Fixed type hints for CosineAnnealingWarmRestarts:
- `T_mult` is not `Optional[int]` but just `int`
- `eta_min` is not `Optional[float]` but just `float`
- removed `step` method specific annotation as it is compatible with the base class
e132f09e88/torch/optim/lr_scheduler.py (L1365-L1375)
Otherwise, computation like this `self.T_i * self.T_mult` in `self.step` is not possible:
```
error: Unsupported operand types for * ("int" and "None")
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102067
Approved by: https://github.com/janeyx99
This PR proposes an optimized way to do Exponential Moving Average (EMA), which is faster than the current way using `swa_utils.AveragedModel` described in https://pytorch.org/docs/stable/optim.html#custom-averaging-strategies.
This implementation is asynchronous, and is built as an optimizer wrapper so that the EMA weight update happens without any additional CPU/GPU sync, just after optimizer steps, and with limited code changes.
Example usage:
```
model = Model().to(device)
opt = torch.optim.Adam(model.parameters())
opt = EMAOptimizer(opt, device, 0.9999)
for epoch in range(epochs):
training_loop(model, opt)
regular_eval_accuracy = evaluate(model)
with opt.swap_ema_weights():
ema_eval_accuracy = evaluate(model)
```
Here are some benchmarks (time per iteration) on various torchvision models:
|model|this PR iteration time |swa_utils.AveragedModel iteration time| iteration speedup |
|-----|-----------------------------|-----------------------|---------------------------------------------|
| | | | |
|regnet_x_1_6gf|62.73 |67.998 |1.08 |
|regnet_x_3_2gf|101.75 |109.422 |1.08 |
|regnet_x_400mf|25.13 |32.005 |1.27 |
|regnet_x_800mf|33.01 |37.466 |1.13 |
|regnet_x_8gf|128.13 |134.868 |1.05 |
|regnet_y_16gf|252.91 |261.292 |1.03 |
|regnet_y_1_6gf|72.14 |84.22 |1.17 |
|regnet_y_3_2gf|99.99 |109.296 |1.09 |
|regnet_y_400mf|29.53 |36.506 |1.24 |
|regnet_y_800mf|37.82 |43.634 |1.15 |
|regnet_y_8gf|196.63 |203.317 |1.03 |
|resnet101|128.80 |137.434 |1.07 |
|resnet152|182.85 |196.498 |1.07 |
|resnet18|29.06 |29.975 |1.03 |
|resnet34|50.73 |53.443 |1.05 |
|resnet50|76.88 |80.602 |1.05 |
|resnext101_32x8d|277.29 |280.759 |1.01 |
|resnext101_64x4d|269.56 |281.052 |1.04 |
|resnext50_32x4d|100.73 |101.102 |1.00 |
|shufflenet_v2_x0_5|10.56 |15.419 |1.46 |
|shufflenet_v2_x1_0|13.11 |18.525 |1.41 |
|shufflenet_v2_x1_5|18.05 |23.132 |1.28 |
|shufflenet_v2_x2_0|25.04 |30.008 |1.20 |
|squeezenet1_1|14.26 |14.325 |1.00 |
|swin_b|264.52 |274.613 |1.04 |
|swin_s|180.66 |188.914 |1.05 |
|swin_t|108.62 |112.632 |1.04 |
|swin_v2_s|220.29 |231.153 |1.05 |
|swin_v2_t|127.27 |133.586 |1.05 |
|vgg11|95.52 |103.714 |1.09 |
|vgg11_bn|106.49 |120.711 |1.13 |
|vgg13|132.94 |147.063 |1.11 |
|vgg13_bn|149.73 |165.256 |1.10 |
|vgg16|158.19 |172.865 |1.09 |
|vgg16_bn|177.04 |192.888 |1.09 |
|vgg19|184.76 |194.194 |1.05 |
|vgg19_bn|203.30 |213.334 |1.05 |
|vit_b_16|217.31 |219.748 |1.01 |
|vit_b_32|69.47 |75.692 |1.09 |
|vit_l_32|223.20 |258.487 |1.16 |
|wide_resnet101_2|267.38 |279.836 |1.05 |
|wide_resnet50_2|145.06 |154.918 |1.07 |
You can see that in all cases it is faster than using `AveragedModel`. In fact in many cases, adding EMA does not add any overhead since the computation is hidden behind the usual iteration flow.
This is a similar implementation to the one currently in [NVIDIA NeMo](https://github.com/NVIDIA/NeMo).
If the team is interested in merging this, let me know and I'll add some documentation similar to `swa_utils` and tests.
Credits to @szmigacz for the implementation.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94820
Approved by: https://github.com/janeyx99
Fixes#95781.
The cause seems to be that the current implementation doesn't correctly pass `found_inf` when `grad_scale` is `None`. Therefore parameters can get mistakenly updated by gradients whose some elements are invalid, i.e. nan or inf.
Related #94060
I forgot about this wrong handling after #94344
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95847
Approved by: https://github.com/janeyx99
Big OOP correction continued. Also added a test this time to verify the defaulting was as expected.
The key here is realizing that the grouping for foreach already assumes that the non-param tensorlists follow suit in dtype and device, so it is too narrow to check that _all_ tensors were on CUDA. The main leeway this allowed was state_steps, which are sometimes cpu tensors. Since foreach _can_ handle cpu tensors, this should not introduce breakage.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95820
Approved by: https://github.com/albanD
Changes:
- #95200
1. Recognize `.py.in` and `.pyi.in` files as Python in VS Code for a better development experience.
2. Fix deep setting merge in `tools/vscode_settings.py`.
- #95267
3. Use `Namedtuple` rather than `namedtuple + __annotations__` for `torch.nn.utils.rnn.PackedSequence_`:
`namedtuple + __annotations__`:
```python
PackedSequence_ = namedtuple('PackedSequence_',
['data', 'batch_sizes', 'sorted_indices', 'unsorted_indices'])
# type annotation for PackedSequence_ to make it compatible with TorchScript
PackedSequence_.__annotations__ = {'data': torch.Tensor, 'batch_sizes': torch.Tensor,
'sorted_indices': Optional[torch.Tensor],
'unsorted_indices': Optional[torch.Tensor]}
```
`Namedtuple`: Python 3.6+
```python
class PackedSequence_(NamedTuple):
data: torch.Tensor
batch_sizes: torch.Tensor
sorted_indices: Optional[torch.Tensor]
unsorted_indices: Optional[torch.Tensor]
```
- => this PR: #95268
4. Sort import statements and remove unnecessary imports in `.pyi`, `.pyi.in` files.
5. Format `.pyi`, `.pyi.in` files and remove unnecessary ellipsis `...` in type stubs.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95268
Approved by: https://github.com/huydhn
Rolling back the default change for Adam and rectifying the docs to reflect that AdamW never defaulted to fused.
Since our fused implementations are relatively newer, let's give them a longer bake-in time before flipping the switch for every user.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95241
Approved by: https://github.com/ngimel
Optimize unnecessary collection cast calls, unnecessary calls to list, tuple, and dict, and simplify calls to the sorted builtin. This should strictly improve speed and improve readability.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94323
Approved by: https://github.com/albanD
This allows it so that ONLY when the users don't set anything for foreach or fused do we switch the default and cascades adam so that we default to fused, then foreach, then single-tensor.
To clarify:
* if the user puts True in foreach _only_, it will run the foreach implementation.
* if the user puts True in fused _only_, it will run the fused implementation.
* if the user puts True in foreach AND for fused, it will run the fused implementation.
And:
* if the user puts False in foreach _only_, it will run the single tensor implementation.
* if the user puts False in fused _only_, it will still run the single tensor implementation.
* if the user puts False in foreach AND for fused, it will run the single tensor implementation.
I also didn't trust myself that much with the helper function, so I ran some local asserts on _default_to_fused_or_foreach. The only point left to really test is the type(p) -- torch.Tensor but I think the distributed tests will catch that in CI.
```
cuda_only_fp_list = [
torch.rand((1, 2), device="cuda", dtype=torch.float32),
torch.rand((1, 2), device="cuda", dtype=torch.float64),
torch.rand((1, 2), device="cuda", dtype=torch.float16),
torch.rand((1, 2), device="cuda", dtype=torch.bfloat16),
]
cuda_only_int_list = [
torch.randint(1024, (1, 2), device="cuda", dtype=torch.int64),
]
cpu_list = [
torch.rand((1, 2), device="cpu", dtype=torch.float32),
torch.rand((1, 2), device="cpu", dtype=torch.float64),
torch.rand((1, 2), device="cpu", dtype=torch.float16),
]
none_list = [None]
# differentiable should always make it return false for both
assert _default_to_fused_or_foreach([cuda_only_fp_list], True, True) == (False, False)
assert _default_to_fused_or_foreach([cuda_only_fp_list], True, False) == (False, False)
# cpu lists should always make it return false for both
assert _default_to_fused_or_foreach([cuda_only_fp_list, cpu_list], False, True) == (False, False)
assert _default_to_fused_or_foreach([cpu_list], False, True) == (False, False)
assert _default_to_fused_or_foreach([cuda_only_fp_list, cpu_list], False, False) == (False, False)
assert _default_to_fused_or_foreach([cpu_list], False, False) == (False, False)
# has fused triggers correctly
assert _default_to_fused_or_foreach([cuda_only_fp_list], False, True) == (True, False)
assert _default_to_fused_or_foreach([cuda_only_fp_list], False, False) == (False, True)
# ints always goes to foreach
assert _default_to_fused_or_foreach([cuda_only_fp_list, cuda_only_int_list], False, True) == (False, True)
assert _default_to_fused_or_foreach([cuda_only_fp_list, cuda_only_int_list], False, False) == (False, True)
# Nones don't error
assert _default_to_fused_or_foreach([cuda_only_fp_list, none_list], False, True) == (True, False)
assert _default_to_fused_or_foreach([cuda_only_fp_list, cuda_only_int_list, none_list], False, True) == (False, True)
assert _default_to_fused_or_foreach([none_list], False, True) == (True, False)
assert _default_to_fused_or_foreach([none_list], False, False) == (False, True)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/93184
Approved by: https://github.com/albanD
Attempts to fix#92656
BC-breaking! This changes the default of zero_grad in optim and in nn to default set grads to None instead of zero tensors. We are changing the default because there are proven perf wins and existing code has typically not regressed due to this change. (will probably have to flesh out this note more).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/92731
Approved by: https://github.com/ngimel
Old behavior would have adadelta foreach sending tensors to the slow path if they were not all the same dtype nor on the same device.
This PR adds grouping for adadelta optimizer so that it would run foreach in batches, allowing more users to benefit from foreach perf.
Of course, we should ensure that the new implementation works, so there are new tests to ensure this behavior is not broken.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/92048
Approved by: https://github.com/albanD
This PR adds the `_profile_using_dynolog` function to `torch/__init__.py`. The `_profile_using_dynolog` method allows registering the optimizer step post hook. This is required to collect iteration based traces using dynolog.
Other related changes for tests to pass:
1. Updated `optimizer.pyi`
1. Updated `overrides.py`
1. The test `test_kineto_profiler_multiple_steppers` in `test_profiler.py` has been broken down into two cases:
- `test_kineto_profiler_multiple_steppers_with_override_True` : this test uses the override argument
- `test_kineto_profiler_multiple_steppers_with_override_False` : this test uses the environment variable
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90101
Approved by: https://github.com/albanD
@mlazos: skips `item()` calls if compiling with dynamo, by defining a helper function `_get_value` which either returns the result of `.item()` or the scalar cpu tensor if compiling with dynamo. This was done because removing `item()` calls significantly regresses eager perf. Additionally, `_dispatch_sqrt` calls the appropriate sqrt function (math.sqrt, or torch.sqrt).
Fixes https://github.com/pytorch/torchdynamo/issues/1083
This PR will no longer be needed once symint support is default.
This PR closes all remaining graph breaks in the optimizers (!!)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88173
Approved by: https://github.com/albanD
Doing some tests with all Optimizer and LRScheduler classes in optim package, I noticed a couple of mistakes in type annotations, so created a pull request to fix them.
- In Optimizer class, incorrectly named parameter `default` instead of `defaults` in pyi file
- In SGD class, type for `maximize` and `differentiable` not available in either py or pyi files
I don't know if there is a plan to move all types from pyi to py files, so wasn't too sure where to fix what.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90216
Approved by: https://github.com/janeyx99
Fixes#84053
As described in the issue, the AveragedModel will deep copy the model during initialization, which means that the buffers in the averaged model cannot be updated together with the model.
