[AOTInductor] Update performance benchmark code #109560
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109560
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 83eaaec with merge base b91ba22 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
desertfire
requested changes
Sep 19, 2023
angelayi
force-pushed
the
angelayi/aot_inductor_benchmark
branch
2 times, most recently
from
35165da to
827bf15
Compare
desertfire
approved these changes
Sep 19, 2023
torch/_inductor/utils.py
Outdated
| void run( | ||
| std::vector<at::Tensor>& input_tensors, | ||
| std::vector<at::Tensor>& output_tensors) { | ||
| class AOTInductorModelContainer { |
There was a problem hiding this comment.
Please use a different name as this can cause confusion with
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 1 mandatory check(s) failed. The first few are: Dig deeper by viewing the failures on hud |
desertfire
force-pushed
the
angelayi/aot_inductor_benchmark
branch
from
b89ce6b to
83eaaec
Compare
|
@angelayi has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
pytorchmergebot
pushed a commit
that referenced
this pull request
Sep 21, 2023
Summary: Same as #109560, made a new PR because we need to land from internal Previously during performance benchmark testing, we would create an AOTInductorModelContainerHandle every time the compiled function is run with new inputs. However after #108473 we now load the constants needed in the runtime when initializing the AOTInductorModelContainerHandle. This resulted in our benchmarks displaying a ~0.4x speedup. This diff moves the initialization of AOTInductorModelContainerHandle outside of the code where we run the compiled function with different inputs. For example, ``` python benchmarks/dynamo/huggingface.py --performance --cold-start-latency --inference --bfloat16 --export-aot-inductor --disable-cudagraphs --device cuda --total-partitions 3 --partition-id 0 --only AlbertForMaskedLM ``` results in `1.359x` speedup. Specifically, this adds a `create_container_handle` and `delete_container_handle` function which need to called before `run`. We call `create_container_handle` to initialize the AOTInductorModelContainerHandle, call `run` to run the compiled .so with different inputs, and then `delete_container_handle` to delete it. [Updated dashboard results](https://hud.pytorch.org/benchmark/compilers?startTime=Wed%2C%2013%20Sep%202023%2021%3A03%3A55%20GMT&stopTime=Wed%2C%2020%20Sep%202023%2021%3A03%3A55%20GMT&granularity=hour&suite=torchbench&mode=inference&dtype=bfloat16&lBranch=angelayi/aot_inductor_benchmark&lCommit=f9aa49c4c9a1a140b6f0c4520d1d6d99b57e12fa&rBranch=main&rCommit=015be4cedba357eb931e24bf188479235db7c5c8) Test Plan: CI Differential Revision: D49513934 Pull Request resolved: #109820 Approved by: https://github.com/desertfire
pytorchmergebot
pushed a commit
that referenced
this pull request
Sep 22, 2023
Summary: Change AOTInductor to directly return output tensors instead of taking pre-allocated output tensors to return the results. This gives several benefits: * It makes sure AOTInductor has the same behavior when managing the output tensors as the default Inductor, which is widely tested and thus more reliable. * As we have debugged before, there are cases we still have to codegen extra copy_ ops to fill the pre-allocated output tensors which doesn't make sense for performance. * With the coming enhanced memory planning, this again will make sure the memory planning logic is the between AOTInductor and Inductor, which will greatly simplify the problem and improve the reliability. This change also combines D49494954 from Yang and #109560 from Angela. Differential Revision: D49502318 Pull Request resolved: #109790 Approved by: https://github.com/chenyang78
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously during performance benchmark testing, we would create an AOTInductorModelContainerHandle every time the compiled function is run with new inputs. However after #108473 we now load the constants needed in the runtime when initializing the AOTInductorModelContainerHandle. This resulted in our benchmarks displaying a ~0.4x speedup.
This diff moves the initialization of AOTInductorModelContainerHandle outside of the code where we run the compiled function with different inputs.
For example,
results in
1.359xspeedup.Specifically, this adds a
create_container_handleanddelete_container_handlefunction which need to called beforerun. We callcreate_container_handleto initialize the AOTInductorModelContainerHandle, callrunto run the compiled .so with different inputs, and thendelete_container_handleto delete it.Updated dashboard results
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @ngimel @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov