Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/64430
The functionalization pass needs `{view}_scatter` versions of the slice/select/diagonal ops in order to correctly propagate mutations from a view to its base. On top of that, the implementations need to be primitive w.r.t. autograd, because they look something like `...slice().copy_()`, and the functionalization pass can't use views + mutations inside of it's own alias-removal machinery!
I added some basic tests that I tried to base off of existing tests for views (particularly around testing the derivative formulas), but I'm wondering if I should add something more comprehensive.
Also, as_strided fits into this category - the functionalization pass will need an `as_strided_scatter` op that's primitive w.r.t. autograd. I didn't add it for now, because it'll involve duplicating a bunch of logic from the current `as_strided_backward()` function, and also writing a derivative formula that I wasn't sure how to write :)
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D31942092
Pulled By: bdhirsh
fbshipit-source-id: c702a57c2748a7c771c14e4bcc3e996b48fcc4c8
Summary:
Adds mixed precision autocasting support between fp32/fp16 to torchscript/JIT. More in depth descriptoin can be found at [torch/csrc/jit/JIT-AUTOCAST.md](https://github.com/pytorch/pytorch/pull/63939/files#diff-1f1772aaa508841c5bb58b74ab98f49a1e577612cd9ea5c386c8714a75db830b)
This PR implemented an autocast optimization pass that inserts casting ops per AMP rule (torch/csrc/jit/passes/autocast.cpp), that mimics the behavior of eager autocast. The pass also takes into consideration the context of `torch.cuda.amp.autocast` and only inserts casting ops within the enabled context manager, giving feature parity as with eager amp autocast.
We currently provide JIT AMP autocast as a prototyping feature, so it is default off and could be turned on via `torch._C._jit_set_autocast_mode(True)`
The JIT support for autocast is subject to different constraints compared to the eager mode implementation (mostly related to the fact that TorchScript is statically typed), restriction on the user facing python code is described in doc torch/csrc/jit/JIT-AUTOCAST.md
This is a prototype, there are also implementation limitation that's necessary to keep this PR small and get something functioning quickly on upstream, so we can iterate on designs.
Few limitation/challenge that is not properly resolved in this PR:
1. Autocast inserts cast operation, which would have impact on scalar type of output tensor feeding downstream operations. We are not currently propagating the updated scalar types, this would give issues/wrong results on operations in promotion rules.
2. Backward for autodiff in JIT misses the casting of dgrad to input scalar type, as what autograd does in eager. This forces us to explicitly mark the casting operation for certain operations (e.g. binary ops), otherwise, we might be feeding dgrad with mismatch scalar type to input. This could potentially break gradient function consuming dgrad. (e.g. gemm backwards, which assumes grad_output to be of same scalar type as input')
3. `torch.autocast` api has an optional argument `dtype` which is not currently supported in the JIT autocast and we require a static value.
Credit goes mostly to:
tlemo
kevinstephano
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63939
Reviewed By: navahgar
Differential Revision: D31093381
Pulled By: eellison
fbshipit-source-id: da6e26c668c38b01e296f304507048d6c1794314
Summary:
Adds `torch.argwhere` as an alias to `torch.nonzero`
Currently, `torch.nonzero` is actually provides equivalent functionality to `np.argwhere`.
From NumPy docs,
> np.argwhere(a) is almost the same as np.transpose(np.nonzero(a)), but produces a result of the correct shape for a 0D array.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/64257
Reviewed By: dagitses
Differential Revision: D31474901
Pulled By: saketh-are
fbshipit-source-id: 335327a4986fa327da74e1fb8624cc1e56959c70
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62030
Remove dtype tracking from Python Storage interface, remove all the different `<type>Storage` classes except for `ByteStorage`, and update serialization accordingly, while maintaining as much FC/BC as possible
Fixes https://github.com/pytorch/pytorch/issues/47442
* **THE SERIALIZATION FORMAT IS FULLY FC/BC.** We worked very hard to make sure this is the case. We will probably want to break FC at some point to make the serialization structure of tensors make more sense, but not today.
* There is now only a single torch.ByteStorage class. Methods like `Tensor.set_` no longer check that the dtype of storage is appropriate.
* As we no longer know what dtype of a storage is, we've **removed** the size method from Storage, replacing it with nbytes. This is to help catch otherwise silent errors where you confuse number of elements with number of bytes.
