AI编辑：我是小将
混合精度训练（mixed precision training）可以让模型训练在尽量不降低性能的情形下提升训练速度，而且也可以降低显卡使用内存。目前主流的深度学习框架都开始支持混合精度训练。对于PyTorch，混合精度训练还主要是采用NVIDIA开源的apex库。但是，PyTorch将迎来重大更新，那就是提供内部支持的混合精度训练，而且是自动混合精度训练：
torch.cuda.amp.autocast：自动为GPU op选择精度来提升训练性能而不降低模型准确度。torch.cuda.amp.GradScaler: 对梯度进行scale来加快模型收敛，因为float16梯度容易出现underflow（梯度过小）
两者结合在一起，可以实现自动混合精度训练：
代码语言：javascript复制# Creates model and optimizer in default precision
model = Net().cuda()
optimizer = optim.SGD(model.parameters(), ...)
# Creates a GradScaler once at the beginning of training.
scaler = GradScaler()
for epoch in epochs:
for input, target in data:
optimizer.zero_grad()
# Runs the forward pass with autocasting.
with autocast():
output = model(input)
loss = loss_fn(output, target)
# Scales loss. Calls backward() on scaled loss to create scaled gradients.
# Backward passes under autocast are not recommended.
# Backward ops run in the same precision that autocast used for corresponding forward ops.
scaler.scale(loss).backward()
# scaler.step() first unscales the gradients of the optimizer's assigned params.
# If these gradients do not contain infs or NaNs, optimizer.step() is then called,
# otherwise, optimizer.step() is skipped.
scaler.step(optimizer)
# Updates the scale for next iteration.
scaler.update()
可以看到，为了防止梯度的underflow，首先scaler.scale(loss).backward()会对loss乘以一个scale因子，然后backward时所有梯度都会乘以相同的scale因子，这样保证梯度有较大的magnitude而不会出现为0。我们不希望这个scale因子对学习速率产生影响，那么scaler.step(optimizer)会先unscale要更新的梯度然后再更新，如果梯度出现infs或者NaNs，optimizer将忽略这次迭代训练。
如果你想在梯度更新前对梯度进行clip，也是可以的：
代码语言：javascript复制scaler = GradScaler()
for epoch in epochs:
for input, target in data:
optimizer.zero_grad()
with autocast():
output = model(input)
loss = loss_fn(output, target)
scaler.scale(loss).backward()
# Unscales the gradients of optimizer's assigned params in-place
scaler.unscale_(optimizer)
# Since the gradients of optimizer's assigned params are unscaled, clips as usual:
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
# optimizer's gradients are already unscaled, so scaler.step does not unscale them,
# although it still skips optimizer.step() if the gradients contain infs or NaNs.
scaler.step(optimizer)
# Updates the scale for next iteration.
scaler.update()
当然，混合精度训练肯定要支持分布式训练，由于autocast是thread local的，所以要注意以下不同的情形：
如果使用torch.nn.DataParallel：
此时只有一个进程，而不同GPU上是各自的线程跑forward过程的，所以下面操作时无效的：
代码语言：javascript复制model = MyModel()
dp_model = nn.DataParallel(model)
# Sets autocast in the main thread
with autocast():
# dp_model's internal threads won't autocast. The main thread's autocast state has no effect.
output = dp_model(input)
# loss_fn still autocasts, but it's too late...
loss = loss_fn(output)
此时你需要对model的forward方法用autocast装饰：
MyModel(nn.Module):
...
@autocast()
def forward(self, input):
...
# Alternatively
MyModel(nn.Module):
...
def forward(self, input):
with autocast():
...
model = MyModel()
dp_model = nn.DataParallel(model)
with autocast():
output = dp_model(input)
loss = loss_fn(output)
如果使用
torch.nn.parallel.DistributedDataParallel：
一般情形下是单GPU进程的，此时原来的用来就没有问题，但是如果是多GPU一个进程那么就和上述问题一样，需要用autocast装饰model的forward。
