A recent PR deprecated `torch.testing.assert_allclose` in favor of `torch.testing.assert_close` and left a `TODO`. This PR follows up to confirm that we do intend to have `check_dtype=False`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90251
Approved by: https://github.com/rohan-varma
`FSDP.clip_grad_norm_()` is tested separately in `test_fsdp_clip_grad_norm.py`. This PR removes the dead non-run code from `common_fsdp.py` and `test_fsdp_core.py` related to `clip_grad_norm_()`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90250
Approved by: https://github.com/rohan-varma
Observed by @aazzolini, some op might have Optional[Tensor] returns
where it return None (i.e. native_layer_norm_backward), it's a mismatch
between C++ aten op signature and python None, but we need to handle it
in the python side
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90241
Approved by: https://github.com/aazzolini
**Summary**
Post op fusion can reduce data movement overhead and improve inference performance. This PR adds fused `linear-leaky_relu` op for `onednn` backend, which will be used for int8 inference with `onednn` backend. Cannot call this op with other quantization backends otherwise an error is thrown.
**Test Plan**
python test_quantization.py TestQuantizedLinear
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88478
Approved by: https://github.com/jgong5, https://github.com/jerryzh168
- Add `full` nvprim to support factory functions because the full reference uses `empty` and `fill` while we have a full factory function.
- Change `full_like` reference to call `full` to avoid defining another nvprim.
- Enable support for new_zeros to enable `cudnn_batch_norm` decomposition.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/89230
Approved by: https://github.com/kevinstephano, https://github.com/mruberry
I need to rebase later after Shen's PRs land.
The idea is to only register the pre/post-forward hook on the _root modules_ among the modules that consume a `FlatParameter`. (Yes, the term _root module_ is heavily overloaded. We may want to clarify that at some point. Here, _root_ is being used in the graph sense, meaning parent-less, and the scope is only among the modules consuming a `FlatParameter`.)
This avoids unnecessary pre/post-forward hooks running, which would lead to errors because the unshard is not truly idempotent.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90201
Approved by: https://github.com/mrshenli, https://github.com/rohan-varma
This PR get rids of torchgen FunctionSchema parsing and parse
it manually, it should resolve torchgen package issue and also
provide some perf wins when running DTensor eagerly
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90106
Approved by: https://github.com/awgu
In pytorch, the optim state_dict will always use number to index optimizer state_dict for parameters.
Now composability workstream need a FQN based way to index optimizer state_dict for parameters..
For example, SGD optimizer might have something in its `state_dict` like:
```
{'state':
{0:
{'momentum_buffer': tensor(...)},
{1:
{'momentum_buffer': tensor(...)},
...
}
'param_groups':
[{'lr': 0.001, 'momentum': 0.9, 'dampening': 0, 'weight_decay': 0, 'nesterov': False, 'maximize': False, 'foreach': None, 'differentiable': False, 'params': [0, 1, 2, 3, 4, 5, 6, 7]}]
}
```
And in NamedOptimizer we want the `state_dict` can be:
```
{'state':
{'net1.0.weight':
{'momentum_buffer': tensor(...)},
{'net1.0.bias':
{'momentum_buffer': tensor(...)},
...
}
'param_groups':
[{'lr': 0.001, 'momentum': 0.9, 'dampening': 0, 'weight_decay': 0, 'nesterov': False, 'maximize': False, 'foreach': None, 'differentiable': False, 'params': ['net1.0.weight', 'net1.0.bias', 'net2.0.weight', 'net2.0.bias', 'net3.weight', 'net3.bias', 'net4.1.weight', 'net4.1.bias']}]
}
```
We also want to support load_state_dict to enable optim `state_dict` override for NameOptimizer.
For the next couple PR/diffs, we also need to:
1. To make `NamedOptimizer` working with FSDP (like registering a hook for model wrapped with FSDP) and other PTD/PT components.
2. Make `NamedOptimizer` works well with apply_optim_in_backward
3. Upstream also `CombinedOptimizer`.
Differential Revision: [D41432088](https://our.internmc.facebook.com/intern/diff/D41432088/)
**NOTE FOR REVIEWERS**: This PR has internal Meta-specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D41432088/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/89480
Approved by: https://github.com/rohan-varma
The current cond implementation is silently incorrect when
there are outstanding side effects, since the locally tracked
side effects are lost when the recursive export call is made.
At least we raise an assert now.
I'm working on a refactor of cond which should be able to sidestep
this problem. Maybe.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Differential Revision: [D41746973](https://our.internmc.facebook.com/intern/diff/D41746973)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90208
Approved by: https://github.com/voznesenskym
Not sure why, but top-level `using namespace` directive causes VC++ fail with (if C++17 standard is used, but everything is fine with C++14):
```
C:\actions-runner\_work\pytorch\pytorch\third_party\pybind11\include\pybind11\detail\../pytypes.h(1520): error C2872: 'attr': ambiguous symbol
C:\actions-runner\_work\pytorch\pytorch\aten\src\ATen/core/interned_strings.h(349): note: could be 'c10::attr'
C:\actions-runner\_work\pytorch\pytorch\torch/csrc/jit/ir/ir.h(75): note: or 'torch::jit::attr'
C:\actions-runner\_work\pytorch\pytorch\cmake\..\third_party\pybind11\include\pybind11/pybind11.h(1094): note: see reference to function template instantiation 'pybind11::str pybind11::str::format<_Ty1&>(_Ty1 &) const' being compiled
with
[
_Ty1=pybind11::handle
]
```
Solve this by replacing global `using namespace torch::jit;` with
specific usages of objects/methods from namespaces
Another prep change for https://github.com/pytorch/pytorch/70188
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90228
Approved by: https://github.com/kit1980, https://github.com/albanD
