Initialize TileGym CI by arjkesh · Pull Request #4 · NVIDIA/TileGym

arjkesh · 2025-12-10T01:07:39Z

Description

Add build/test CI, code formatter (along with formatting), PR config, GHCR cleanup job

Relevant files:
There are a lot of formatting changes due to addition of darker formatter. The files below have functional changes

tests/benchmark/run_all.sh --> modify benchmark runner script to improve reporting
.github/* --> home of CI infra and associated tests
modeling/transformers/Dockerfile --> Restructure dockerfile to maximize build-over-build layer overlap

Features:

parallelize ops tests with xdist (added da1f37a)
debug parser logic (fixed 42c632f)
add junitxml reporting (added da1f37a)
debug image tagging (fixed 42c632f)
use darker instead of black (fixed in d653790)
speed up build time (caching and dockerfile refactoring, dropped build time by 60% da1f37a)
swap out main push flows for nightly (fixed in d653790)
Add tests for infra scripts (passing in 8de6ac0)
Add "verified" tag after build/test completes to signal that testing succeeded on the image (will run nightly)

CI Configuration

config:
  build: true
  # valid options are "ops" and "benchmark"
  test: ["ops", "benchmark"]

Checklist

Documentation updated (if needed)
CI configuration reviewed

arjkesh · 2025-12-10T01:15:22Z

/ok to test 6264b2c

xjmxyt · 2025-12-10T04:58:57Z

.github/workflows/docker-tests.yml

+            --gpus all \
+            -w /workspace/tilegym \
+            tilegym-transformers:latest \
+            pytest -s tests/ops -v -k test_op


We can try to run in parallel if it does not OOM.
Use pip install pytest-xdist and pytest -n job_number to automatically run tests in parallel.

copy-pr-bot · 2025-12-10T15:17:18Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

arjkesh · 2025-12-10T15:19:35Z

/ok to test 9c9e570

arjkesh · 2025-12-10T15:47:14Z

/ok to test 0987a19

arjkesh · 2025-12-10T16:08:04Z

/ok to test 1a2e7ce

arjkesh · 2025-12-10T17:50:41Z

Remaining changes

parallelize ops tests with xdist (added da1f37a)
parallelize benchmark tests (added da1f37a)
debug parser logic (fixed 42c632f)
add junitxml reporting (added da1f37a)
debug image tagging (fixed 42c632f)
use darker instead of black (fixed in d653790)
speed up build time (caching and dockerfile refactoring, dropped build time by 60% da1f37a)
swap out main push flows for nightly (fixed in d653790)

arjkesh · 2025-12-10T17:55:29Z

/ok to test cdeba44

arjkesh · 2025-12-10T18:02:47Z

/ok to test 42c632f

arjkesh · 2025-12-10T18:42:31Z

/ok to test da1f37a

arjkesh · 2025-12-10T19:29:52Z

/ok to test da1f37a

copy-pr-bot · 2025-12-10T19:29:56Z

/ok to test da1f37a

@arjkesh, there was an error processing your request: E2

See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/2/

arjkesh · 2025-12-10T19:42:56Z

/ok to test 173ca7b

arjkesh · 2025-12-10T22:18:15Z

/ok to test d653790

arjkesh · 2025-12-11T01:43:07Z

/ok to test 00e7f0e

arjkesh · 2025-12-11T02:21:55Z

/ok to test 5e36b90

tests/benchmark/run_all.sh

hannahli-nv · 2025-12-11T10:18:42Z

Could you please rename the README.md file in the .github directory or move it elsewhere? This will ensure that the README.md file in the root directory is displayed properly on the project's homepage.

hannahli-nv · 2025-12-11T10:19:23Z

Could you please rename the README.md file in the .github directory or move it elsewhere? This will ensure that the README.md file in the root directory is displayed properly on the project's homepage.

arjkesh · 2025-12-11T12:40:08Z

.github/workflows/tilegym-ci.yml

+
+env:
+  # PR images go to a temp repo, main/nightly go to main repo
+  IMAGE_NAME_PR: tilegym-transformers-pr


change to tilegym and tilegym-pr

…ame GHCR repos

arjkesh · 2025-12-11T15:02:23Z

/ok to test d3ba035

arjkesh · 2025-12-11T15:53:05Z

/ok to test c03370b

arjkesh · 2025-12-11T16:18:39Z

/ok to test 11c023b

arjkesh · 2025-12-12T04:06:07Z

/ok to test a2f303b

arjkesh · 2025-12-12T12:16:33Z

/ok to test 0daef68

arjkesh · 2025-12-12T12:17:44Z

/ok to test fd88ad2

This reverts commit 0daef68.