* `Storage._new_shared` takes a `nbytes` kwarg and will reject previous positional only calls. `Storage._new_with_file` and `_set_from_file` require explicit element size arguments.
* It's no longer possible to convert storages to different types using the float/double/etc methods. Instead, do the conversion using a tensor.
* It's no longer possible to allocate a typed storage directly using FloatStorage/DoubleStorage/etc constructors. Instead, construct a tensor and extract its storage. The classes still exist but they are used purely for unpickling.
* The preexisting serialization format stores dtype with storage, and in fact this dtype is used to determine the dtype of the tensor overall.
To accommodate this case, we introduce a new TypedStorage concept that exists only during unpickling time which is used to temporarily store the dtype so we can construct a tensor. **If you overrode the handling of pickling/unpickling, you MUST add handling for TypedStorage** or your serialization code will degrade to standard file-based serialization.
Original pull request: https://github.com/pytorch/pytorch/pull/59671
Reviewed By: soulitzer, ngimel
Differential Revision: D29466819
Pulled By: ezyang
fbshipit-source-id: 4a14e5d3c2b08e06e558683d97f7378a3180b00e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65621
Add a new attribute to the FusedMovingAvgObsFakeQuantize that controls if the Fake Quant operation should be applied at the output of a particular layer. The motivation is to give the users additional control to control the numerics of the fake_quant operators during training. It defaults to always fake quant the output (True).
Note: We will still observer the tensors as before (only the fake_quant operation is controlled using this flag)
For example
```
input model
x -> fc1 -> fc2 -> non_quantizable_op -> fc3
After fake_quant
x -> fake_quant(x) -> fc1 -> fake_quant(fc1) -> fc2 -> fake_quant(fc2) -> non_quantizable_op -> fake_quant() -> fc3 -> fake_quantize(fc3)
With output_fake_quant disabled at the output of fc2 and fc3 (since their outputs are non-quantizable)
x -> fake_quant(x) -> fc1 -> fake_quant(fc1) -> fc2 -> non_quantizable_op -> fake_quant() -> fc3
```
Test Plan: ./buck-out/gen/caffe2/test/quantization_fx\#binary.par -r test_disable_output_fake_quant
Reviewed By: jerryzh168
Differential Revision: D31174526
fbshipit-source-id: bffe776216d041fb09133a6fb09bfc2c0bb46b89
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65340
I thought about a few possible ways of doing this. The main hazard is
that if I create a CPU tensor that doesn't have any real storage, the
moment I actually try to access the data on the tensor I will segfault.
So I don't want to use _make_subclass on a "cpu meta tensor" because
the CPU meta tensor (with no subclass) is radioactive: printing it
will immediately cause a segfault. So instead, I have to create
the CPU meta tensor AND subclass all in one go, and that means I need
another function for it. One downside to doing it this way is
I need another overload for explicit strides, and in general it is
difficult to get the view relationships to all work out properly;
tracked at https://github.com/pytorch/pytorch/issues/65339
Fixes https://github.com/pytorch/pytorch/issues/62972
Fixes https://github.com/pytorch/pytorch/issues/62730
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D31057231
Pulled By: ezyang
fbshipit-source-id: 73522769e093ae8a1bf0c7f7e594659bfb827b28
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62671
Very crude first implementation of `torch.nanmean`. The current reduction kernels do not have good support for implementing nan* variants. Rather than implementing new kernels for each nan* operator, I will work on new reduction kernels with support for a `nan_policy` flag and then I will port `nanmean` to use that.
**TODO**
- [x] Fix autograd issue
Test Plan: Imported from OSS
Reviewed By: malfet
Differential Revision: D30515181
Pulled By: heitorschueroff
fbshipit-source-id: 303004ebd7ac9cf963dc4f8e2553eaded5f013f0
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/64689
This brings it in line with the C++ implementation.
Fixes https://github.com/pytorch/pytorch/issues/64687
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D30816215
Pulled By: ezyang
fbshipit-source-id: ed36af6c35467ae678d9548197efd97c36d38dec
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63552
In this PR, we want to exclude these 2 cases in the `Autocast` weight cache usages:
- Using `torch.jit.trace` under the `Autocast`
As report in https://github.com/pytorch/pytorch/issues/50231 and several other discussions, using `torch.jit.trace` under the `Autocast`, the trace process would hit Autocast's weight cache and fails. So we should disable weight cache under the trace process.