One solution is to make the buffers equal to the source model every time when calling `update_parameters`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84054
Approved by: https://github.com/samdow
Hi, we noticed in our team that by using CyclicLR, there is a problem with memory clearance on GPU (probably it will be the case without the GPU as well, but that was our use case) After initializing CyclicLR, GPU memory is not cleared even after the model, optimizer and scheduler are out of scope (e.g. reference count is zero). This is because `__init__` method inside `CyclicLR` creates reference to its own methods and it will not get removed until `gc.collect()` is called manually. This is a problem if people want to test multiple models in one run of a script, after testing the first model, second one will fail on `CUDA out of memory error` because the first one is not cleared from the memory.
I propose a simple fix by using `weakref`, similarly as in `_LRScheduler` base class, but if you have any comments I am happy to change it.
Here is the code to reproduce the bug:
```
import torch
import weakref
from transformers import DetrForObjectDetection
class X:
def __init__(self, optimizer):
self.optimizer = optimizer
# Will cause cyclic reference.
self.func = self.dummy
# Will work as expected, memory cleared after instance count is zero.
# self.func = weakref.WeakMethod(self.dummy)
def dummy(self, x):
return 1.
def test():
model = DetrForObjectDetection.from_pretrained('facebook/detr-resnet-50')
model.to('cuda')
optimizer = torch.optim.Adam(model.parameters())
x = X(optimizer)
test()
print(f'{torch.cuda.memory_reserved()}, {torch.cuda.memory_allocated()}') # Should print (<some memory>, 0), but with cyclic reference, it will print (<some memory>, <some memory>).
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85462
Approved by: https://github.com/albanD
Hello there 👋
As discussed in #84485, this PR enables more flexibility on the optimizers that are wrapped by LR schedulers in PyTorch. Currently, it is incompatible with optimizers that have a number of betas different than 2. This PR fixes that with minimal modifications.
Fixes#84485
Any feedback is welcome!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84486
Approved by: https://github.com/Lezcano, https://github.com/soulitzer
This is a new version of #15648 based on the latest master branch.
Unlike the previous PR where I fixed a lot of the doctests in addition to integrating xdoctest, I'm going to reduce the scope here. I'm simply going to integrate xdoctest, and then I'm going to mark all of the failing tests as "SKIP". This will let xdoctest run on the dashboards, provide some value, and still let the dashboards pass. I'll leave fixing the doctests themselves to another PR.
In my initial commit, I do the bare minimum to get something running with failing dashboards. The few tests that I marked as skip are causing segfaults. Running xdoctest results in 293 failed, 201 passed tests. The next commits will be to disable those tests. (unfortunately I don't have a tool that will insert the `#xdoctest: +SKIP` directive over every failing test, so I'm going to do this mostly manually.)
Fixes https://github.com/pytorch/pytorch/issues/71105
@ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82797
Approved by: https://github.com/ezyang
### Description
Across PyTorch's docstrings, both `callable` and `Callable` for variable types. The Callable should be capitalized as we are referring to the `Callable` type, and not the Python `callable()` function.
### Testing
There shouldn't be any testing required.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82487
Approved by: https://github.com/albanD
### Description
PR #80336 introduced a new parameter to the Sparse Adam optimizer. The new parameter is accessed inside the `step` method of the optimizer. If we try to deserialize and run an older version of the optimizer before this change was introduced, it fails in the step that tries to access the missing parameter.
I have added a workaround to set a default value in case the parameter is unavailable in the optimizer.
### Issue
<!-- Link to Issue ticket or RFP -->
### Testing
* Testing on PyTorch CI
* Manual validation against existing serialized models to make sure they continue to work
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82273
Approved by: https://github.com/mehtanirav, https://github.com/albanD
Adds the `differentiable` argument, a method for updating parameters in an existing optimizer, and a template for testing the differentiability of multiple optimizers.
