Torch.empty() does not allocate memory?

1

I found out that calling torch.empty() does not allocate memory to fit future data. Is it an intended behavior? I believed that torch.empty() works like any other construction methods, but without filling in any values.

Docs says:

Returns a tensor filled with uninitialized data.

Which is ambiguous to me.

2

torch.empty does allocate memory, but does not initialize this memory space.
I.e. the values in your tensor will just be interpreted as the bits which are currently in the allocated memory space.

3

Are you sure?

Here’s the output from the memory profiler:

Line #    Mem usage    Increment   Line Contents
================================================
     5     53.1 MiB     53.1 MiB   @profile
     6                             def main():
     7     53.2 MiB      0.1 MiB       t = torch.empty((1000000, 1000), dtype=torch.float, device='cpu')
     8     53.2 MiB      0.0 MiB       del t
     9   3868.3 MiB   3815.1 MiB       t = torch.zeros((1000000, 1000), dtype=torch.float, device='cpu')
    10     53.6 MiB      0.0 MiB       del t
4

Yes, otherwise you couldn’t index the tensor or work in some other way.
I’m not sure, how the memory profiler works, but using the process_memory_info() data you’ll get the expected result:

import torch

import os
import psutil
process = psutil.Process(os.getpid())
mem_t0 = process.memory_info().data
print(mem_t0)

x = torch.empty(1000, 1000, dtype=torch.float32)

mem_t1 =  process.memory_info().data
mem_expected = x.numel() * 4 /1024 # in kBytes
print('Expected {} kBytes'.format(mem_expected))
print('Actual {} kBytes'.format((mem_t1 - mem_t0) / 1024))
> Expected 3906.25 kBytes
> Actual 3908.0 kBytes
5

This is strange. Using the following code, I encounter a “Out Of Memory” error after a specific number of iterations.

a = torch.empty(b.shape)
for i, batch in enumerate(train_dl):
    print(i, flush=True)
    a[batch_size*i : batch_size*(i+1)] = model(batch).detach()

When I change torch.empty to torch.zeros, I get the OOM error immediately without a single iteration.

6

Based on your code snippet it seems you are initializing a on the CPU, so it’s unclear why a change from torch.empty to torch.zeros would cause an OOM on the GPU as both tensors will be allocated on the CPU by default or are you seeing a host OOM?

7

Here, I am not using a GPU. It’s a host OOM.

8

How much RAM does your system have and what does the error message tell you?