- Using `Autocast` with `Grad mode`
- Usually we are using `Grad mode` for training. Since in the training phase, the weight will change in every step. So we doesn't need to cache the weight.
- For the recommended `Autocast` training case in the [doc](https://pytorch.org/docs/stable/amp.html), `Autocast` will clear the cache every step leaving the context. We should disable it to save the clear operations.
```
model = Net().cuda()
optimizer = optim.SGD(model.parameters(), ...)
for input, target in data:
optimizer.zero_grad()
with autocast():
output = model(input)
loss = loss_fn(output, target)
loss.backward()
optimizer.step()
```
Test Plan: Imported from OSS
Reviewed By: mrshenli
Differential Revision: D30644913
Pulled By: ezyang
fbshipit-source-id: ad7bc87372e554e7aa1aa0795e9676871b3974e7
Summary:
Fixes https://github.com/pytorch/pytorch/issues/62811
Add `torch.linalg.matmul` alias to `torch.matmul`. Note that the `linalg.matmul` doesn't have a `method` variant.
Also cleaning up `torch/_torch_docs.py` when formatting is not needed.
cc IvanYashchuk Lezcano mruberry rgommers
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63227
Reviewed By: mrshenli
Differential Revision: D30770235
Pulled By: mruberry
fbshipit-source-id: bfba77dfcbb61fcd44f22ba41bd8d84c21132403
Summary:
Fixes https://github.com/pytorch/pytorch/issues/61767
## Changes
- [x] Add `torch.concat` alias to `torch.cat`
- [x] Add OpInfo for `cat`/`concat`
- [x] Fix `test_out` skips (Use `at::native::resize_output` or `at::native::resize_output_check`)
- [x] `cat`/`concat`
- [x] `stack`
- [x] `hstack`
- [x] `dstack`
- [x] `vstack`/`row_stack`
- [x] Remove redundant tests for `cat`/`stack`
~I've not added `cat`/`concat` to OpInfo `op_db` yet, since cat is a little more tricky than other OpInfos (should have a lot of tests) and currently there are no OpInfos for that. I can try to add that in a subsequent PR or maybe here itself, whatever is suggested.~
**Edit**: cat/concat OpInfo has been added.
**Note**: I've added the named tensor support for `concat` alias as well, maybe that's out of spec in `array-api` but it is still useful for consistency in PyTorch.
Thanks to krshrimali for guidance on my first PR :))
cc mruberry rgommers pmeier asmeurer leofang AnirudhDagar asi1024 emcastillo kmaehashi heitorschueroff krshrimali
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62560
Reviewed By: saketh-are
Differential Revision: D30762069
Pulled By: mruberry
fbshipit-source-id: 6985159d1d9756238890488a0ab3ae7699d94337
Summary:
This PR implements the necessary hooks/stubs/enums/etc for complete ONNX Runtime (ORT) Eager Mode integration. The actual extension will live out of tree at https://github.com/pytorch/ort.
We have been [working on this at Microsoft](https://github.com/microsoft/onnxruntime-pytorch/tree/eager-ort/torch_onnxruntime) for the last few months, and are finally ready to contribute the PyTorch core changes upstream (nothing major or exciting, just the usual boilerplate for adding new backends).
The ORT backend will allow us to ferry [almost] all torch ops into granular ONNX kernels that ORT will eagerly execute against any devices it supports (therefore, we only need a single ORT backend from a PyTorch perspective).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58248
Reviewed By: astaff
Differential Revision: D30344992
Pulled By: albanD
fbshipit-source-id: 69082b32121246340d686e16653626114b7714b2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59077Fixes#58549
`from_buffer` constructs a tensor object from an already allocated buffer through
CPython's buffer protocol. Besides the standard `dtype`, `count`, and `offset` parameters,
this function also accepts:
- `device`: where the buffer lives
- `requires_grad`: should autograd record operations on the new tensor
A new test file _test_buffer_protocol.py_ was created. Currently, only CPU tests were
implemented. That's because neither PyTorch nor Numba implements CPython's buffer
protocol. Therefore, there's no way to create a CUDA buffer with the existing
dependencies (could use PyCUDA for that, though).
