Summary: In Pytorch 2.5 we added source code attribution to PT2 traces. Each Torch-Compiled Region will now have its frame id and frame compile id associated with it. Update the image in the doc and add a description of this in the doc itself
Test Plan:
{F1980179183}
Rollback Plan:
Differential Revision: D78118228
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158066
Approved by: https://github.com/aaronenyeshi
Generate source tarball with PEP 517 conform build tools instead of the custom routine in place right now.
Closes#150461.
The current procedure for generating the source tarball consists in creation of a source tree by manual copying and pruning of source files.
This PR replaces that with a call to the standard [build tool](https://build.pypa.io/en/stable/), which works with the build backend to produce an sdist. For that to work correctly, the build backend also needs to be configured. In the case of Pytorch, the backend currently is (the legacy version of) the setuptools backend, the source dist part of which is mostly configured via the `MANIFEST.in` file.
The resulting source distribution can be used to install directly from source with `pip install ./torch-{version}.tar.gz` or to build wheels directly from source with `pip wheel ./torch-{version}.tar.gz`; both should be considered experimental for now.
## Issues
### sdist name
According to PEP 517, the name of the source distribution file must coincide with the project name, or [more precisely](https://peps.python.org/pep-0517/#source-distributions), the source distribution of a project that generates `{NAME}-{...}.whl` wheels are required to be named `{NAME}-{...}.tar.gz`. Currently, the source tarball is called `pytorch-{...}.tar.gz`, but the generated wheels and python package are called `torch-{...}`.
### Symbolic Links
The source tree at the moment contains a small number of symbolic links. This [has been seen as problematic](https://github.com/pypa/pip/issues/5919) largely because of lack of support on Windows, but also because of [a problem in setuptools](https://github.com/pypa/setuptools/issues/4937). Particularly unfortunate is a circular symlink in the third party `ittapi` module, which can not be resolved by replacing it with a copy.
PEP 721 (now integrated in the [Source Distribution Format Specification](https://packaging.python.org/en/latest/specifications/source-distribution-format/#source-distribution-archive-features)) allows for symbolic links, but only if they don't point outside the destination directory and if they don't contain `../` in their target.
The list of symbolic links currently is as follows:
<details>
|source|target|problem|solution|
|-|-|-|-|
| `.dockerignore` | `.gitignore` | ✅ ok (individual file) ||
| `docs/requirements.txt` | `../.ci/docker/requirements-docs.txt` |❗`..` in target|swap source and target[^1]|
| `functorch/docs/source/notebooks` | `../../notebooks/` |❗`..` in target|swap source and target[^1]|
| `.github/ci_commit_pins/triton.txt` | `../../.ci/docker/ci_commit_pins/triton.txt` | ✅ ok (omitted from sdist)||
| `third_party/flatbuffers/docs/source/CONTRIBUTING.md` | `../../CONTRIBUTING.md` |❗`..` in target|omit from sdist[^2]|
| `third_party/flatbuffers/java/src/test/java/DictionaryLookup` | `../../../../tests/DictionaryLookup` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/java/src/test/java/MyGame` | `../../../../tests/MyGame` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/java/src/test/java/NamespaceA` | `../../../../tests/namespace_test/NamespaceA` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/java/src/test/java/NamespaceC` | `../../../../tests/namespace_test/NamespaceC` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/java/src/test/java/optional_scalars` | `../../../../tests/optional_scalars` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/java/src/test/java/union_vector` | `../../../../tests/union_vector` |❗`..` in target|omit from sdist[^3]|
| `third_party/flatbuffers/kotlin/benchmark/src/jvmMain/java` | `../../../../java/src/main/java` |❗`..` in target|omit from sdist[^3]|
| `third_party/ittapi/rust/ittapi-sys/c-library` | `../../` |❗`..` in target|omit from sdist[^4]|
| `third_party/ittapi/rust/ittapi-sys/LICENSES` | `../../LICENSES` |❗`..` in target|omit from sdist[^4]|
| `third_party/opentelemetry-cpp/buildscripts/pre-merge-commit` | `./pre-commit` |✅ ok (individual file)||
| `third_party/opentelemetry-cpp/third_party/prometheus-cpp/cmake/project-import-cmake/sample_client.cc` | `../../push/tests/integration/sample_client.cc` |❗`..` in target|omit from sdist[^5]|
| `third_party/opentelemetry-cpp/third_party/prometheus-cpp/cmake/project-import-cmake/sample_server.cc` | `../../pull/tests/integration/sample_server.cc` |❗`..` in target|omit from sdist[^5]|
| `third_party/opentelemetry-cpp/third_party/prometheus-cpp/cmake/project-import-pkgconfig/sample_client.cc` | `../../push/tests/integration/sample_client.cc` |❗`..` in target|omit from sdist[^5]|
| `third_party/opentelemetry-cpp/third_party/prometheus-cpp/cmake/project-import-pkgconfig/sample_server.cc` | `../../pull/tests/integration/sample_server.cc` |❗`..` in target|omit from sdist[^5]|
| `third_party/XNNPACK/tools/xngen` | `xngen.py` | ✅ ok (individual file)||
</details>
The introduction of symbolic links inside the `.ci/docker` folder creates a new problem, however, because Docker's `COPY` command does not allow symlinks in this way. We work around that by using `tar ch` to dereference the symlinks before handing them over to `docker build`.
[^1]: These resources can be naturally considered to be part of the docs, so moving the actual files into the place of the current symlinks and replacing them with (unproblematic) symlinks can be said to improve semantics as well.
[^2]: The flatbuffers docs already actually use the original file, not the symlink and in the most recent releases, starting from flatbuffers-25.1.21 the symlink is replaced by the actual file thanks to a documentation overhaul.
[^3]: These resources are flatbuffers tests for java and kotlin and can be omitted from our sdist.
[^4]: We don't need to ship the rust bindings for ittapi.
[^5]: These are demonstration examples for how to link to prometheus-cpp using cmake and can be omitted.
### Nccl
Nccl used to be included as a submodule. However, with #146073 (first released in v2.7.0-rc1), the submodule was removed and replaced with a build time checkout procedure in `tools/build_pytorch_libs.py`, which checks out the required version of nccl from the upstream repository based on a commit pin recorded in `.ci/docker/ci_commit_pins/nccl-cu{11,12}.txt`.
This means that a crucial third party dependency is missing from the source distribution and as the `.ci` folder is omitted from the source distribution, it is not possible to use the build time download.
However, it *is* possible to use a system provided Nccl using the `USE_SYSTEM_NCCL` environment variable, which now also is the default for the official Pytorch wheels.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152098
Approved by: https://github.com/atalman
This PR adds a new config `backward_pass_autocast`, to set the backward autocast
behavior. It does not change the existing behavior.
The reason why we need this is that torch.compile acquires a forward and
backward graph at the time of the forward pass. This means that
implemented naively, if there are any context managers active outside
the call to torch.compile, the backward graph will also get the
behaviors from those context managers. This PR gives users a way to
tweak the autocast behavior of the backward pass.
Please see torch._functorch.config for the options to the
`backward_pass_autocast` config.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156356
Approved by: https://github.com/bdhirsh
ghstack dependencies: #155354
The dtype documentation has not been updated in awhile, let's do a revamp.
