…es (#97737)

Slack thread: https://pytorch.slack.com/archives/GEEQ2K4MD/p1679962409906099

I was seeing some massive (~2x) slowdowns on a job after running it on PyTorch 2.0. From some profiling in `py-spy` it looked like the pin_memory thread was doing a lot more work than before. Looking at a trace in `nsys` I saw the thread doing the forward pass having a bunch of `pthread_cond_timedwait` with GIL reacquire calls in it’s call stack, and it seemed like the thread doing the forward pass was getting blocked (waiting for the GIL) by the pin memory thread (which was holding the GIL).

After some debugging I found out the issue. If a `bytes` was passed into `pin_memory`, previously in 1.13 (before #94709) it would short-circuit and return here
https://github.com/pytorch/pytorch/blob/d922c29a22e4bf0fba49526f7536395eb8cd66f4/torch/utils/data/_utils/pin_memory.py#L54-L55
since `bytes` was in `torch._six.string_classes`:
```
>>> from torch._six import string_classes
>>> string_classes
(<class 'str'>, <class 'bytes'>)
>>>
```

However after #94709, if a `bytes` was passed into `pin_memory` it would fall into here instead
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L68-L73
because the previous check is now doing `isinstance(data, str)` instead of `isinstance(data, (str, bytes))`!
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L56-L57

As a result, `pin_memory` gets called recursively for each element in the `bytes` leading to a ton of wasted recursion. This also explains the slowdown / GIL contention I was seeing.

This PR simply changes `isinstance(data, str)` to `isinstance(data, (str, bytes))` to match the behavior before #94709

Pull Request resolved: #97737
Approved by: https://github.com/albanD, https://github.com/NivekT

Revisit `torch._six.string_classes` (which is `(str, bytes)`) removal: `isinstance(obj, string_classes) -> isinstance(obj, str)`.

Both `str` and `bytes` are `Sequence` classes.

```python
In [1]: from typing import Sequence

In [2]: issubclass(bytes, Sequence)
Out[2]: True

In [3]: issubclass(str, Sequence)
Out[3]: True
```

Re-add `bytes` to type guards like:

```python
def is_seq(obj):
    return isinstance(obj, Sequence) and not isinstance(obj, (str, bytes))
```

Ref:

- #94709 (comment)
- #97737
- #97789
Pull Request resolved: #97863
Approved by: https://github.com/Skylion007, https://github.com/albanD

…es (pytorch#97737)

Slack thread: https://pytorch.slack.com/archives/GEEQ2K4MD/p1679962409906099

I was seeing some massive (~2x) slowdowns on a job after running it on PyTorch 2.0. From some profiling in `py-spy` it looked like the pin_memory thread was doing a lot more work than before. Looking at a trace in `nsys` I saw the thread doing the forward pass having a bunch of `pthread_cond_timedwait` with GIL reacquire calls in it’s call stack, and it seemed like the thread doing the forward pass was getting blocked (waiting for the GIL) by the pin memory thread (which was holding the GIL).

After some debugging I found out the issue. If a `bytes` was passed into `pin_memory`, previously in 1.13 (before pytorch#94709) it would short-circuit and return here
https://github.com/pytorch/pytorch/blob/d922c29a22e4bf0fba49526f7536395eb8cd66f4/torch/utils/data/_utils/pin_memory.py#L54-L55
since `bytes` was in `torch._six.string_classes`:
```
>>> from torch._six import string_classes
>>> string_classes
(<class 'str'>, <class 'bytes'>)
>>>
```

However after pytorch#94709, if a `bytes` was passed into `pin_memory` it would fall into here instead
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L68-L73
because the previous check is now doing `isinstance(data, str)` instead of `isinstance(data, (str, bytes))`!
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L56-L57

As a result, `pin_memory` gets called recursively for each element in the `bytes` leading to a ton of wasted recursion. This also explains the slowdown / GIL contention I was seeing.

This PR simply changes `isinstance(data, str)` to `isinstance(data, (str, bytes))` to match the behavior before pytorch#94709

Pull Request resolved: pytorch#97737
Approved by: https://github.com/albanD, https://github.com/NivekT

…97863)

Revisit `torch._six.string_classes` (which is `(str, bytes)`) removal: `isinstance(obj, string_classes) -> isinstance(obj, str)`.

Both `str` and `bytes` are `Sequence` classes.

```python
In [1]: from typing import Sequence

In [2]: issubclass(bytes, Sequence)
Out[2]: True

In [3]: issubclass(str, Sequence)
Out[3]: True
```

Re-add `bytes` to type guards like:

```python
def is_seq(obj):
    return isinstance(obj, Sequence) and not isinstance(obj, (str, bytes))
```

Ref:

- pytorch#94709 (comment)
- pytorch#97737
- pytorch#97789
Pull Request resolved: pytorch#97863
Approved by: https://github.com/Skylion007, https://github.com/albanD

Torch does not support torch._six anymore, so removed from torch._six import string_classes

Replaced string_classes with str.

