8000 [Regression] Local performance regression by hjjq · Pull Request #321 · hidet-org/hidet · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

[Regression] Local performance regression #321

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 15 commits into from
Jul 20, 2023
Merged

[Regression] Local performance regression #321

merged 15 commits into from
Jul 20, 2023

Conversation

hjjq
Copy link
Collaborator
@hjjq hjjq commented Jul 19, 2023

python ./scripts/regression/run.py -h at the repo root for usage.
Currently only some operator shapes and models, with 3090 latency data, are added. More operator/shapes/models/GPUs to be added.
Tolerance to report speedup/slowdown is set at +-8%. May adjust in the future accordingly.

yaoyaoding
yaoyaoding reviewed Jul 19, 2023
View reviewed changes
Copy link
Collaborator
@yaoyaoding yaoyaoding left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @hjjq, thanks!

Just one suggestion. Can we store the shapes to profile in a seperate json file instead of using the regression_data.json, which also records the profile data for previous runs.

  • configs.json: store all the shape information that we want to benchmark. (This information can also be written in the python source code directly).
  • regression_data.json: (optional) store the benchmark data for previous run, if have. We can generate one for the first run.

@hjjq
Copy link
Collaborator Author
hjjq commented Jul 19, 2023

Hi @yaoyaoding , I generated this current version of regression_data.json with a local run on my 3090.
What I did was similar to what triton did 8000 , where the latency numbers are stored in the repository together with the shapes.
I can create configs.json as you described, but regression_data.json will still be needed in order to compare against a reference run, and the reference latencies need to be updated in the repo every time we achieve a speedup (It will always contain the all-time-high peformance that Hidet can achieve on a particular GPU.)

@yaoyaoding
Copy link
Collaborator

I see, that makes sense.

@yaoyaoding yaoyaoding merged commit 73b7fdb into hidet-org:main Jul 20, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@yaoyaoding yaoyaoding yaoyaoding approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@hjjq @yaoyaoding
0