1. combine the duplicated docs for dtypes from `tensors.rst` and `tensor_attributes.rst` to live in `tensor_attributes.rst`, and link to that page from `tensors.rst`
2. split the dtype table into floating point and integer dtypes
3. add the definition of shell dtype
4. add the float8 and MX dtypes as shell dtypes to the dtype table
5. remove legacy quantized dtypes from the table
6. add the definition of various dtype suffixes ("fn", etc)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156087
Approved by: https://github.com/albanD
Based on the [conversation](https://github.com/pytorch/pytorch/issues/121791), we plan to drop the "highest, high, medium" to represent fp32 internal computation data types . Instead, we will directly use the algorithm to represent it.
### Design Choice: Directly use algorithms name like "TF32", "BF16".
#### Pros
- The names are more informative. 'tf32' is more informative than a simple "high".
- Easier to extend new algorithm like `tf32x3`
#### Cons
- "HIGHEST, HIGH, MEDIUM" indicated the relative precision between different algorithms. However, we can have more documents to discuss them.
### We provide a layered structure for backends/operators.
('f32' is short for 'fp32_precision')
### We provide 3 fp32 compute precision can be set:
- **"ieee"**: Not allowed to use any other internal computation data types .
- **"tf32"**: Allowed to use tf32 as internal computation data types.
- **"bf16"**: Allowed to use bf16 as internal computation data types.
- **"none"**: Precision's are not set. Can be override by its father node.
### Overriding Precision Settings
Child node can be override by its father node if it is set to default.
For current default settings:
```
backend = generic, op = all, precision setting = none
backend = cuda, op = all, precision setting = none
backend = cuda, op = conv, precision setting = tf32
backend = cuda, op = rnn, precision setting = tf32
backend = cuda, op = matmul, precision setting = none
backend = matmul, op = all, precision setting = none
backend = matmul, op = conv, precision setting = none
backend = matmul, op = rnn, precision setting = none
backend = matmul, op = matmul, precision setting = none
```
- If the user set `torch.backends.mkldnn.fp32_precision="bf16"`, his child nodes `torch.backends.mkldnn.matmul.fp32_precision` / `torch.backends.mkldnn.conv.fp32_precision` / `torch.backends.mkldnn.rnn.fp32_precision` will also be override to "bf16".
- If the user set `torch.backends.fp32_precision="bf16"`, `torch.backends.mkldnn.fp32_precision` and his child nodes will also we override to "bf16".
### Backward Compatible
Since new API allow user to have more fine-grained control. There will be some conflict. For example, previous `torch.backends.cudnn.allow_tf32` are not enough to represent the status for `torch.backends.cudnn.rnn.fp32_precision="ieee"` and `torch.backends.cudnn.conv.fp32_precision="tf32"`. Therefore, our goal for backward compatible is
- If the user only uses previous APIs, it will work as previous expectations.
- If the user use **new** API to change the status to an **un-representable** status for old API, and try to access the status by **old** API. We will raise Runtime Error and point the document for user.
### Test Plan
```
python test/test_cuda.py -k test_fp32_precision_with_tf32
python test/test_cuda.py -k test_fp32_precision_with_float32_matmul_precision
python test/test_cuda.py -k test_invalid_status_for_legacy_api
python test/test_mkldnn.py -k test_mlkdnn_get_set
python test/test_mkldnn.py -k test_generic_precision
python test/test_mkldnn.py -k test_invalid
python test/test_mkldnn.py -k test_default_use_parent
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125888
Approved by: https://github.com/jgong5, https://github.com/albanD
Co-authored-by: Jiang, Yanbing <yanbing.jiang@intel.com>
# Motivation
Update the doc, to make `torch.device`'s constructor officially support the following methods:
- A device string, which is a string representation of the device type and optionally the device ordinal.
- A device type and a device ordinal.
- A device ordinal, which is treated as the current accelerator type.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156686
Approved by: https://github.com/albanD
This is a remix of https://github.com/pytorch/pytorch/pull/155558
Instead of mediating guard collective via a config option, in this one it's done via a `set_stance` like API. The motivation is that checking for the config value on entry on torch.compile is apparently quite expensive, according to functorch_maml_omniglot. So this makes it a bit cheaper.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156562
Approved by: https://github.com/Microve
Implement Attention-23 using sdpa and flexattention.
- I used copilot for this.
- Also updated the conversion logic to remove trailing None inputs.
@gramalingam @kunal-vaishnavi @titaiwangms
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156431
Approved by: https://github.com/titaiwangms
Co-authored-by: kunal-vaishnavi <115581922+kunal-vaishnavi@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Fixes#128796
This PR adds documentation about the behavior of division by zero operations in PyTorch's autograd system. The documentation explains:
1. How division by zero produces `inf` values following IEEE-754 floating point arithmetic
2. How autograd handles these cases and why masking after division can lead to `nan` gradients
3. Provides concrete examples showing the issue
4. Recommends two solutions:
- Masking before division
- Using MaskedTensor (experimental API)
The documentation is added to the autograd notes section, making it easily discoverable for users who encounter this common issue.
This addresses the original issue #128796 which requested better documentation of this behavior to help users avoid common pitfalls when dealing with division by zero in their models.
dditional changes:
- Fixed formatting consistency by replacing curly apostrophes with straight apostrophes in the existing documentation
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155987
Approved by: https://github.com/soulitzer
Co-authored-by: sekyondaMeta <127536312+sekyondaMeta@users.noreply.github.com>
Fixes#155028
This pull request updates the documentation by transitioning from .rst to .md format. It introduces new Markdown files for the documentation of named_tensor, nested, nn.attention.bias, nn.attention.experimental, and nn.attention.flex_attention
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155696
Approved by: https://github.com/svekars
Co-authored-by: Svetlana Karslioglu <svekars@meta.com>
Fixes#155036
This pull request updates the documentation for several modules by transitioning from .rst to .md format, improving readability and usability. It introduces new Markdown files for the documentation of torch.ao.ns._numeric_suite, torch.ao.ns._numeric_suite_fx, AOTInductor, AOTInductor Minifier, and the torch.compiler API
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155377
Approved by: https://github.com/svekars
Co-authored-by: Svetlana Karslioglu <svekars@meta.com>
Fixes https://github.com/pytorch/pytorch/issues/155023
Related PR: #155781
Description:
As discussed, this PR is a follow-up update for `jit_language_reference_v2.md` by deleting the code chunk indentation.
Checklist:
- [x] The issue being fixed is referenced above (Fixes https://github.com/pytorch/pytorch/issues/155023)
- [x] Only one issue is addressed in this pull request
- [x] Labels from the issue that this PR is fixing are added to this pull request
- [x] No unnecessary issues are included into this pull request.
@pytorchbot label "topic: docs"
@pytorchbot label "topic: not user facing"
@pytorchbot label docathon-h1-2025
@pytorchbot label "module: docs"
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155937
Approved by: https://github.com/jingsh, https://github.com/svekars
Summary: Change HF classes to not have an underscore, there-by making them public, we will add documentation to them following this
Test Plan:
ensure existing tests pass
Rollback Plan:
Differential Revision: D76364024
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155837
Approved by: https://github.com/saumishr
Fixes https://github.com/pytorch/pytorch/issues/155023
Description:
converted `jit_language_reference_v2.rst` to `jit_language_reference_v2.md`
**I indented the code blocks to minimize the file difference to pass the sanity check for no more than 2000 lines of change. I will submit another PR to fix the indentation after this PR is merged.**
Checklist:
- [x] The issue being fixed is referenced above (Fixes https://github.com/pytorch/pytorch/issues/155023)
- [x] Only one issue is addressed in this pull request
- [x] Labels from the issue that this PR is fixing are added to this pull request
- [x] No unnecessary issues are included into this pull request.
