-
Notifications
You must be signed in to change notification settings - Fork 4.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add a new tutorial for compile time caching #3277
base: main
Are you sure you want to change the base?
Changes from 4 commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -691,3 +691,7 @@ XPU | |
XPUs | ||
impl | ||
overrideable | ||
TorchServe | ||
Inductor’s | ||
onwards | ||
recompilations |
Original file line number | Diff line number | Diff line change | ||||||||
---|---|---|---|---|---|---|---|---|---|---|
@@ -0,0 +1,76 @@ | ||||||||||
Compile Time Caching Configurations | ||||||||||
========================================================= | ||||||||||
**Authors:** `Oguz Ulgen <https://github.com/oulgen>`_ and `Sam Larsen <https://github.com/masnesral>`_ | ||||||||||
|
||||||||||
Introduction | ||||||||||
------------------ | ||||||||||
|
||||||||||
PyTorch Compiler implements several caches to reduce compilation latency. | ||||||||||
This recipe demonstrates how you can configure various parts of the caching in ``torch.compile``. | ||||||||||
|
||||||||||
Prerequisites | ||||||||||
------------------- | ||||||||||
|
||||||||||
Before starting this recipe, make sure that you have the following: | ||||||||||
|
||||||||||
* Basic understanding of ``torch.compile``. See: | ||||||||||
|
||||||||||
* `torch.compiler API documentation <https://pytorch.org/docs/stable/torch.compiler.html#torch-compiler>`__ | ||||||||||
* `Introduction to torch.compile <https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html>`__ | ||||||||||
* `Compile Time Caching in torch.compile <https://pytorch.org/tutorials/recipes/torch_compile_caching_tutorial.html>`__ | ||||||||||
|
||||||||||
* PyTorch 2.4 or later | ||||||||||
|
||||||||||
Inductor Cache Settings | ||||||||||
---------------------------- | ||||||||||
|
||||||||||
Most of these caches are in-memory, only used within the same process, and are transparent to the user. An exception is caches that store compiled FX graphs (``FXGraphCache``, ``AOTAutogradCache``). These caches allow Inductor to avoid recompilation across process boundaries when it encounters the same graph with the same Tensor input shapes (and the same configuration). The default implementation stores compiled artifacts in the system temp directory. An optional feature also supports sharing those artifacts within a cluster by storing them in a Redis database. | ||||||||||
|
||||||||||
There are a few settings relevant to caching and to FX graph caching in particular. | ||||||||||
The settings are accessible via environment variables listed below or can be hard-coded in Inductor’s config file. | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_FX_GRAPH_CACHE | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
This setting enables the local FX graph cache feature, i.e., by storing artifacts in the host’s temp directory. ``1`` enables, and any other value disables it. By default, the disk location is per username, but users can enable sharing across usernames by specifying ``TORCHINDUCTOR_CACHE_DIR`` (below). | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_AUTOGRAD_CACHE | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
This setting extends FXGraphCache to store cached results at the AOTAutograd level, instead of at the Inductor level. ``1`` enables, and any other value disables it. | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
By default, the disk location is per username, but users can enable sharing across usernames by specifying ``TORCHINDUCTOR_CACHE_DIR`` (below). | ||||||||||
`TORCHINDUCTOR_AUTOGRAD_CACHE` requires `TORCHINDUCTOR_FX_GRAPH_CACHE` to work. The same cache dir stores cache entries for ``AOTAutogradCache`` (under `{TORCHINDUCTOR_CACHE_DIR}/aotautograd`) and ``FXGraphCache`` (under `{TORCHINDUCTOR_CACHE_DIR}/fxgraph`). | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_CACHE_DIR | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
This setting specifies the location of all on-disk caches. By default, the location is in the system temp directory under ``torchinductor_<username>``, for example, ``/tmp/torchinductor_myusername``. | ||||||||||
|
||||||||||
Note that if ``TRITON_CACHE_DIR`` is not set in the environment, Inductor sets the ``Triton`` cache directory to this same temp location, under the Triton sub-directory. | ||||||||||
|
||||||||||
TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
This setting enables the remote FX graph cache feature. The current implementation uses ``Redis``. ``1`` enables caching, and any other value disables it. The following environment variables configure the host and port of the Redis server: | ||||||||||
