A simple and accurate CUDA memory management laboratory for pytorch, it consists of different parts about the memory:
-
Features:
- Memory Profiler: A
line_profiler
style CUDA memory profiler with simple API. - Memory Reporter: A reporter to inspect tensors occupying the CUDA memory.
- Courtesy: An interesting feature to temporarily move all the CUDA tensors into CPU memory for courtesy, and of course the backward transferring.
- IPython support through
%mlrun
/%%mlrun
line/cell magic commands.
- Memory Profiler: A
-
Table of Contents
- Released version:
pip install pytorch_memlab
- Newest version:
pip install git+https://github.com/stonesjtu/pytorch_memlab
Out-Of-Memory errors in pytorch happen frequently, for new-bees and experienced programmers. A common reason is that most people don't really learn the underlying memory management philosophy of pytorch and GPUs. They wrote memory in-efficient codes and complained about pytorch eating too much CUDA memory.
In this repo, I'm going to share some useful tools to help debugging OOM, or to inspect the underlying mechanism if anyone is interested in.
The memory profiler is a modification of python's line_profiler
, it gives
the memory usage info for each line of code in the specified function/method.
import torch
from pytorch_memlab import LineProfiler
def inner():
torch.nn.Linear(100, 100).cuda()
def outer():
linear = torch.nn.Linear(100, 100).cuda()
linear2 = torch.nn.Linear(100, 100).cuda()
linear3 = torch.nn.Linear(100, 100).cuda()
work()
After the script finishes or interrupted by keyboard, it gives the following profiling info if you're in a Jupyter notebook:
or the following info if you're in a text-only terminal:
## outer
active_bytes reserved_bytes line code
all all
peak peak
0.00B 0.00B 7 def outer():
40.00K 2.00M 8 linear = torch.nn.Linear(100, 100).cuda()
80.00K 2.00M 9 linear2 = torch.nn.Linear(100, 100).cuda()
120.00K 2.00M 10 inner()
## inner
active_bytes reserved_bytes line code
all all
peak peak
80.00K 2.00M 4 def inner():
120.00K 2.00M 5 torch.nn.Linear(100, 100).cuda()
An explanation of what each column means can be found in the Torch documentation. The name of any field from memory_stats()
can be passed to display()
to view the corresponding statistic.
If you use profile
decorator, the memory statistics are collected during
multiple runs and only the maximum one is displayed at the end.
We also provide a more flexible API called profile_every
which prints the
memory info every N times of function execution. You can simply replace
@profile
with @profile_every(1)
to print the memory usage for each
execution.
The @profile
and @profile_every
can also be mixed to gain more control
of the debugging granularity.
- You can also add the decorator in the module class:
class Net(torch.nn.Module):
def __init__(self):
super().__init__()
@profile
def forward(self, inp):
#do_something
- The Line Profiler profiles the memory usage of CUDA device 0 by default,
you may want to switch the device to profile by
set_target_gpu
. The gpu selection is globally, which means you have to remember which gpu you are profiling on during the whole process:
import torch
from pytorch_memlab import profile, set_target_gpu
@profile
def func():
net1 = torch.nn.Linear(1024, 1024).cuda(0)
set_target_gpu(1)
net2 = torch.nn.Linear(1024, 1024).cuda(1)
set_target_gpu(0)
net3 = torch.nn.Linear(1024, 1024).cuda(0)
func()
More samples can be found in test/test_line_profiler.py
Make sure you have IPython
installed, or have installed pytorch-memlab
with
pip install pytorch-memlab[ipython]
.
First, load the extension:
%load_ext pytorch_memlab
This makes the %mlrun
and %%mlrun
line/cell magics available for use. For
example, in a new cell run the following to profile an entire cell
%%mlrun -f func
import torch
from pytorch_memlab import profile, set_target_gpu
def func():
net1 = torch.nn.Linear(1024, 1024).cuda(0)
set_target_gpu(1)
net2 = torch.nn.Linear(1024, 1024).cuda(1)
set_target_gpu(0)
net3 = torch.nn.Linear(1024, 1024).cuda(0)
Or you can invoke the profiler for a single statement on via the %mlrun
cell
magic.
import torch
from pytorch_memlab import profile, set_target_gpu
def func(input_size):
net1 = torch.nn.Linear(input_size, 1024).cuda(0)
%mlrun -f func func(2048)
See %mlrun?
for help on what arguments are supported. You can set the GPU
device to profile, dump profiling results to a file, and return the
LineProfiler
object for post-profile inspection.