@pytorchbot label "topic: docs"
@pytorchbot label "topic: not user facing"
@pytorchbot label docathon-h1-2025
@pytorchbot label module: docs
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155781
Approved by: https://github.com/svekars
This PR registers the RotaryEmbedding op in the `torch.ops.onnx` name spaces and allows the exporter to recognize and export onnx operators.
## Design
ONNX operators of their respective opset version is implemented in torch/onnx/ops/_impl.py, and are registered in the torch.ops.onnx namespace following the following rule:
`OpType-version => torch.ops.onnx.OpType.opset{version}`
For example, `RotaryEmbedding-23` becomes `torch.ops.onnx.RotaryEmbedding.opset23`
This name is parsed by the exporter to create an onnx node in the graph without having to go through translation.
When users use the ops in the model, we provide more convenient, unversioned functions under `torch.onnx.ops` that will dispatch to the implementations based on user input (type and provided attributes). For example, users can directly call `torch.onnx.ops.rotary_embedding()` to use the op natively in their pytorch models. I chose snake case naming to make the functions more pythonic and aligned with other torch apis.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154745
Approved by: https://github.com/titaiwangms
Which inherits from `RuntimeError` and contains `error_code`, which in case of CUDA should contain error returned by `cudaGetLastError`
`torch::detail::_new_accelerator_error_object(c10::AcceleratorError&)` follows the pattern of CPython's [`PyErr_SetString`](cb8a72b301/Python/errors.c (L282)), namely
- Convert cstr into Python string with `PyUnicode_FromString`
- Create new exception object using `PyObject_CallOneArg` just like it's done in [`_PyErr_CreateException`](cb8a72b301/Python/errors.c (L32))
- Set `error_code` property using `PyObject_SetAttrString`
- decref all temporary references
Test that it works and captures CPP backtrace (in addition to CI) by running
```python
import os
os.environ['TORCH_SHOW_CPP_STACKTRACES'] = '1'
import torch
x = torch.rand(10, device="cuda")
y = torch.arange(20, device="cuda")
try:
x[y] = 2
print(x)
except torch.AcceleratorError as e:
print("Exception was raised", e.args[0])
print("Captured error code is ", e.error_code)
```
which produces following output
```
Exception was raised CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
Exception raised from c10_cuda_check_implementation at /home/ubuntu/pytorch/c10/cuda/CUDAException.cpp:41 (most recent call first):
C++ CapturedTraceback:
#4 std::_Function_handler<std::shared_ptr<c10::LazyValue<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > const> (), c10::SetStackTraceFetcher(std::function<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > ()>)::{lambda()#1}>::_M_invoke(std::_Any_data const&) from Logging.cpp:0
#5 c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >) from ??:0
#6 c10::cuda::c10_cuda_check_implementation(int, char const*, char const*, int, bool) [clone .cold] from CUDAException.cpp:0
#7 void at::native::gpu_kernel_impl<at::native::AbsFunctor<float> >(at::TensorIteratorBase&, at::native::AbsFunctor<float> const&) [clone .isra.0] from tmpxft_000191fc_00000000-6_AbsKernel.cudafe1.cpp:0
#8 at::native::abs_kernel_cuda(at::TensorIteratorBase&) from ??:0
#9 at::Tensor& at::native::unary_op_impl_with_complex_to_float_out<at::native::abs_stub_DECLARE_DISPATCH_type>(at::Tensor&, at::Tensor const&, at::native::abs_stub_DECLARE_DISPATCH_type&, bool) [clone .constprop.0] from UnaryOps.cpp:0
#10 at::(anonymous namespace)::(anonymous namespace)::wrapper_CUDA_out_abs_out(at::Tensor const&, at::Tensor&) from RegisterCUDA_0.cpp:0
#11 at::_ops::abs_out::call(at::Tensor const&, at::Tensor&) from ??:0
#12 at::native::abs(at::Tensor const&) from ??:0
#13 c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (at::Tensor const&), &at::(anonymous namespace)::(anonymous namespace)::wrapper_CompositeExplicitAutograd__abs>, at::Tensor, c10::guts::typelist::typelist<at::Tensor const&> >, at::Tensor (at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&) from RegisterCompositeExplicitAutograd_0.cpp:0
#14 at::_ops::abs::redispatch(c10::DispatchKeySet, at::Tensor const&) from ??:0
#15 torch::autograd::VariableType::(anonymous namespace)::abs(c10::DispatchKeySet, at::Tensor const&) from VariableType_1.cpp:0
#16 c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (c10::DispatchKeySet, at::Tensor const&), &torch::autograd::VariableType::(anonymous namespace)::abs>, at::Tensor, c10::guts::typelist::typelist<c10::DispatchKeySet, at::Tensor const&> >, at::Tensor (c10::DispatchKeySet, at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&) from VariableType_1.cpp:0
#17 at::_ops::abs::call(at::Tensor const&) from ??:0
#18 at::native::isfinite(at::Tensor const&) from ??:0
#19 c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (at::Tensor const&), &at::(anonymous namespace)::(anonymous namespace)::wrapper_CompositeImplicitAutograd__isfinite>, at::Tensor, c10::guts::typelist::typelist<at::Tensor const&> >, at::Tensor (at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&) from RegisterCompositeImplicitAutograd_0.cpp:0
#20 at::_ops::isfinite::call(at::Tensor const&) from ??:0
#21 torch::autograd::THPVariable_isfinite(_object*, _object*, _object*) from python_torch_functions_2.cpp:0
#22 PyObject_CallFunctionObjArgs from ??:0
#23 _PyObject_MakeTpCall from ??:0
#24 _PyEval_EvalFrameDefault from ??:0
#25 _PyObject_FastCallDictTstate from ??:0
#26 _PyStack_AsDict from ??:0
#27 _PyObject_MakeTpCall from ??:0
#28 _PyEval_EvalFrameDefault from ??:0
#29 _PyFunction_Vectorcall from ??:0
#30 _PyEval_EvalFrameDefault from ??:0
#31 _PyFunction_Vectorcall from ??:0
#32 _PyEval_EvalFrameDefault from ??:0
#33 _PyFunction_Vectorcall from ??:0
#34 _PyEval_EvalFrameDefault from ??:0
#35 PyFrame_GetCode from ??:0
#36 PyNumber_Xor from ??:0
#37 PyObject_Str from ??:0
#38 PyFile_WriteObject from ??:0
#39 _PyWideStringList_AsList from ??:0
#40 _PyDict_NewPresized from ??:0
#41 _PyEval_EvalFrameDefault from ??:0
#42 PyEval_EvalCode from ??:0
#43 PyEval_EvalCode from ??:0
#44 PyUnicode_Tailmatch from ??:0
#45 PyInit__collections from ??:0
#46 PyUnicode_Tailmatch from ??:0
#47 _PyRun_SimpleFileObject from ??:0
#48 _PyRun_AnyFileObject from ??:0
#49 Py_RunMain from ??:0
#50 Py_BytesMain from ??:0
#51 __libc_init_first from ??:0
#52 __libc_start_main from ??:0
#53 _start from ??:0
Captured error code is 710
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152023
Approved by: https://github.com/eqy, https://github.com/mradmila, https://github.com/ngimel
ghstack dependencies: #154436
The goal of this multigraph work is to enable a compiled region that has a single dynamo trace but multiple backend specializations. This work was inspired by vLLM which does this in a somewhat hacky way where they use a custom backend to capture a dynamo graph and then manually invoke compile_fx multiple times to get specialized graphs.
