Lightning AI 快速入门：多 GPU 加速与零代码部署实战指南

def forward(self, x):
    x = x.cuda()
    layer_1.cuda()
    x_hat = layer_1(x)

def forward(self, x):
    x_hat = layer_1(x)

class LitAutoEncoder(pl.LightningModule):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Sequential(nn.Linear(*, ), nn.ReLU(), nn.Linear(, ))
        self.decoder = nn.Sequential(nn.Linear(, ), nn.ReLU(), nn.Linear(, *))

    def forward(self, x):
        return self.decoder(self.encoder(x))

    def training_step(self, batch, batch_idx):
        x, y = batch
        x = x.view(x.size(), -)
        z = self.encoder(x)
        x_hat = self.decoder(z)
        loss = F.mse_loss(x_hat, x)
        self.log('train_loss', loss)
        return loss

    def configure_optimizers(self):
        return torch.optim.Adam(self.parameters(), lr=1e-3)

model = LitAutoEncoder()
trainer = Trainer(accelerator="gpu", devices=, max_epochs=)
trainer.fit(model, train_loader)

import litserve as ls

class SimpleLitAPI(ls.LitAPI):
    def setup(self, device):
        # 加载模型等初始化操作
        pass

    def predict(self, prompt):
        # 处理输入，生成输出
        yield "This is a sample generated output"

api = SimpleLitAPI()
server = ls.LitServer(api, spec=ls.OpenAISpec())
server.run(port=)