Find out more by checking out the demo Jupyter notebook
As Memory Profiler only gives the overall memory usage information by lines, a more low-level memory usage information can be obtained by Memory Reporter.
Memory reporter iterates all the Tensor
objects and gets the underlying
UntypedStorage
(previously Storage
) object to get the actual memory usage instead of the surface
Tensor.size
.
see UntypedStorage for detailed information
- A minimal one:
import torch
from pytorch_memlab import MemReporter
linear = torch.nn.Linear(1024, 1024).cuda()
reporter = MemReporter()
reporter.report()
outputs:
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
Parameter0 (1024, 1024) 4.00M
Parameter1 (1024,) 4.00K
-------------------------------------------------------------------------------
Total Tensors: 1049600 Used Memory: 4.00M
The allocated memory on cuda:0: 4.00M
-------------------------------------------------------------------------------
- You can also pass in a model object for automatically name inference.
import torch
from pytorch_memlab import MemReporter
linear = torch.nn.Linear(1024, 1024).cuda()
inp = torch.Tensor(512, 1024).cuda()
# pass in a model to automatically infer the tensor names
reporter = MemReporter(linear)
out = linear(inp).mean()
print('========= before backward =========')
reporter.report()
out.backward()
print('========= after backward =========')
reporter.report()
outputs:
========= before backward =========
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
weight (1024, 1024) 4.00M
bias (1024,) 4.00K
Tensor0 (512, 1024) 2.00M
Tensor1 (1,) 512.00B
-------------------------------------------------------------------------------
Total Tensors: 1573889 Used Memory: 6.00M
The allocated memory on cuda:0: 6.00M
-------------------------------------------------------------------------------
========= after backward =========
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
weight (1024, 1024) 4.00M
weight.grad (1024, 1024) 4.00M
bias (1024,) 4.00K
bias.grad (1024,) 4.00K
Tensor0 (512, 1024) 2.00M
Tensor1 (1,) 512.00B
-------------------------------------------------------------------------------
Total Tensors: 2623489 Used Memory: 10.01M
The allocated memory on cuda:0: 10.01M
-------------------------------------------------------------------------------
- The reporter automatically deals with the sharing weights parameters:
import torch
from pytorch_memlab import MemReporter
linear = torch.nn.Linear(1024, 1024).cuda()
linear2 = torch.nn.Linear(1024, 1024).cuda()
linear2.weight = linear.weight
container = torch.nn.Sequential(
linear, linear2
)
inp = torch.Tensor(512, 1024).cuda()
# pass in a model to automatically infer the tensor names
out = container(inp).mean()
out.backward()
# verbose shows how storage is shared across multiple Tensors
reporter = MemReporter(container)
reporter.report(verbose=True)
outputs:
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
0.weight (1024, 1024) 4.00M
0.weight.grad (1024, 1024) 4.00M
0.bias (1024,) 4.00K
0.bias.grad (1024,) 4.00K
1.bias (1024,) 4.00K
1.bias.grad (1024,) 4.00K
Tensor0 (512, 1024) 2.00M
Tensor1 (1,) 512.00B
-------------------------------------------------------------------------------
Total Tensors: 2625537 Used Memory: 10.02M
The allocated memory on cuda:0: 10.02M
-------------------------------------------------------------------------------
- You can better understand the memory layout for more complicated module:
import torch
from pytorch_memlab import MemReporter
lstm = torch.nn.LSTM(1024, 1024).cuda()
reporter = MemReporter(lstm)
reporter.report(verbose=True)
inp = torch.Tensor(10, 10, 1024).cuda()
out, _ = lstm(inp)
out.mean().backward()
reporter.report(verbose=True)
As shown below, the (->)
indicates the re-use of the same storage back-end
outputs:
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
weight_ih_l0 (4096, 1024) 32.03M
weight_hh_l0(->weight_ih_l0) (4096, 1024) 0.00B
bias_ih_l0(->weight_ih_l0) (4096,) 0.00B
bias_hh_l0(->weight_ih_l0) (4096,) 0.00B
Tensor0 (10, 10, 1024) 400.00K
-------------------------------------------------------------------------------
Total Tensors: 8499200 Used Memory: 32.42M