There's really two parts of this work:
**The frontend changes:**
1) we introduce an optional kwarg `specialize_on` to mark_{dynamic,unbacked} that takes in a list of specializations. I debated other methods including specifying specializations via decorators, but ultimately decided this approach was more harmonious. The big issue with decorators is the difficulty of composing well with the rest of the torch.compile ecosystem including graph breaks, lazy initialization of variable trackers and symbolic variables, etc.
**The backend changes (this PR):**
1) We capture the backend_specialization specified in the mark_{dynamic,unbacked} API into a SymbolicContext. See changes in `/_dynamo/variables/builder.py`
2) After we are done dynamo tracing, we will lazily (more on this later) invoke `call_user_compiler` up to N + 1 times for N specializations and 1 generic graph. Under the hood this will call compile_fx, which composes nicely with both Async Compile and AOTAutogradCache. We do this by using a context manager to patch in specialization specific axioms into the ShapeEnv before invoking the user compiler.
3) When we have specializations, we install a lazy specialized dispatch function that checks each specialization and dispatches to the first one that matches. Instead of doing all of the specialization compiles up front, we do the compiles lazily. The first time a specialization is invoked, we will do the compilation and save it in a cache so subsequent invocations are fast. If none of the specializations match, we dispatch to the generic graph. I decided to do this over returning N different GuardedCodes since 1) it doesn't pollute the dynamo cache (eg. if you have 8 specializations, you would hit the cache limit) 2) it naturally incorporates the hierarchical lattice structure of the guards since the specializations are always necessarily stricter than the generic region's guards.
I benchmarked this PR stack with #152596 and found around a 50% reduction when dispatching to the specialized regions:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153449
Approved by: https://github.com/zou3519
ghstack dependencies: #153433
Removes MemPoolContext from custom user mempools. The ground truth for which pool should be used is in graph_pools active pool, and MemPoolContext just introduced an opportunity for the pool pointed to by MemPoolContext and active pool in graph_pools to go out of sync (see all the asserts in the code to make sure that happens, and yet it still could happen in a multithread scenario, see my recent PRs (#153990).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154042
Approved by: https://github.com/albanD, https://github.com/syed-ahmed
- Move community and language binding links to the horizontal bar
- Add an intro to the community page.
- Fix the link in the ogp_image
- Fix the link in the version switcher
- Clean up unneeded links
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153090
Approved by: https://github.com/albanD
Based on the [conversation](https://github.com/pytorch/pytorch/issues/121791), we plan to drop the "highest, high, medium" to represent fp32 internal computation data types . Instead, we will directly use the algorithm to represent it.
### Design Choice: Directly use algorithms name like "TF32", "BF16".
#### Pros
- The names are more informative. 'tf32' is more informative than a simple "high".
- Easier to extend new algorithm like `tf32x3`
#### Cons
- "HIGHEST, HIGH, MEDIUM" indicated the relative precision between different algorithms. However, we can have more documents to discuss them.
### We provide a layered structure for backends/operators.
('f32' is short for 'fp32_precision')
### We provide 3 fp32 compute precision can be set:
- **"ieee"**: Not allowed to use any other internal computation data types .
- **"tf32"**: Allowed to use tf32 as internal computation data types.
- **"bf16"**: Allowed to use bf16 as internal computation data types.
- **"none"**: Precision's are not set. Can be override by its father node.
### Overriding Precision Settings
Child node can be override by its father node if it is set to default.
For current default settings:
```
backend = generic, op = all, precision setting = none
backend = cuda, op = all, precision setting = none
backend = cuda, op = conv, precision setting = tf32
backend = cuda, op = rnn, precision setting = tf32
backend = cuda, op = matmul, precision setting = none
backend = matmul, op = all, precision setting = none
backend = matmul, op = conv, precision setting = none
backend = matmul, op = rnn, precision setting = none
backend = matmul, op = matmul, precision setting = none
```
- If the user set `torch.backends.mkldnn.fp32_precision="bf16"`, his child nodes `torch.backends.mkldnn.matmul.fp32_precision` / `torch.backends.mkldnn.conv.fp32_precision` / `torch.backends.mkldnn.rnn.fp32_precision` will also be override to "bf16".
- If the user set `torch.backends.fp32_precision="bf16"`, `torch.backends.mkldnn.fp32_precision` and his child nodes will also we override to "bf16".
### Backward Compatible
Since new API allow user to have more fine-grained control. There will be some conflict. For example, previous `torch.backends.cudnn.allow_tf32` are not enough to represent the status for `torch.backends.cudnn.rnn.fp32_precision="ieee"` and `torch.backends.cudnn.conv.fp32_precision="tf32"`. Therefore, our goal for backward compatible is
- If the user only uses previous APIs, it will work as previous expectations.
- If the user use **new** API to change the status to an **un-representable** status for old API, and try to access the status by **old** API. We will raise Runtime Error and point the document for user.
### Test Plan
```
python test/test_cuda.py -k test_fp32_precision_with_tf32
python test/test_cuda.py -k test_fp32_precision_with_float32_matmul_precision
python test/test_cuda.py -k test_invalid_status_for_legacy_api
python test/test_mkldnn.py -k test_mlkdnn_get_set
python test/test_mkldnn.py -k test_generic_precision
python test/test_mkldnn.py -k test_invalid
python test/test_mkldnn.py -k test_default_use_parent
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125888
Approved by: https://github.com/jgong5, https://github.com/albanD
Co-authored-by: Jiang, Yanbing <yanbing.jiang@intel.com>
definitely_true is almost same as guard_or_false, the potential differences are not meaningful to a degree that justify the
existence of both. same for definitely_false, it can be expressed with guard_or_true and guard_or_false.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152463
Approved by: https://github.com/bobrenjc93
Summary:
# Context:
When memory leak happens, it usually trigger the OOM in the later iterations. The snapshot of full iteration will be huge and hard to interpret.
On CUDA side, they provide OOM observer which generates snapshot when OOM happens with latest 1,500,000 entries for debugging.
In this diff, we want to implement the feature on MTIA side
Test Plan:
Run this test with last diff in the stack.
```
buck run @//mode/opt kineto/libkineto/fb/mtia/integration_tests:mtia_memory_auto_trace_test
```
As shown, the memory_snapshot is generated when oom happens
Log: P1794792326
Snapshot: https://fburl.com/pytorch_memory_visualizer/lx73y6s3 {F1977402355}
Differential Revision: D71993315
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152160
Approved by: https://github.com/sraikund16
MemPool is a separate pool of memory handled by the caching allocator. This PR adds the option let the caching allocator try to use this pool as a last resort instead of OOMing by associating a use_on_oom bool with each MemPool.
Usage:
Users can optionally specify a ``use_on_oom`` bool (which is False by default) during MemPool creation. If true, then the CUDACachingAllocator will be able to use memory in this pool as a last resort instead of OOMing.
```
pool = torch.cuda.MemPool(allocator, use_on_oom=True)
with torch.cuda.use_mem_pool(pool):
a = torch.randn(40 * 1024 * 1024, dtype=torch.uint8, device="cuda")
del a
# at the memory limit, this will succeed by using pool's memory in order to avoid the oom
b = torch.randn(40 * 1024 * 1024, dtype=torch.uint8, device="cuda")
```
Testing:
```
python test/test_cuda.py -k test_mempool_limited_memory_with_allocator
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151487
Approved by: https://github.com/eqy, https://github.com/syed-ahmed, https://github.com/ngimel
# Motivation
We propose adding support for the Python with statement on `torch.accelerator.device_index` to enable device switching functionality. This enhancement would simplify writing device-agnostic code and provide benefits across all accelerators. Its device-specific counterparts include [`torch.cuda.device`](00199acdb8/torch/cuda/__init__.py (L482)) and [`torch.cuda._DeviceGuard`](00199acdb8/torch/cuda/__init__.py (L469)).