At the moment, if `device` differs from the device the buffer actually lives, two things
may happen:
- `RuntimeError`, if `device='cuda'`
- Segmentation fault (not tested -- see above), if `device='cpu'`
Test Plan: Imported from OSS
Reviewed By: jbschlosser
Differential Revision: D29870914
Pulled By: mruberry
fbshipit-source-id: 9fa8611aeffedfe39c9af74558178157a11326bb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61570
Fused operator that computes moving average min/max values (in-place) of the input tensor and fake-quantizes it.
It expects the qmin/qmax values to reflect the range of the quantized tensor (instead of reduce_range)
Motivation for adding this operator is for performance reasons, since moving the computation from python to C++/CUDA can increase the performance of QAT.
Test Plan:
python test/test_quantization.py TestFusedObsFakeQuant
Imported from OSS
Reviewed By: vkuzo
Differential Revision: D29682762
fbshipit-source-id: 28e4c50e77236d6976fe4b326c9a12103ed95840
Summary:
This PR un-reverts https://github.com/pytorch/pytorch/issues/61475 + fixes compilation with MSVC, that does not recognize alternative operator spellings (i.e. using `or` instead of `||` )
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61937
Reviewed By: albanD
Differential Revision: D29805941
Pulled By: malfet
fbshipit-source-id: 01e5963c6717c1b44b260300d87ba0bf57f26ce9
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56058
User facing changes:
1. Adds a negative bit and corresponding new API (`is_neg()`,`resolve_neg()`)
2. `tensor.conj().imag` now returns a floating point tensor with neg bit set to 1 instead of a tensor with no notion of negative bit. Note that imag is still a view and all the view properties still hold for imag.
Non user facing changes:
1. Added a new Negative dispatch key and a backend fallback to handle it
2. Updated copy kernel to handle negative bit
3. Merged conjugate and negative bit fallback kernel
4. fixed https://github.com/pytorch/pytorch/issues/60478 (caused due to https://github.com/pytorch/pytorch/pull/54987)
Testing:
1. Added a new OpInfo based test `test_neg_view` (verifies that out-of-place and in-place operations work correctly for all operations when the input is a neg view tensor by checking the result against an actually negated tensor, verifies that autograd returns the same output for both neg view and actually negated tensors as well as it works fine when grad_out is a neg view).
2. Added a new test class containing `test_conj_view`, `test_neg_view`.
Test Plan: Imported from OSS
Reviewed By: soulitzer
Differential Revision: D29636403
fbshipit-source-id: 12214c9dc4806c51850f4a72a109db9527c0ca63
Summary:
Based from https://github.com/pytorch/pytorch/pull/50466
Adds the initial implementation of `torch.cov` similar to `numpy.cov`. For simplicity, we removed support for many parameters in `numpy.cov` that are either redundant such as `bias`, or have simple workarounds such as `y` and `rowvar`.
cc PandaBoi
closes https://github.com/pytorch/pytorch/issues/19037
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58311
Reviewed By: jbschlosser
Differential Revision: D29431651
Pulled By: heitorschueroff
fbshipit-source-id: 167dea880f534934b145ba94291a9d634c25b01b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58059
Add CUDA.used vital sign which is true only if CUDA was "used" which technically means the context was created.
Also adds the following features:
- Force vitals to be written even if vitals are disabled, to enable testing when the env variable is not set from the start of execution
- Add a read_vitals call for python to read existing vital signs.
Test Plan: buck test mode/dbg caffe2/test:torch -- --regex basic_vitals
Reviewed By: xuzhao9
Differential Revision: D28357615
fbshipit-source-id: 681bf9ef63cb1458df9f1c241d301a3ddf1e5252
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59760
See https://github.com/pytorch/pytorch/issues/59049
There are some moving parts to this PR, I'll structure this explanation so the straightforward parts go first, and then the less straightforward parts.
**The actual dispatch to Python.** The core logic of dispatch to Python lives in `concrete_dispatch_fn` in `torch/csrc/autograd/python_variable.cpp`. It takes the input IValue stack, scans all the arguments for Tensor arguments, and defers most of the heavy lifting to `handle_torch_function_no_python_arg_parser` which actually does all of the logic for calling out to torch dispatch (in particular, this function handles multiple dispatch situations for you). Because we have a different function name than regular `__torch_function__` handling, `handle_torch_function_no_python_arg_parser` is generalized to accept a magic method name to look for when testing if Tensors have custom handling or not. Unlike `__torch_function__`, by default there is no `__torch_dispatch__` on Tensor classes.