|
||||||||||
``TORCHINDUCTOR_REDIS_HOST`` (defaults to ``localhost``) | ||||||||||
``TORCHINDUCTOR_REDIS_PORT`` (defaults to ``6379``) | ||||||||||
|
||||||||||
Note that if Inductor locates a remote cache entry, it stores the compiled artifact in the local on-disk cache; that local artifact would be served on subsequent runs on the same machine. | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_AUTOGRAD_REMOTE_CACHE | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
Like ``TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE``, this setting enables the remote ``AOTAutogradCache`` feature. The current implementation uses Redis. ``1`` enables caching, and any other value disables it. The following environment variables configure the host and port of the ``Redis`` server: | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
``TORCHINDUCTOR_REDIS_HOST`` (defaults to ``localhost``) | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
``TORCHINDUCTOR_REDIS_PORT`` (defaults to ``6379``) | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
`TORCHINDUCTOR_AUTOGRAD_REMOTE_CACHE`` depends on ``TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE`` to be enabled to work. The same Redis server can store both AOTAutograd and FXGraph cache results. | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_AUTOTUNE_REMOTE_CACHE | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
This setting enables a remote cache for ``TorchInductor``’s autotuner. As with the remote FX graph cache, the current implementation uses Redis. ``1`` enables caching, and any other value disables it. The same host / port environment variables listed above apply to this cache. | ||||||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||||||
|
||||||||||
TORCHINDUCTOR_FORCE_DISABLE_CACHES | ||||||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||||||
Set this value to ``1`` to disable all Inductor caching. This setting is useful for tasks like experimenting with cold-start compile times or forcing recompilation for debugging purposes. | ||||||||||
|
||||||||||
Conclusion | ||||||||||
------------- | ||||||||||
In this recipe, we have learned how to configure PyTorch Compiler's caching mechanisms. Additionally, we explored the various settings and environment variables that allow users to configure and optimize these caching features according to their specific needs. | ||||||||||
|
Original file line number | Diff line number | Diff line change | ||||
---|---|---|---|---|---|---|
@@ -1,12 +1,16 @@ | ||||||
Compile Time Caching in ``torch.compile`` | ||||||
========================================================= | ||||||
**Authors:** `Oguz Ulgen <https://github.com/oulgen>`_ and `Sam Larsen <https://github.com/masnesral>`_ | ||||||
**Author:** `Oguz Ulgen <https://github.com/oulgen>`_ | ||||||
|
||||||
Introduction | ||||||
------------------ | ||||||
|
||||||
PyTorch Inductor implements several caches to reduce compilation latency. | ||||||
This recipe demonstrates how you can configure various parts of the caching in ``torch.compile``. | ||||||
PyTorch Compiler provides several caching offerings to reduce compilation latency. | ||||||
oulgen marked this conversation as resolved.
Show resolved
Hide resolved
|
||||||
This recipe will explain these offerings in detail to help users pick the best option for their use case. | ||||||
|
||||||
Check out `Compile Time Caching Configurations <https://pytorch.org/tutorials/recipes/torch_compile_caching_configuration_tutorial.html>` for how to configure these caches. | ||||||
|
||||||
Also check out our caching benchmark at `PT CacheBench Benchmarks <https://hud.pytorch.org/benchmark/llms?repoName=pytorch%2Fpytorch&benchmarkName=TorchCache+Benchmark>`. | ||||||
|
||||||
Prerequisites | ||||||
------------------- | ||||||
|
@@ -17,60 +21,83 @@ Before starting this recipe, make sure that you have the following: | |||||
|
||||||
* `torch.compiler API documentation <https://pytorch.org/docs/stable/torch.compiler.html#torch-compiler>`__ | ||||||
* `Introduction to torch.compile <https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html>`__ | ||||||
* `Triton language documentation <https://triton-lang.org/main/index.html>`__ | ||||||
|
||||||
* PyTorch 2.4 or later | ||||||
|
||||||
Inductor Cache Settings | ||||||
---------------------------- | ||||||
Caching Offerings | ||||||
--------------------- | ||||||
|
||||||
``torch.compile`` provides the following caching offerings: | ||||||
|
||||||