**Design Philosophy**
It accepts either an `Int` or `None` as input. When `None` is passed, no device switch is performed. Supporting `None` is important for compatibility, as it's possible to encounter `None` values from `torch.device.index`.
Therefore, with this PR, we can do like this
```python
src = 0
dst = 1
# Set src to current device
torch.accelerator.set_device_index(src)
with torch.accelerator.device_index(dst):
# Inside with statement, we set dst to current device
assert torch.accelerator.get_device_index() == dst
# Here the current device should be src
assert torch.accelerator.get_device_index() == src
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148864
Approved by: https://github.com/albanD
This has been pretty helpful for the size-oblivious rewrite. Wanted the variadic args version to avoid `sym_or(a, sym_or(b, sym_or(c, d)))` in favor of `sym_or(a, b, c, d)`. Happy to change this to ban the 1-arg version.
This is better than plain and/or because the whole symbolic expression gets preserved, and if we guard on it or defer as a runtime assert, we preserve all branches.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150456
Approved by: https://github.com/laithsakka
This adds lazy initialization support to ProcessGroupGloo via `TORCH_GLOO_LAZY_INIT` or via `create_device(..., lazy_init=True)`
This is still a draft PR as there's one race condition when doing coalesced operations that needs to be fixed upstream in Gloo first. Depends on https://github.com/facebookincubator/gloo/pull/427 landing first
This also updates the gloo submodule to include the required changes.
Test plan:
added lazy init test variants
```
pytest -v test/distributed/test_c10d_gloo.py -k Lazy
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150801
Approved by: https://github.com/fduwjj
Summary:
Instead of explicitly specifying dynamic shapes, it is possible to infer them from additional example inputs. Together with the example inputs provided to export, we can basically make any varying dim dynamic and keep any fixed dim static. This should be useful for prod scenarios that have access to tests and/or profiling data, yet are somewhat removed from the model authoring process.
However this alone is not satisfactory: the exported program by design has only one graph, representing one path through the model, and we cannot necessarily guarantee that this graph works for the additional example inputs because different guards might have been created if we had exported with them instead (corresponding to different traced paths). However, checking that the additional example inputs satisfy the guards created by the original export should be sufficient for generalization.
Now, while we don't preserve all guards in the exported program, we do check a subset of them as part of input matching. So we add a verification step at the end of export when such additional example inputs are provided. This should be enough for now.
Test Plan: added test (positive and negative cases)
Differential Revision: D72001771
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150144
Approved by: https://github.com/bobrenjc93
Needed this class for because `parallelize_module` takes a dict, which doesn't allow `PrepareModuleInput` and `PrepareModuleOutput` to be applied at the same time.
The `PrepareModuleInputOutput` in this PR initializes two variables `prepare_module_input` and `prepare_module_output` and uses them to process module / inputs / outputs.
I had another implementation which put all code in `PrepareModuleInputOutput` and let `PrepareModuleInput` and `PrepareModuleOutput` inherit the monolithic `PrepareModuleInputOutput`. But it is
1. less cleaner
2. conceptually abusing inheritance because `PrepareModuleInput` shouldn't be able to access class methods of `PrepareModuleOutput` and vice versa
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150372
Approved by: https://github.com/wanchaol
**Summary**
This PR adds a lowering pass for x86 backend
- Patterns of `dequantize -> conv/linear (-> quantize)` are fused to corresponding quantized onednn ops.
- Weights are prepacked ahead of time.
- Post ops of conv/linear are fused if supported.
- The pass returns a `GraphModule` with the modifications mentioned above.
**Test plan**
```
pytest test/quantization/pt2e/test_x86inductor_quantizer.py -k test_lowering_to_x86
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149708
Approved by: https://github.com/jerryzh168, https://github.com/leslie-fang-intel
some context in this document:
https://docs.google.com/document/d/18nJsj-F2C_QXO7ClwzPcAUENQ-B440B43W7DdDnlDt4/edit?tab=t.0#heading=h.pgebnyi7pocj
But TLDR;
`guard_or_true`, `guard_or_false` are better than `guard_size_oblivious` due to :
- Easier to reason about what assumptions we are making while reading the code.
- Avoid size_oblivious complexity that is not needed.
- Avoid unsoundness that could make `guard_size_oblivious(a==1)` be true when its not true for some vaue `a` during runtime.
- Less data dependent errors for some cases: ex, when doing `guard_size_oblivious(a==1)` and we know `a` is a tensor size, if it's traced with `a=u1-u2` `guard_size_oblivious(a==1)` will throw a data dependent error but `guard_else_false` will just return `False`.
### How is it different from statically_known_true??
**`if(cond)`:** (normal guarding) will try to evaluate statically and guard on the condition, willing to restrict input space to evaluate cond. if it fails to evaluate due to data dependent error will throw an exception (that could be converted to graph break in some situations).
**`statically_known_true(cond)`:** would be used when you never want to add a guard (restrict your input space), but just want to do a best effort check to see if you can infer that something is true/false ONLY based on existing constraints.
**`guard_or_true(cond)`/`guard_or_false(cond)`:** Those would be used in situations you prefer to guard and know the result of the expression over not guarding, but in case you hit a data dependent error you are ok with just returning true or false.
Some reasons you might be ok with returning true/false instead could be:
1. It's an optimization I do not want to fail for not performing optimization.
2. I am willing to deviate from the normal semantics when I have unbacked for the benefit of not failing (See the doc above for more details).
**`definitely_true(cond)`**: same as `guard_or_false(cond)` except does not try to do static eval for unbacked (planning to deprecate it and replace uses with `guard_or_false` or make it alias to `guard_or_false`)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148430
Approved by: https://github.com/bobrenjc93
some context in this document:
https://docs.google.com/document/d/18nJsj-F2C_QXO7ClwzPcAUENQ-B440B43W7DdDnlDt4/edit?tab=t.0#heading=h.pgebnyi7pocj
But TLDR;
`guard_or_true`, `guard_or_false` are better than `guard_size_oblivious` due to :
- Easier to reason about what assumptions we are making while reading the code.
- Avoid size_oblivious complexity that is not needed.
- Avoid unsoundness that could make `guard_size_oblivious(a==1)` be true when its not true for some vaue `a` during runtime.
- Less data dependent errors for some cases: ex, when doing `guard_size_oblivious(a==1)` and we know `a` is a tensor size, if it's traced with `a=u1-u2` `guard_size_oblivious(a==1)` will throw a data dependent error but `guard_else_false` will just return `False`.
### How is it different from statically_known_true??
**`if(cond)`:** (normal guarding) will try to evaluate statically and guard on the condition, willing to restrict input space to evaluate cond. if it fails to evaluate due to data dependent error will throw an exception (that could be converted to graph break in some situations).
**`statically_known_true(cond)`:** would be used when you never want to add a guard (restrict your input space), but just want to do a best effort check to see if you can infer that something is true/false ONLY based on existing constraints.
**`guard_or_true(cond)`/`guard_or_false(cond)`:** Those would be used in situations you prefer to guard and know the result of the expression over not guarding, but in case you hit a data dependent error you are ok with just returning true or false.
Some reasons you might be ok with returning true/false instead could be:
1. It's an optimization I do not want to fail for not performing optimization.
2. I am willing to deviate from the normal semantics when I have unbacked for the benefit of not failing (See the doc above for more details).