arjkesh · 2025-12-12T12:36:28Z

/ok to test a276ae1

xjmxyt

Overall LGTM.

arjkesh added 3 commits December 9, 2025 16:59

Init tilegym CI

82ddfdb

black code formatting

bb5f66c

Merge branch 'main' of github.com:NVIDIA/TileGym into tilegym_ci_init

6264b2c

azazhu previously approved these changes Dec 10, 2025

View reviewed changes

xjmxyt reviewed Dec 10, 2025

View reviewed changes

Test signed commit

9c9e570

update image names

0987a19

arjkesh dismissed azazhu’s stale review via 0987a19 December 10, 2025 15:45

NVIDIA deleted a comment from copy-pr-bot bot Dec 10, 2025

reduce image size

1a2e7ce

debug parsing logic

cdeba44

grab PR number from branch name

42c632f

Parallelize tests, utilize build caches, optimize dockerfile

da1f37a

empty trigger

173ca7b

arjkesh added 2 commits December 10, 2025 14:15

update linter, move pr images to separate repo

3d48592

use darker instead of black formatting

d653790

Add infra tests, modularize further

b91fde2

test trigger ci

5e36b90

xjmxyt reviewed Dec 11, 2025

View reviewed changes

tests/benchmark/run_all.sh Outdated Show resolved Hide resolved

arjkesh commented Dec 11, 2025

View reviewed changes

arjkesh added 2 commits December 11, 2025 06:57

sequential-ize benchmark tests, improve reporting, rename readme, ren…

63c6639

…ame GHCR repos

Merge branch 'main' of github.com:NVIDIA/TileGym into tilegym_ci_init

d3ba035

Fix benchmark reporting

c03370b

restructure benchmark file checkout, add logs

11c023b

arjkesh enabled auto-merge (squash) December 11, 2025 16:39

arjkesh added 3 commits December 11, 2025 19:55

reformat 120 line length

6806a7b

update

b66efcb

add line length 120 to dark checker

a2f303b

arjkesh disabled auto-merge December 12, 2025 04:17

revert dockerfile changes, install xdist in CI yaml

0daef68

retrigger build

fd88ad2

arjkesh added 2 commits December 12, 2025 04:30

Revert "revert dockerfile changes, install xdist in CI yaml"

8a19e9b

This reverts commit 0daef68.

go back to optimized dockerfile config

a276ae1

xjmxyt approved these changes Dec 12, 2025

View reviewed changes

arjkesh merged commit c15b012 into NVIDIA:main Dec 12, 2025
9 checks passed

arjkesh deleted the tilegym_ci_init branch December 12, 2025 14:36

Conversation

arjkesh commented Dec 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

CI Configuration

Checklist

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

xjmxyt Dec 10, 2025

Choose a reason for hiding this comment

Uh oh!

copy-pr-bot bot commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

copy-pr-bot bot commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 10, 2025

Uh oh!

arjkesh commented Dec 11, 2025

Uh oh!

arjkesh commented Dec 11, 2025

Uh oh!

Uh oh!

hannahli-nv commented Dec 11, 2025

Uh oh!

hannahli-nv commented Dec 11, 2025

Uh oh!

arjkesh Dec 11, 2025

Choose a reason for hiding this comment

Uh oh!

arjkesh commented Dec 11, 2025

Uh oh!

arjkesh commented Dec 11, 2025

Uh oh!

arjkesh commented Dec 11, 2025

Uh oh!

arjkesh commented Dec 12, 2025

Uh oh!

arjkesh commented Dec 12, 2025

Uh oh!

arjkesh commented Dec 12, 2025

Uh oh!

arjkesh commented Dec 12, 2025

Uh oh!

xjmxyt left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

arjkesh commented Dec 10, 2025 •

edited

Loading

arjkesh commented Dec 10, 2025 •

edited

Loading