给出对CLIP预训练模型知识蒸馏的训练代码，要求有kl_div loss和soft loss和hard loss，只给出训练代码即可，重点体现soft loss和hard loss

以下是一个基于PyTorch的CLIP知识蒸馏训练代码，包括KL散度loss、soft loss和hard loss：

import torch
import torch.nn.functional as F
from torch.utils.data import DataLoader
from torchvision.datasets import CIFAR10
from torchvision.transforms import Compose, ToTensor, Normalize

from models import StudentModel, TeacherModel


def kl_div_loss(p, q):
    """
    KL散度loss
    """
    return F.kl_div(F.log_softmax(p, dim=1), F.softmax(q, dim=1), reduction='batchmean')


def soft_loss(p, q, t=3):
    """
    soft loss
    """
    return F.kl_div(F.log_softmax(p/t, dim=1), F.softmax(q/t, dim=1), reduction='batchmean') * t * t


def hard_loss(p, q):
    """
    hard loss
    """
    return F.cross_entropy(p, q)


def train():
    # 加载数据
    transform = Compose([ToTensor(), Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
    dataset = CIFAR10('./data', train=True, download=True, transform=transform)
    dataloader = DataLoader(dataset, batch_size=128, shuffle=True, num_workers=4, pin_memory=True)

    # 加载模型
    teacher = TeacherModel()
    student = StudentModel()

    # 设置优化器
    teacher_optimizer = torch.optim.Adam(teacher.parameters(), lr=0.001)
    student_optimizer = torch.optim.Adam(student.parameters(), lr=0.001)

    # 设置温度
    t = 3

    # 训练
    for epoch in range(10):
        for i, (images, labels) in enumerate(dataloader):
            # 前向传播
            teacher_outputs = teacher(images)
            student_outputs = student(images)

            # 计算loss
            kl_loss = kl_div_loss(student_outputs, teacher_outputs.detach())
            soft_loss = soft_loss(student_outputs, teacher_outputs.detach(), t)
            hard_loss = hard_loss(student_outputs, labels)
            total_loss = kl_loss + soft_loss + hard_loss

            # 反向传播
            teacher_optimizer.zero_grad()
            student_optimizer.zero_grad()
            total_loss.backward()
            teacher_optimizer.step()
            student_optimizer.step()

            # 打印结果
            if (i+1) % 100 == 0:
                print(f'Epoch [{epoch+1}/10], Step [{i+1}/391], Loss: {total_loss.item():.4f}')

        # 降低温度
        t *= 0.9

    # 保存模型
    torch.save(student.state_dict(), 'student.pth')

在上面的代码中，`kl_div_loss()`函数计算KL散度loss，`soft_loss()`函数计算soft loss，`hard_loss()`函数计算hard loss。在训练过程中，我们使用了这三种loss的加权和作为总的loss进行反向传播。同时，我们还使用了不同的温度t来控制soft loss的权重，以达到更好的效果。最终，我们保存了训练好的student模型。