**`definitely_true(cond)`**: same as `guard_or_false(cond)` except does not try to do static eval for unbacked (planning to deprecate it and replace uses with `guard_or_false` or make it alias to `guard_or_false`)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148430
Approved by: https://github.com/bobrenjc93
This adds a `reduce_scatter` implementation for ProcessGroupGloo. This is a pretty naive implementation as it does 1 allreduce per rank but may be useful for testing in FSDP etc. There was an existing implementation of reduce_scatter_tensor/reduce_scatter_tensor_coalesed that has a very similar implementation but requires a fixed tensor size per rank.
If users find these functions to be too slow we can address them as issues arise.
Gloo now supports all major distributed operations. Quite a few of these were added by @rohan-varma and @yifuwang but they didn't update the support chart. We also have `CUDAWork` variants of most operations so those were also added to the chart.
Test plan:
```
pytest -v test/distributed/test_c10d_gloo.py -k reduce_scatter
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149869
Approved by: https://github.com/fduwjj
Summary:
1\ The current write item structure does not contain the amount of data that needs to be written.
2\ the planner.item already has a size primitive 'tensor_storage_size'. https://fburl.com/code/7a0gsmw7 But only for tensors.
3\ Right now, the only way the writer layer get hold of this property (fro non tensor data)
first do a lookup in to the actual tensor/bytes
then calculate the nbytes.
This change introduce a way to capture non-tensor data size within a write-plan item.
Test Plan: Existing UT.
Differential Revision: D71599725
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149699
Approved by: https://github.com/MeetVadakkanchery
In the old exporter we allow users to define a symbolic() method to bypass JIT tracing for a block of logic. We can allow users to do similar things by creating symbolic ops at export.
This PR implements `torch.onnx.ops.symbolic` and `torch.onnx.ops.symbolic_multi_out` to allow users to create onnx nodes symbolically with pt2 & fx. The custom pytorch ops were designed such that the attributes are encoded to be part of a valid fx op. Users provide shape and dtype for the meta function to produce the currect fake tensor during export.
An example is
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148905
Approved by: https://github.com/titaiwangms
This change does 2 important things:
(a) Instead of relying on IValue type as source of truth, we use the schema as the source of truth, which is important as IValue types are overloaded and can ambiguously convert incorrectly. For example, a MemoryFormat will look like an int + get converted to an int64_t vs a MemoryFormat!
(b) This PR expands support for many more types to encompass way more schemas, e.g., Optional, Device, dtype, etc. The main win from this PR is the ability for aoti_torch_call_dispatcher to call TensorFactory ops like ones_like/empty_like!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149052
Approved by: https://github.com/albanD
The sub-gradient of minimum norm is the least steep descent direction.
```python
import torch
x = torch.tensor([-2, -1, 0, 1, 2.], requires_grad=True)
torch.relu(x).sum().backward()
print(x.grad) # tensor([0., 0., 0., 1., 1.])
y = torch.tensor([-2, -1, 0, 1, 2.], requires_grad=True)
torch.abs(y).sum().backward()
print(y.grad) # tensor([-1., -1., 0., 1., 1.])
```
(How can I request a reviewer? I don't have the button on the right)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148658
Approved by: https://github.com/lezcano
This PR adds two main parts:
- shim.h stable C APIs into torch::Library APIs
- a higher level API in torch/csrc/stable/library.h that calls into this shim.h + otherwise is self contained
Goal: custom kernel writers should be able to call the apis in the directories above in order to register their library in a way that allows their custom extension to run with a different libtorch version than it was built with.
Subplots resolved:
- Do we want a whole separate StableLibrary or do we want to freeze torch::Library and add `m.stable_impl(cstring, void (*fn)(void **, int64_t, int64_t)` into it
- Yes, we want a separate StableLibrary. We cannot freeze Library and it is NOT header only.
- Should I use unint64_t as the common denominator instead of void* to support 32bit architectures better?
- Yes, and done
- Should I add a stable `def` and `fragment` when those can be done in python?
- I think we do want these --- and now they're done
- Where should library_stable_impl.cpp live? -- no longer relevant
- I need some solid test cases to make sure everything's going ok. I've intentionally thrown in a bunch of random dtypes into the signature, but I still haven't tested returning multiple things, returning nothing, complex dtypes, etc.
- Have since tested all the torch library endpoints. the others can be tested in a followup to separate components that need to be in shim.h vs can be added later
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148124
Approved by: https://github.com/albanD, https://github.com/zou3519, https://github.com/atalman
This PR adds two main parts:
- shim.h stable C APIs into torch::Library APIs
- a higher level API in torch/csrc/stable/library.h that calls into this shim.h + otherwise is self contained
Goal: custom kernel writers should be able to call the apis in the directories above in order to register their library in a way that allows their custom extension to run with a different libtorch version than it was built with.
Subplots resolved:
- Do we want a whole separate StableLibrary or do we want to freeze torch::Library and add `m.stable_impl(cstring, void (*fn)(void **, int64_t, int64_t)` into it
- Yes, we want a separate StableLibrary. We cannot freeze Library and it is NOT header only.
- Should I use unint64_t as the common denominator instead of void* to support 32bit architectures better?
- Yes, and done
- Should I add a stable `def` and `fragment` when those can be done in python?
- I think we do want these --- and now they're done
- Where should library_stable_impl.cpp live? -- no longer relevant
- I need some solid test cases to make sure everything's going ok. I've intentionally thrown in a bunch of random dtypes into the signature, but I still haven't tested returning multiple things, returning nothing, complex dtypes, etc.
- Have since tested all the torch library endpoints. the others can be tested in a followup to separate components that need to be in shim.h vs can be added later
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148124
Approved by: https://github.com/albanD, https://github.com/zou3519
Installing PyTorch from binaries will automatically install the runtime packages of Intel® Deep Learning Essentials. In this case, if we activate oneAPI in a standalone installation of Intel® Deep Learning Essentials, there will be an environment issue. Therefore, add a note to remind users to avoid this situation.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148168
Approved by: https://github.com/janeyx99
Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com>
Co-authored-by: Jane (Yuan) Xu <31798555+janeyx99@users.noreply.github.com>
This is an initial attempt to provide some statistics for the pinned host memory allocations flowing through CachingHostAllocator. Many times in the past we have had inexplicable slowdowns that would be much easier to diagnose if we had some host memory characteristics.
This change tries very hard not to disrupt the initial design of the allocator, and it uses existing locking mechanism, whenever possible, to gather statistics "for free". Only deviation from that is on the "slow path" where we incur CUDA calls anyway, so taking a short lock is not going to hurt the performance much, especially in the steady state where most allocations will come from cache.
As mentioned before, this is the first PR, to introduce the concept and to see if it fits the right paradigm. We can always add more later.
Metrics that would require more involved changes to the code base and locks, like requested memory, have been punted for now. I also tried to reuse the Stat structure used in CUDA caching allocator, in order to maintain symmetry.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147660
Approved by: https://github.com/ngimel
Summary:
### Context
Background checkpoint upload thread interfering with trainer thread:
In [async save API](https://github.com/pytorch/pytorch/blob/main/torch/distributed/checkpoint/state_dict_saver.py#L239-L248), the background thread spends a considerable amount of time on CPU-bound tasks (pickling/unpickling several metada objects a.k.a SavePlans) on rank0 during the collective operation; this kind of asymmetric computation heavily contends for GIL with the trainer thread causing GPU util to suffer significantly for the E2E checkpoint duration.
### Solution:
Introduce async save via a checkpoint daemon process. This daemon process will be created once (during the first save attempt) and can serve async checkpoint requests for the remainder of training lifetime.
Test Plan: Added E2E UTs for process based async save.