**Maintaining the Python dispatch key.** In order to get to the dispatch to Python logic, we must tag Tensors with the `__torch_dispatch__` magic method with the newly added Python dispatch key (separated from PythonFuncTorch to allow for a transitional period while they migrate to this mechanism). We expose a new private property `_is_python_dispatch` that assists in debugging if a Tensor is participating in Python dispatch or not. We apply the Python dispatch key the first time a PyObject for a Tensor is constructed (THPVariable_NewWithVar), testing if `__torch_dispatch__` exists with then newly added `check_has_torch_dispatch`.
**Shallow copy and detach.** For the simple examples tested in this PR, most creations of Tensor route through the dispatcher. The exception to this is `shallow_copy_and_detach`, which bypasses the dispatcher and is used when saving tensors for backwards. When a Tensor is Python dispatch, we override the behavior of `shallow_copy_and_detach` to instead directly call into `__torch_dispatch__` to perform a `detach` operation (in the same way it would be invoked if you called `detach` directly). Because this Python call is triggered directly from c10::TensorImpl, it must be indirected through `PyInterpreter::detach`, which is the general mechanism for dynamic dispatching to the Python interpreter associated with a TensorImpl.
**torchdeploy compatibility.** The dispatch to Python logic cannot be directly registered to the dispatcher as it is compiled in the Python library, which will get loaded multiple times per torchdeploy interpreter. Thus, we must employ a two phase process. First, we register a fallback inside a non-Python library (aten/src/ATen/core/PythonFallbackKernel.cpp). Its job is to determine the appropriate PyInterpreter to handle the Python dispatch by going through all of the arguments and finding the first argument that has a PyObject/PyInterpreter. With this PyInterpreter, it makes another dynamic dispatch via "dispatch" which will go to the correct torchdeploy interpreter to handle dispatching to actual Python.
**Testing.** We provide a simple example of a LoggingTensor for testing, which can be used to generate TorchScript-like traces to observe what operations are being called when a Tensor is invoked. Although a LoggingTensor would be better implemented via an is-a relationship rather than a has-a relationship (as is done in the test), we've done it this way to show that arbitrarily complex compositions of tensors inside a tensor work properly.
**Known limitations.**
* We haven't adjusted any operator code, so some patterns may not work (as they lose the Python subclass in an unrecoverable way)
* `__torch_function__` must be explicitly disabled with `_disabled_torch_function_impl` otherwise things don't work quite correctly (in particular, what is being disabled is default subclass preservation behavior.)
* We don't ever populate kwargs, even when an argument is kwarg-only
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Differential Revision:
D29017912
D29017912
Test Plan: Imported from OSS
Reviewed By: bdhirsh
Pulled By: ezyang
fbshipit-source-id: a67714d9e541d09203a8cfc85345b8967db86238
Summary:
Reference https://github.com/pytorch/pytorch/issues/50345
`zeta` was already present in the codebase to support computation of `polygamma`.
However, `zeta` only had `double(double, double)` signature **for CPU** before the PR (which meant that computation `polygamma` were always upcasted to `double` for zeta part).
With this PR, float computations will take place in float and double in double.
Have also refactored the code and moved the duplicate code from `Math.cuh` to `Math.h`
**Note**: For scipy, q is optional, and if it is `None`, it defaults `1` which corresponds to Reimann-Zeta. However, for `torch.specia.zeta`, I made it mandatory cause for me it feels odd without `q` this is Reimann-Zeta and with `q` it is the general Hurwitz Zeta. I think sticking to just general made more sense as passing `1` for q sounds trivial.
Verify:
* [x] Docs https://14234587-65600975-gh.circle-artifacts.com/0/docs/special.html#torch.special.zeta
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59623
Reviewed By: ngimel
Differential Revision: D29348269
Pulled By: mruberry
fbshipit-source-id: a3f9ebe1f7724dbe66de2b391afb9da1cfc3e4bb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/60464
Fixes https://github.com/szagoruyko/pytorchviz/issues/65
An alternate implementation of this PR would be to remove the
__torch_function__ interposition points for these accessors entirely.
In the end, I decided to opt for extra expressivity. See
torch.overrides for the criterion on how I decided which accessors
should get the nowrap treatment.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D29302835
Pulled By: ezyang
fbshipit-source-id: fbe0ac4530a6cc9d6759a3fdf5514d4d7b1f7690