This is all based in discussions with @albanD & @jbschlosser
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80938
Approved by: https://github.com/albanD
Generator comprehensions with any/all are less verbose and potentially help to save memory/CPU : https://eklitzke.org/generator-comprehensions-and-using-any-and-all-in-python
To make JIT work with this change, I added code to convert GeneratorExp to ListComp. So the whole PR is basically NoOp for JIT, but potentially memory and speed improvement for eager mode.
Also I removed a test from test/jit/test_parametrization.py. The test was bad and had a TODO to actually implement and just tested that UnsupportedNodeError is thrown, and with GeneratorExp support a different error would be thrown.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/78142
Approved by: https://github.com/malfet, https://github.com/albanD
This is causing issues if the user has the step on cuda for a good reason.
These assert prevents code that used to run just fine to fail.
Note that this is a pretty bad thing to do for performance though so it is ok to try and push users away from doing it.
For the 1.12.1 milestone: this is not asking for a dot release to fix this (as this is bad practice anyways). But it would be a great thing to add if we do one: it is very low risk and will prevent breakage for users.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80222
Approved by: https://github.com/jbschlosser, https://github.com/ngimel
What was happening is that when we have multiple learning rate schedulers, the order in which they are being initialized is not being taken into account. This is a problem if they were being initialized in sequential order (as one might intuitively do).
Each scheduler calls `step()` on initialization and sets the `lr` in its optimizer's `params_groups`. However, this means that step 0 will be using the `lr` that was set by the very last scheduler (in the case of initializing schedulers sequentially) instead of the first scheduler.
The fix in this PR, addresses the above bug by performing a call to the appropriate scheduler on initialization after decrementing the `last_epoch` values in order to keep them the same post-step. This will ensure that the correct scheduler is the one setting the `lr` values for the optimizer's `param_groups`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72856
Approved by: https://github.com/jbschlosser
### Goal
Fixes https://github.com/pytorch/pytorch/issues/79720
### Approach
replace `Chains list of learning rate schedulers. It takes a list of chainable learning rate schedulers and performs consecutive step() functions` **`belong`** `to them by just one call.` with `Chains list of learning rate schedulers. It takes a list of chainable learning rate schedulers and performs consecutive step() functions` **`belonging`** `to them by just one call.`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/79775
Approved by: https://github.com/albanD
Near term fix for https://github.com/pytorch/pytorch/issues/76368.
Q. Why does the user need to request `capturable=True` in the optimizer constructor? Why can't capture safety be completely automatic?
A. We need to set up capture-safe (device-side) state variables before capture. If we don't, and step() internally detects capture is underway, it's too late: the best we could do is create a device state variable and copy the current CPU value into it, which is not something we want baked into the graph.
Q. Ok, why not just do the capture-safe approach with device-side state variables all the time?
A. It incurs several more kernel launches per parameter, which could really add up and regress cpu overhead for ungraphed step()s. If the optimizer won't be captured, we should allow step() to stick with its current cpu-side state handling.
Q. But cuda RNG is a stateful thing that maintains its state on the cpu outside of capture and replay, and we capture it automatically. Why can't we do the same thing here?
A. The graph object can handle RNG generator increments because its capture_begin, capture_end, and replay() methods can see and access generator object. But the graph object has no explicit knowledge of or access to optimizer steps in its capture scope. We could let the user tell the graph object what optimizers will be stepped in its scope, ie something like
```python
graph.will_use_optimizer(opt)
graph.capture_begin()
...
```
but that seems clunkier than an optimizer constructor arg.
I'm open to other ideas, but right now I think constructor arg is necessary and the least bad approach.
Long term, https://github.com/pytorch/pytorch/issues/71274 is a better fix.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77862
Approved by: https://github.com/ezyang
Fixes#60265
The initial LR for this scheduler is not consistent when a new instance is created with `last_epoch != -1`
Maybe we can refactor the testing code to test `last_epoch != -1` in schedulers that can recreate their state from the current epoch?
Pull Request resolved: https://github.com/pytorch/pytorch/pull/60339
Approved by: https://github.com/albanD
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73215
Fixing an issue in optimizers from _multi_tensor, for `sgd_mt` introduced in 2cb03e926f
Reviewed By: mikaylagawarecki
Differential Revision: D34389034
fbshipit-source-id: ede153d52dca15909c6c022853589707f18dc8d1
(cherry picked from commit cc8a58e584)