Differential Revision: D69272583
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147039
Approved by: https://github.com/saumishr
Summary:
Generate AOTI size and stride input check by default. But the checks are only run if `AOT_INDUCTOR_DEBUG_COMPILE` env variable is set (to avoid slowing down the performance).
Example output:
```cpp
bool _check_aoti_runtime_check_inputs_env() {
const static char* env_var_value = getenv("AOTI_RUNTIME_CHECK_INPUTS");
const static bool result = env_var_value != nullptr && env_var_value[0] != '\0';
return result;
}
AOTI_NOINLINE static void __check_inputs_outputs(
AtenTensorHandle* input_handles,
AtenTensorHandle* output_handles) {
if (!_check_aoti_runtime_check_inputs_env()){
return;
}
//rest of the check
}
```
Test Plan: CI
Differential Revision: D70260490
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148005
Approved by: https://github.com/hl475, https://github.com/desertfire, https://github.com/jingsh
This is an initial attempt to provide some statistics for the pinned host memory allocations flowing through CachingHostAllocator. Many times in the past we have had inexplicable slowdowns that would be much easier to diagnose if we had some host memory characteristics.
This change tries very hard not to disrupt the initial design of the allocator, and it uses existing locking mechanism, whenever possible, to gather statistics "for free". Only deviation from that is on the "slow path" where we incur CUDA calls anyway, so taking a short lock is not going to hurt the performance much, especially in the steady state where most allocations will come from cache.
As mentioned before, this is the first PR, to introduce the concept and to see if it fits the right paradigm. We can always add more later.
Metrics that would require more involved changes to the code base and locks, like requested memory, have been punted for now. I also tried to reuse the Stat structure used in CUDA caching allocator, in order to maintain symmetry.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147660
Approved by: https://github.com/ngimel
Fix a link to numpy documentation that has moved and now 404's
I"ve checked other numpy doc links that point to docs.scipy.org (which then redirects to numpy.org) and they do work, so I am fixing just this 404.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147697
Approved by: https://github.com/soulitzer
This patch adds support for sycl kernels build via `torch.utils.cpp_extension.load`, `torch.utils.cpp_extension.load_inline` and (new) `class SyclExtension` APIs. Files having `.sycl` extension are considered to have sycl kernels and are compiled with `icpx` (dpc++ sycl compiler from Intel). Files with other extensions, `.cpp`, `.cu`, are handled as before. API supports building sycl along with other file types into single extension.
Note that `.sycl` file extension is a PyTorch convention for files containing sycl code which I propose to adopt. We did follow up with compiler team to introduce such file extension in the compiler, but they are opposed to this. At the same time discussion around sycl file extension and adding sycl language support into such tools as cmake is ongoing. Eventually cmake also considers to introduce some file extension convention for sycl. I hope we can further influence cmake and compiler communities to broader adopt `.sycl` file extension.
By default SYCL kernels are compiled for all Intel GPU devices for which pytorch native aten SYCL kernels are compiled. At the moment `pvc,xe-lpg`. This behavior can be overridden by setting `TORCH_XPU_ARCH_LIST` environment variables to the comma separated list of desired devices to compile for.
Fixes: #132944
CC: @gujinghui @EikanWang @fengyuan14 @guangyey @jgong5
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132945
Approved by: https://github.com/albanD, https://github.com/guangyey, https://github.com/malfet
Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com>
This patch adds support for sycl kernels build via `torch.utils.cpp_extension.load`, `torch.utils.cpp_extension.load_inline` and (new) `class SyclExtension` APIs. Files having `.sycl` extension are considered to have sycl kernels and are compiled with `icpx` (dpc++ sycl compiler from Intel). Files with other extensions, `.cpp`, `.cu`, are handled as before. API supports building sycl along with other file types into single extension.
Note that `.sycl` file extension is a PyTorch convention for files containing sycl code which I propose to adopt. We did follow up with compiler team to introduce such file extension in the compiler, but they are opposed to this. At the same time discussion around sycl file extension and adding sycl language support into such tools as cmake is ongoing. Eventually cmake also considers to introduce some file extension convention for sycl. I hope we can further influence cmake and compiler communities to broader adopt `.sycl` file extension.
By default SYCL kernels are compiled for all Intel GPU devices for which pytorch native aten SYCL kernels are compiled. At the moment `pvc,xe-lpg`. This behavior can be overridden by setting `TORCH_XPU_ARCH_LIST` environment variables to the comma separated list of desired devices to compile for.
Fixes: #132944
CC: @gujinghui @EikanWang @fengyuan14 @guangyey @jgong5
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132945
Approved by: https://github.com/albanD, https://github.com/guangyey
Use it to unwrap any functorch-wrapped tensor. I don't recommend using
the output in a program since it breaks the semantics of the transforms,
but it seems useful for debugging.
I will note that some people have wanted to get intermediate values out
of an e.g. grad transform, so this might be a way to do that...
Test Plan:
- tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/146528
Approved by: https://github.com/Chillee
## Background
This PR adds `torch.utils.serialization.config.load.calculate_storage_offsets`. This option relies on the previous PR in this stack, where storage order was changed to non lexicographical. A `.format_version` entry was added to the zipfile and `calculate_storage_offsets` will only work on checkpoints with `.format_version`.
When this is turned on, for `torch.load(mmap=True)`, offsets of each storage record (other than the 0th storage will be calculated instead of relying on `miniz` APIs to determine this).
The existing APIs will issue multiple random reads (reading the end of central directory record, then reading the zipfile header for the record) to determine the storage offset where the record starts. This can greatly degrade `torch.load(mmap=True)` performance for non-filesystem cases.
6aaae9d78f/caffe2/serialize/inline_container.cc (L589-L605)
## How does this work
The format for the checkpoint is as such
```
archive_name/
|_ data.pkl
|_.format_version
|_byteorder
|_data/
|_ 0
|_ 1
|_ 2
|_ ...
|_
```
Each `data/i` record represents a storage, where storages are written in the order that the Pickler encounters them.
For each storage, our `persistent_load` logic saves the following metadata to the pickle file `dtype, numel, key, location` where `numel` is the number of bytes in the storage.
Note that we always use `miniz` writer in the zip64 mode per [here](7796e308d0/caffe2/serialize/inline_container.cc (L701)) A zipfile record written by miniz looks as such
```
---------------- ----------------- ------------------- ---------------- --------- ------------------------------
| 30 byte header | n byte filename | zip64_extra_data | m byte padding | storage | 16 or 24 byte local dir footer |
---------------- ----------------- ------------------- ---------------- --------- ------------------------------
```
- The header size (30) is given by [`MZ_ZIP_LOCAL_DIR_HEADER_SIZE`](https://github.com/pytorch/pytorch/blob/main/third_party/miniz-3.0.2/miniz.c?fbclid=IwZXh0bgNhZW0CMTEAAR2O8Vysd--UoSCxW70gabXIS1dbz733oHwuUQ5_Ff1hY2WU6PL2i6CSH4A_aem_J9oaU2HpDeWtJKOU9EnVqw#L3290)
- filename will be `"{archive_name}/{filepath}"`
- `zip64_extra_data` is determined by [`mz_zip_writer_create_zip64_extra_data`](7796e308d0/third_party/miniz-3.0.2/miniz.c (L6202)). Note that [we only create zip64_extra_data if storage_size >= 0xFFFFFFFF or the offset of the start of the header >= 0xFFFFFFFF](7796e308d0/third_party/miniz-3.0.2/miniz.c (L6519-L6524))
- `m` is determined by [`getPadding`](7796e308d0/caffe2/serialize/inline_container.cc (L254)), which accounts for filename, zip64_extra_data to determine `m` such that the start of `storage` is aligned to 64 bytes. The `m` bytes will always start with `F B padding_size" as the first 4 bytes
- The local dir footer size is determined based on [this snippet ](7796e308d0/third_party/miniz-3.0.2/miniz.c (L6610-L6632)): if the buffer size is 0 it is skipped. If the zip64_extra_data was created, it is 24, otherwise it is 16.