* End to end caching (also known as ``Mega-Cache``) | ||||||
* Modular caching of ``TorchDynamo``, ``TorchInductor``, and ``Triton`` | ||||||
|
||||||
It is important to note that caching validates that the cache artifacts are used with the same PyTorch and Triton version, as well as, same GPU when device is set to be cuda. | ||||||
|
||||||
``torch.compile`` end-to-end caching (a.k.a. ``Mega-Cache``) | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
------------------------------------------------------------ | ||||||
|
||||||
End to end caching, from here onwards referred to Mega-Cache, is the ideal solution for users looking for a portable caching solution that can be stored in a database and can later be fetched possibly on a separate machine. | ||||||
|
||||||
``Mega-Cache`` provides two compiler APIs | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
* ``torch.compiler.save_cache_artifacts()`` | ||||||
* ``torch.compiler.load_cache_artifacts()`` | ||||||
|
||||||
The intended use case is after compiling and executing a model, the user calls ``torch.compiler.save_cache_artifacts()`` which will return the compiler artifacts in a portable form. Later, potentially on a different machine, the user may call ``torch.compiler.load_cache_artifacts()`` with these artifacts to pre-populate the ``torch.compile`` caches in order to jump-start their cache. | ||||||
|
||||||
An example to this is as follows. First, compile and save the cache artifacts. | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
.. code-block:: python | ||||||
|
||||||
@torch.compile | ||||||
def fn(x, y): | ||||||
return x.sin() @ y | ||||||
|
||||||
a = torch.rand(100, 100, dtype=dtype, device=device) | ||||||
b = torch.rand(100, 100, dtype=dtype, device=device) | ||||||
|
||||||
result = fn(a, b) | ||||||
|
||||||
artifacts = torch.compiler.save_cache_artifacts() | ||||||
|
||||||
# Now, potentially store these artifacts in a database | ||||||
|
||||||
Later, the user can jump-start their cache by the following. | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
Most of these caches are in-memory, only used within the same process, and are transparent to the user. An exception is caches that store compiled FX graphs (FXGraphCache, AOTAutogradCache). These caches allow Inductor to avoid recompilation across process boundaries when it encounters the same graph with the same Tensor input shapes (and the same configuration). The default implementation stores compiled artifacts in the system temp directory. An optional feature also supports sharing those artifacts within a cluster by storing them in a Redis database. | ||||||
.. code-block:: python | ||||||
|
||||||
There are a few settings relevant to caching and to FX graph caching in particular. | ||||||
The settings are accessible via environment variables listed below or can be hard-coded in Inductor’s config file. | ||||||
# Potentially download/fetch the artifacts from the database | ||||||
assert artifacts is not None | ||||||
artifact_bytes, cache_info = artifacts | ||||||
|
||||||
TORCHINDUCTOR_FX_GRAPH_CACHE | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
This setting enables the local FX graph cache feature, i.e., by storing artifacts in the host’s temp directory. ``1`` enables, and any other value disables it. By default, the disk location is per username, but users can enable sharing across usernames by specifying ``TORCHINDUCTOR_CACHE_DIR`` (below). | ||||||
torch.compiler.load_cache_artifacts(artifact_bytes) | ||||||
|
||||||
TORCHINDUCTOR_AUTOGRAD_CACHE | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
This setting extends FXGraphCache to store cached results at the AOTAutograd level, instead of at the Inductor level. ``1`` enables, and any other value disables it. | ||||||
By default, the disk location is per username, but users can enable sharing across usernames by specifying ``TORCHINDUCTOR_CACHE_DIR`` (below). | ||||||
`TORCHINDUCTOR_AUTOGRAD_CACHE` requires `TORCHINDUCTOR_FX_GRAPH_CACHE` to work. The same cache dir stores cache entries for AOTAutogradCache (under `{TORCHINDUCTOR_CACHE_DIR}/aotautograd`) and FXGraphCache (under `{TORCHINDUCTOR_CACHE_DIR}/fxgraph`). | ||||||
This operation populates all the modular caches that will be discussed in the next section, including ``PGO``, ``AOTAutograd``, ``Inductor``, ``Triton``, and ``Autotuning``. | ||||||
|
||||||
TORCHINDUCTOR_CACHE_DIR | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
This setting specifies the location of all on-disk caches. By default, the location is in the system temp directory under ``torchinductor_<username>``, for example, ``/tmp/torchinductor_myusername``. | ||||||