Reference: pytorch/pytorch#94709

…es (#97737)

Slack thread: https://pytorch.slack.com/archives/GEEQ2K4MD/p1679962409906099

I was seeing some massive (~2x) slowdowns on a job after running it on PyTorch 2.0. From some profiling in `py-spy` it looked like the pin_memory thread was doing a lot more work than before. Looking at a trace in `nsys` I saw the thread doing the forward pass having a bunch of `pthread_cond_timedwait` with GIL reacquire calls in it’s call stack, and it seemed like the thread doing the forward pass was getting blocked (waiting for the GIL) by the pin memory thread (which was holding the GIL).

After some debugging I found out the issue. If a `bytes` was passed into `pin_memory`, previously in 1.13 (before #94709) it would short-circuit and return here
https://github.com/pytorch/pytorch/blob/d922c29a22e4bf0fba49526f7536395eb8cd66f4/torch/utils/data/_utils/pin_memory.py#L54-L55
since `bytes` was in `torch._six.string_classes`:
```
>>> from torch._six import string_classes
>>> string_classes
(<class 'str'>, <class 'bytes'>)
>>>
```

However after #94709, if a `bytes` was passed into `pin_memory` it would fall into here instead
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L68-L73
because the previous check is now doing `isinstance(data, str)` instead of `isinstance(data, (str, bytes))`!
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L56-L57

As a result, `pin_memory` gets called recursively for each element in the `bytes` leading to a ton of wasted recursion. This also explains the slowdown / GIL contention I was seeing.

This PR simply changes `isinstance(data, str)` to `isinstance(data, (str, bytes))` to match the behavior before #94709

Pull Request resolved: #97737
Approved by: https://github.com/albanD, https://github.com/NivekT

…97789, #97863) (#98055)

* [DataLoader] Short circuit pin_memory recursion when operating on bytes (#97737)

Slack thread: https://pytorch.slack.com/archives/GEEQ2K4MD/p1679962409906099

I was seeing some massive (~2x) slowdowns on a job after running it on PyTorch 2.0. From some profiling in `py-spy` it looked like the pin_memory thread was doing a lot more work than before. Looking at a trace in `nsys` I saw the thread doing the forward pass having a bunch of `pthread_cond_timedwait` with GIL reacquire calls in it’s call stack, and it seemed like the thread doing the forward pass was getting blocked (waiting for the GIL) by the pin memory thread (which was holding the GIL).

After some debugging I found out the issue. If a `bytes` was passed into `pin_memory`, previously in 1.13 (before #94709) it would short-circuit and return here
https://github.com/pytorch/pytorch/blob/d922c29a22e4bf0fba49526f7536395eb8cd66f4/torch/utils/data/_utils/pin_memory.py#L54-L55
since `bytes` was in `torch._six.string_classes`:
```
>>> from torch._six import string_classes
>>> string_classes
(<class 'str'>, <class 'bytes'>)
>>>
```

However after #94709, if a `bytes` was passed into `pin_memory` it would fall into here instead
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L68-L73
because the previous check is now doing `isinstance(data, str)` instead of `isinstance(data, (str, bytes))`!
https://github.com/pytorch/pytorch/blob/c263bd43e8e8502d4726643bc6fd046f0130ac0e/torch/utils/data/_utils/pin_memory.py#L56-L57

As a result, `pin_memory` gets called recursively for each element in the `bytes` leading to a ton of wasted recursion. This also explains the slowdown / GIL contention I was seeing.

This PR simply changes `isinstance(data, str)` to `isinstance(data, (str, bytes))` to match the behavior before #94709

Pull Request resolved: #97737
Approved by: https://github.com/albanD, https://github.com/NivekT

* [DataLoader] Fix  collation logic (#97789)

Similar to #97737, a previous auto-refactor changed how `bytes` are handled during collation, which can potentially lead to performance regression. This PR undoes that.
Pull Request resolved: #97789
Approved by: https://github.com/albanD

* Revisit `torch._six.string_classes` removal (#94709) (#97863)

Revisit `torch._six.string_classes` (which is `(str, bytes)`) removal: `isinstance(obj, string_classes) -> isinstance(obj, str)`.

Both `str` and `bytes` are `Sequence` classes.

```python
In [1]: from typing import Sequence

In [2]: issubclass(bytes, Sequence)
Out[2]: True

In [3]: issubclass(str, Sequence)
Out[3]: True
```

Re-add `bytes` to type guards like:

```python
def is_seq(obj):
    return isinstance(obj, Sequence) and not isinstance(obj, (str, bytes))
```

Ref:

- #94709 (comment)
- #97737
- #97789
Pull Request resolved: #97863
Approved by: https://github.com/Skylion007, https://github.com/albanD

---------

Co-authored-by: Eric Zhang <[email protected]>
Co-authored-by: Kevin Tse <[email protected]>

This commit fixes a deprecated import statement in the codebase that uses the string_classes module from PyTorch. The deprecated import statement from torch import string_classes has been replaced with a simpler and more Pythonic alternative string_classes = str.

The string_classes module was previously used to define the type of a string, but it has been removed in more recent versions of PyTorch. This commit provides a cleaner and more compatible solution that works with newer versions of PyTorch.

See pytorch/pytorch#94709

See pytorch/pytorch#94709 (comment)

See pytorch/pytorch#94709 (comment)

See pytorch/pytorch#94709 (comment)