When `torch.utils.serialization.config.load.calculate_storage_offsets` is set we do the following
- We keep track of where the "cursor" is in the file using `current_offset`, after each persistent_load call, it will be at the offset where the header for the next record starts
- for the 0th storage, "data/0", we use the regular get_record_offset to determine the start of the storage
- for any other storage, (where the storages will be in order encountered by the unpickler, 0, 1, 2, 3, ...) we use `get_record_offset_no_read`, which re-uses the `getPadding` logic to determine the offset of the storage
- Note that `load_tensor` will only ever be called again with the same key if the storage's `._data_ptr()` is 0 [[pointer1](https://github.com/pytorch/pytorch/blob/main/torch/serialization.py#L1917-L1918)][[pointer2](https://github.com/pytorch/pytorch/blob/main/torch/serialization.py#L1936-L1937)], so we cache the offsets for this edge case
- After each storage, if the storage is non-zero, we account for the local dir footer based on the logic described above
## Testing strategy
The agreed upon testing strategy was as follows:
- Add debug code gated by an environment flag `TORCH_SERIALIZATION_DEBUG` that will run this offset calculation logic and verify it against getRecordOffset for each storage (when mmap=False)
- This flag is set throughout CI, which means that every time `torch.load` is called, the offset calculation logic is implicitly being tested.
Differential Revision: [D67673026](https://our.internmc.facebook.com/intern/diff/D67673026)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143880
Approved by: https://github.com/albanD
ghstack dependencies: #143879
Triton 2.2 and greater have a bug where allowing TF32 generation for a GPU that does not support TF32 will cause code generation errors. Patch around this problem by:
1. Adding a function to `torch.cuda` that determines whether CUDA hardware is capable of using the TF32 format.
2. Using that function to explicitly disable TF32 generation when calling Triton, where needed.
To demonstrate that this fix works, try running `test/inductor/test_max_autotune.py` on a GPU with CUDA compute capability < 8 (e.g. any NVIDIA consumer GPU) without this fix.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145684
Approved by: https://github.com/eqy
## Background
This PR adds `torch.utils.serialization.config.load.calculate_storage_offsets`. This option relies on the previous PR in this stack, where storage order was changed to non lexicographical. A `.format_version` entry was added to the zipfile and `calculate_storage_offsets` will only work on checkpoints with `.format_version`.
When this is turned on, for `torch.load(mmap=True)`, offsets of each storage record (other than the 0th storage will be calculated instead of relying on `miniz` APIs to determine this).
The existing APIs will issue multiple random reads (reading the end of central directory record, then reading the zipfile header for the record) to determine the storage offset where the record starts. This can greatly degrade `torch.load(mmap=True)` performance for non-filesystem cases.
6aaae9d78f/caffe2/serialize/inline_container.cc (L589-L605)
## Testing strategy
The agreed upon testing strategy was as follows:
- Add debug code gated by an environment flag `TORCH_SERIALIZATION_DEBUG` that will run this offset calculation logic and verify it against getRecordOffset for each storage (when mmap=False)
- This flag is set throughout CI, which means that every time `torch.load` is called, the offset calculation logic is implicitly being tested.
Differential Revision: [D67673026](https://our.internmc.facebook.com/intern/diff/D67673026)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143880
Approved by: https://github.com/albanD
ghstack dependencies: #143879
Updated nested tensor docs to be NJT-centric (instead of NST-centric). They now include:
* High-level description of NST vs. NJT + a recommendation to use NJT
* General NJT construction / usage
* torch.compile() integration w/ dynamic shapes
* Common errors and how to fix them
* Contribution guide
* Data layout / shape information (with diagram)
* Links to more extensive tutorials involving Transformers / SDPA / FlexAttention
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145402
Approved by: https://github.com/soulitzer
I.e. when `MTL_CAPTURE_ENABLED` environment variable is set to 1, one should be able to invoke wrap the code with `torch.mps.profiler.capture_metal` to generate gputrace for shaders invoked inside the context manager.
For example, code below:
```python
import torch
import os
def foo(x):
return x[:,::2].sin() + x[:, 1::2].cos()
if __name__ == "__main__":
os.environ["MTL_CAPTURE_ENABLED"] = "1"
x = torch.rand(32, 1024, device="mps")
with torch.mps.profiler.metal_capture("compiled_shader"):
torch.compile(foo)(x)
```
should capture the execution of a `torch.compile` generated shader
<img width="734" alt="image" src="https://github.com/user-attachments/assets/718ff64e-103b-4b11-b66c-c89cfc770b5d" />
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144561
Approved by: https://github.com/manuelcandales
ghstack dependencies: #144559, #144560
Replace https://github.com/pytorch/pytorch/pull/138947 for re-import.
Replaces https://github.com/ROCm/pytorch/pull/1592
This PR contains the initial implementation of SDPA with composable_kernel backend. The CK path can be forced by simply calling torch.backends.cuda.preferred_rocm_fa_library("ck"). Similarly, you can force the incumbent aotriton implementation by passing in "aotriton" or "default". As you'd expect, not setting this option will result in aotriton to be used as the backend. In the case of CK, if pytorch deems flash attention usable, then it will use the CK path in all the same places aotriton would have been used. This PR makes no changes to the heuristics which select which attention scheme to use (i.e. flash attention vs memory efficient attention vs math etc etc). It only gets called when flash attention is both enabled (via USE_FLASH_ATTENTION) and is selected at runtime by the existing heuristics.
Files located in pytorch/aten/src/ATen/native/transformers/hip/flash_attn/ck/mha* have been pulled from https://github.com/Dao-AILab/flash-attention courtesy of @tridao's hard work who is the co-author
NOTE: In order to use this backend, the user MUST set USE_CK_FLASH_ATTENTION=1 in their environment when they build PyTorch.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143695
Approved by: https://github.com/malfet
Co-authored-by: Andy Lugo <Andy.LugoReyes@amd.com>
Co-authored-by: Jithun Nair <jithun.nair@amd.com>
as titled, this PR expose this dunder method as a public API in the doc,
so that different checkpoint implementations can leverage this protocol,
instead of exposing a separate API
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144100
Approved by: https://github.com/awgu
ghstack dependencies: #144099
Changes by apply order:
1. Replace all `".."` and `os.pardir` usage with `os.path.dirname(...)`.
2. Replace nested `os.path.dirname(os.path.dirname(...))` call with `str(Path(...).parent.parent)`.
3. Reorder `.absolute()` ~/ `.resolve()`~ and `.parent`: always resolve the path first.
`.parent{...}.absolute()` -> `.absolute().parent{...}`
4. Replace chained `.parent x N` with `.parents[${N - 1}]`: the code is easier to read (see 5.)
`.parent.parent.parent.parent` -> `.parents[3]`
5. ~Replace `.parents[${N - 1}]` with `.parents[${N} - 1]`: the code is easier to read and does not introduce any runtime overhead.~
~`.parents[3]` -> `.parents[4 - 1]`~
6. ~Replace `.parents[2 - 1]` with `.parent.parent`: because the code is shorter and easier to read.~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129374
Approved by: https://github.com/justinchuby, https://github.com/malfet
