, depth, heads, mlp_dim)
        self.to_cls_token = nn.Identity()
        self.mlp_head = nn.Sequential(
            nn.Linear(dim, mlp_dim),
            nn.GELU(),
            nn.Linear(mlp_dim, num_classes)
        )
    def forward(self, img, mask=None):
        p = self.patch_size
        x = rearrange(img, 'b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = p, p2 = p)
        x = self.patch_to_embedding(x)
        cls_tokens = self.cls_token.expand(img.shape[0], -1, -1)
        x = torch.cat((cls_tokens, x), dim=1)
        x += self.pos_embedding
        x = self.transformer(x, mask)
        x = self.to_cls_token(x[:, 0])
        return self.mlp_head(x)

model = ViT(image_size=28, patch_size=7, num_classes=10, channels=1,
            dim=64, depth=6, heads=8, mlp_dim=128)
optimizer = optim.Adam(model.parameters(), lr=0.003)
train_loss_history, test_loss_history = [], []
for epoch in range(1, N_EPOCHS + 1):
print('Epoch:', epoch)
    train_epoch(model, optimizer, train_loader, train_loss_history)
    evaluate(model, test_loader, test_loss_history)
print('Execution time:', '{:5.2f}'.format(time.time() - start_time), 'seconds')