![]() * [Inference]ADD Bench Chatglm2 script (#4963) * add bench chatglm * fix bug and make utils --------- Co-authored-by: CjhHa1 <cjh18671720497outlook.com> * [Pipeline inference] Combine kvcache with pipeline inference (#4938) * merge kvcache with pipeline inference and refactor the code structure * support ppsize > 2 * refactor pipeline code * do pre-commit * modify benchmark * fix bench mark * polish code * add docstring and update readme * refactor the code * fix some logic bug of ppinfer * polish readme * fix typo * skip infer test * updated c++17 compiler flags (#4983) * [Inference] Dynamic Batching Inference, online and offline (#4953) * [inference] Dynamic Batching for Single and Multiple GPUs (#4831) * finish batch manager * 1 * first * fix * fix dynamic batching * llama infer * finish test * support different lengths generating * del prints * del prints * fix * fix bug --------- Co-authored-by: CjhHa1 <cjh18671720497outlook.com> * [inference] Async dynamic batching (#4894) * finish input and output logic * add generate * test forward * 1 * [inference]Re push async dynamic batching (#4901) * adapt to ray server * finish async * finish test * del test --------- Co-authored-by: yuehuayingxueluo <867460659@qq.com> * Revert "[inference]Re push async dynamic batching (#4901)" (#4905) This reverts commit |
||
---|---|---|
.. | ||
bookkeeping | ||
__init__.py | ||
_utils.py | ||
low_level_optim.py | ||
readme.md |
readme.md
Low Level ZeRO
Low Level ZeRO == ZeRO-DP stage 1 and 2, we would denote it as ZeRO.
Examples of ZeRO and gradient accumulation
The code below only shows a typical gradient accumulation process, and it drops a lot of details, such as the processing of loss.
# examples of ZeRO1 with gradient accumulation
...
outputs = model(input)
loss = SomeLoss(outputs)
if (idx + 1) % ACCUMULATE_STEP != 0:
with booster.no_sync(model, optimizer):
# under this context, the gradient would not sync when backward,
# left each rank having different gradient.
# It saves the backward time
booster.backward(loss, optimizer)
continue
else:
# need to sync all the accumulated gradient
booster.backward(loss, optimizer):
optimizer.step()
...
# example of ZeRO2 with gradient accumulation
...
outputs = model(input)
loss = SomeLoss(outputs)
# ZeRO2 split the gradients and can NOT accumulate gradient with syncing.
booster.backward(loss, optimizer)
if (idx + 1) % ACCUMULATE_STEP == 0:
optimizer.step()
...
Design:
Notion
p32
denotes the param copy in the optimizer
p
denotes the model param
g
denotes the gradient
INIT
In low level zero(1, 2), p32
is split. Different from the previous implement, we split each p32
evenly by world_size. Thus, rank0 got a param list as [p00, p10]
, rank1 got a param list as [p-01, p-11]
, etc.
For the detailed implementation, we first pad p
for it can be split by world_size if needed. Then, we would view it to the shape [world_size, -1]
, and each rank got its own part p32
by cloning.
BWD
To leverage the communication, a gradient would be added to a bucket first. When the bucket is full, each g
in it would be reshaped as [world_size, -1]
. And the [local_rank]
parts would be united.
The data structure looks like this:
{
0: [g-00, g-10],
1: [g-01, g-11],
2: [g-02, g-12]
}
After that, the gradients would be flattened by rank, and the data structure looks like this:
# g-X0 means flatten([g-00, g-10])
{
0: [g-X0],
1: [g-X1],
2: [g-X2]
}
For zero1, we iterate the dictionary and do all_reduce
. For zero2, we can just do reduce-scatter
.
Optim
For each rank gets its own p32
and the counterpart g
, it is quite easy to do optim.step()
.
However, we have to consider a situation of layer drop, for instance:
class MlpModel(nn.Module):
def __init__(self):
super(MlpModel, self).__init__()
self.linear1 = nn.Linear(128, 256)
self.drop_linear = nn.Linear(256, 256)
self.linear2 = nn.Linear(256, 512)
def forward(self, x):
x = self.linear1(x)
x = self.linear2(x)
return x
And the solution is to build a mapping of p32
, p
, and g
. Before optim.step()
, we collect p
which requires_grad=True
and p.grad != None
as a real working param. And select the counterpart p32
and g
.