|
||||||
Note that if ``TRITON_CACHE_DIR`` is not set in the environment, Inductor sets the Triton cache directory to this same temp location, under the Triton subdirectory. | ||||||
Modular caching of ``TorchDynamo``, ``TorchInductor``, and ``Triton`` | ||||||
----------------------------------------------------------- | ||||||
|
||||||
TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
This setting enables the remote FX graph cache feature. The current implementation uses Redis. ``1`` enables caching, and any other value disables it. The following environment variables configure the host and port of the Redis server: | ||||||
The above described MegaCache is also compromised of individual components that can be used without any user intervention. By default, PyTorch Compiler comes with local on-disk caches for ``TorchDynamo``, ``TorchInductor``, and ``Triton``. These caches are as following. | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
``TORCHINDUCTOR_REDIS_HOST`` (defaults to ``localhost``) | ||||||
``TORCHINDUCTOR_REDIS_PORT`` (defaults to ``6379``) | ||||||
* ``FXGraphCache``: cache of graph-based IR components used in compilation | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
* ``Triton Cache``: cache of Triton-compilation results (``cubin`` files generated by ``Triton`` as well as other caching artifacts) | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
* ``InductorCache``: bundling of ``FXGraphCache`` and ``Triton`` cache | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
* ``AOTAutogradCache``: caching of joint graph artifacts | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
* ``PGO-cache``: cache of dynamic shape decisions to reduce number of recompilations | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
Note that if Inductor locates a remote cache entry, it stores the compiled artifact in the local on-disk cache; that local artifact would be served on subsequent runs on the same machine. | ||||||
All these cache artifacts are written to ``TORCHINDUCTOR_CACHE_DIR`` which by default will look like ``/tmp/torchinductor_myusername``. | ||||||
|
||||||
TORCHINDUCTOR_AUTOGRAD_REMOTE_CACHE | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
Like TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE, this setting enables the remote AOT AutogradCache feature. The current implementation uses Redis. ``1`` enables caching, and any other value disables it. The following environment variables configure the host and port of the Redis server: | ||||||
``TORCHINDUCTOR_REDIS_HOST`` (defaults to ``localhost``) | ||||||
``TORCHINDUCTOR_REDIS_PORT`` (defaults to ``6379``) | ||||||
|
||||||
`TORCHINDUCTOR_AUTOGRAD_REMOTE_CACHE`` depends on `TORCHINDUCTOR_FX_GRAPH_REMOTE_CACHE` to be enabled to work. The same Redis server can store both AOTAutograd and FXGraph cache results. | ||||||
Remote Caching | ||||||
---------------- | ||||||
|
||||||
TORCHINDUCTOR_AUTOTUNE_REMOTE_CACHE | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
This setting enables a remote cache for Inductor’s autotuner. As with the remote FX graph cache, the current implementation uses Redis. ``1`` enables caching, and any other value disables it. The same host / port environment variables listed above apply to this cache. | ||||||
We also provide a remote caching option for users who would like to take advantage of a ``Redis`` based cache. Check out `Compile Time Caching Configurations <https://pytorch.org/tutorials/recipes/torch_compile_caching_configuration_tutorial.html>` to learn more about how to enable the ``Redis`` based caching. | ||||||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
Suggested change
|
||||||
|
||||||
TORCHINDUCTOR_FORCE_DISABLE_CACHES | ||||||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||||||
Set this value to ``1`` to disable all Inductor caching. This setting is useful for tasks like experimenting with cold-start compile times or forcing recompilation for debugging purposes. | ||||||
|
||||||
Conclusion | ||||||
------------- | ||||||
In this recipe, we have learned that PyTorch Inductor's caching mechanisms significantly reduce compilation latency by utilizing both local and remote caches, which operate seamlessly in the background without requiring user intervention. | ||||||
Additionally, we explored the various settings and environment variables that allow users to configure and optimize these caching features according to their specific needs. | ||||||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.