The allocated memory on cuda:0: 32.52M
Memory differs due to the matrix alignment
-------------------------------------------------------------------------------
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
weight_ih_l0 (4096, 1024) 32.03M
weight_ih_l0.grad (4096, 1024) 32.03M
weight_hh_l0(->weight_ih_l0) (4096, 1024) 0.00B
weight_hh_l0.grad(->weight_ih_l0.grad) (4096, 1024) 0.00B
bias_ih_l0(->weight_ih_l0) (4096,) 0.00B
bias_ih_l0.grad(->weight_ih_l0.grad) (4096,) 0.00B
bias_hh_l0(->weight_ih_l0) (4096,) 0.00B
bias_hh_l0.grad(->weight_ih_l0.grad) (4096,) 0.00B
Tensor0 (10, 10, 1024) 400.00K
Tensor1 (10, 10, 1024) 400.00K
Tensor2 (1, 10, 1024) 40.00K
Tensor3 (1, 10, 1024) 40.00K
-------------------------------------------------------------------------------
Total Tensors: 17018880 Used Memory: 64.92M
The allocated memory on cuda:0: 65.11M
Memory differs due to the matrix alignment
-------------------------------------------------------------------------------
NOTICE:
When forwarding with
grad_mode=True
, pytorch maintains tensor buffers for future Back-Propagation, in C level. So these buffers are not going to be managed or collected by pytorch. But if you store these intermediate results as python variables, then they will be reported.
-
You can also filter the device to report on by passing extra arguments:
report(device=torch.device(0))
-
A failed example due to pytorch's C side tensor buffers
In the following example, a temp buffer is created at inp * (inp + 2)
to
store both inp
and inp + 2
, unfortunately python only knows the existence
of inp, so we have 2M memory lost, which is the same size of Tensor inp
.
import torch
from pytorch_memlab import MemReporter
linear = torch.nn.Linear(1024, 1024).cuda()
inp = torch.Tensor(512, 1024).cuda()
# pass in a model to automatically infer the tensor names
reporter = MemReporter(linear)
out = linear(inp * (inp + 2)).mean()
reporter.report()
outputs:
Element type Size Used MEM
-------------------------------------------------------------------------------
Storage on cuda:0
weight (1024, 1024) 4.00M
bias (1024,) 4.00K
Tensor0 (512, 1024) 2.00M
Tensor1 (1,) 512.00B
-------------------------------------------------------------------------------
Total Tensors: 1573889 Used Memory: 6.00M
The allocated memory on cuda:0: 8.00M
Memory differs due to the matrix alignment or invisible gradient buffer tensors
-------------------------------------------------------------------------------
Sometimes people would like to preempt your running task, but you don't want to save checkpoint and then load, actually all they need is GPU resources ( typically CPU resources and CPU memory is always spare in GPU clusters), so you can move all your workspaces from GPU to CPU and then halt your task until a restart signal is triggered, instead of saving&loading checkpoints and bootstrapping from scratch.
Still developing..... But you can have fun with:
from pytorch_memlab import Courtesy
iamcourtesy = Courtesy()
for i in range(num_iteration):
if something_happens:
iamcourtesy.yield_memory()
wait_for_restart_signal()
iamcourtesy.restore()
- As is stated above in
Memory_Reporter
, intermediate tensors are not covered properly, so you may want to insert such courtesy logics afterbackward
or beforeforward
. - Currently the CUDA context of pytorch requires about 1 GB CUDA memory, which means even all Tensors are on CPU, 1GB of CUDA memory is wasted, :-(. However it's still under investigation if I can fully destroy the context and then re-init.
I suffered a lot debugging weird memory usage during my 3-years of developing efficient Deep Learning models, and of course learned a lot from the great open source community.
- Fix
DataFrame.drop
for pandas 1.5+
- Fix colab error (#35)
- Support python3.8 (#38)
- Support sparse tensor (#30)
- Fix name mapping in
MemReporter
(#24) - Fix reporter without model input (#22 #25)
- Fix memory leak in
MemReporter
- Fix
line_profiler
not found
- Add jupyter notebook figure and ipython support
- Add ipython magic support (#8)
- Add gpu switch for line-profiler(#2)
- Add device filter for reporter
- Install dependency for pip installation
- Fix statistics shift in loop
- initial release