Repository navigation
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/4527
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Oct 8, 2026
Contributor
Author
huydhn
added a commit
that referenced
this pull request
Oct 8, 2026
Moves the Linux test and lint jobs across 11 workflows onto OSDC pods. | before | after | |---|---| | `linux.g5.4xlarge.nvidia.gpu` | `mt-l-x86aavx2-11-41-a10g` | | `linux.g6.4xlarge.experimental.nvidia.gpu` | `mt-l-x86aavx2-11-41-l4` | | `linux.12xlarge` | `mt-l-x86iavx512-48-384` | | `linux.4xlarge` | `mt-l-x86iavx512-16-128` | Each pod sits on the same instance the label it replaces resolved to, so GPU model and count are unchanged and only the Kubernetes overhead comes off. `linux_job_v2` -> `v3` comes with it, per job: OSDC rejects a job with no container. The lint jobs set no `runner:` at all, so they were quietly taking v2's `linux.2xlarge` default. They now pin an OSDC pod explicitly. On OSDC the checkout runs inside the job container and no stock nvidia/cuda image ships git, so 42 jobs move to the images test-infra publishes for this -- the same bases plus git: nvidia/cuda:12.4.0-devel-ubuntu22.04 -> osdc-cuda:cuda12.4.1-cudnn-devel-ubuntu22.04 nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04 -> osdc-cuda:cuda13.0.3-cudnn-devel-ubuntu24.04 nvidia/cuda:12.8.x-*-ubuntu22.04 -> osdc-cuda:cuda12.8.1-cudnn-devel-ubuntu24.04 nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 -> osdc-cuda:cuda11.8.0-cudnn8-devel-ubuntu22.04 pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel -> osdc-cuda:cuda12.9.2-cudnn-devel-ubuntu24.04 unittests-robohive keeps nvidia/cudagl:11.4.0-base -- it is the only image providing the EGL stack that suite renders with -- so it runs standalone with an explicit git install instead of through v3. test-linux-libs.yml's `unittests-isaaclab` stays on EC2: it calls a fork's reusable workflow that cannot be confirmed to set a container. docs.yml is handled in #4524. Wheel builds are in #4487. Authored with Claude Code. ghstack-source-id: b70316e Pull-Request: #4527
huydhn
added a commit
that referenced
this pull request
Oct 8, 2026
Moves the Linux test and lint jobs across 11 workflows onto OSDC pods. | before | after | |---|---| | `linux.g5.4xlarge.nvidia.gpu` | `mt-l-x86aavx2-11-41-a10g` | | `linux.g6.4xlarge.experimental.nvidia.gpu` | `mt-l-x86aavx2-11-41-l4` | | `linux.12xlarge` | `mt-l-x86iavx512-48-384` | | `linux.4xlarge` | `mt-l-x86iavx512-16-128` | Each pod sits on the same instance the label it replaces resolved to, so GPU model and count are unchanged and only the Kubernetes overhead comes off. `linux_job_v2` -> `v3` comes with it, per job: OSDC rejects a job with no container. The lint jobs set no `runner:` at all, so they were quietly taking v2's `linux.2xlarge` default. They now pin an OSDC pod explicitly. On OSDC the checkout runs inside the job container and no stock nvidia/cuda image ships git, so 42 jobs move to the images test-infra publishes for this -- the same bases plus git: nvidia/cuda:12.4.0-devel-ubuntu22.04 -> osdc-cuda:cuda12.4.1-cudnn-devel-ubuntu22.04 nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04 -> osdc-cuda:cuda13.0.3-cudnn-devel-ubuntu24.04 nvidia/cuda:12.8.x-*-ubuntu22.04 -> osdc-cuda:cuda12.8.1-cudnn-devel-ubuntu24.04 nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 -> osdc-cuda:cuda11.8.0-cudnn8-devel-ubuntu22.04 pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel -> osdc-cuda:cuda12.9.2-cudnn-devel-ubuntu24.04 unittests-robohive keeps nvidia/cudagl:11.4.0-base -- it is the only image providing the EGL stack that suite renders with -- so it runs standalone with an explicit git install instead of through v3. test-linux-libs.yml's `unittests-isaaclab` stays on EC2: it calls a fork's reusable workflow that cannot be confirmed to set a container. docs.yml is handled in #4524. Wheel builds are in #4487. Authored with Claude Code. ghstack-source-id: b70316e Pull-Request: #4527
huydhn
added a commit
that referenced
this pull request
Oct 9, 2026
Moves the Linux test and lint jobs across 11 workflows onto OSDC pods. | before | after | |---|---| | `linux.g5.4xlarge.nvidia.gpu` | `mt-l-x86aavx2-11-41-a10g` | | `linux.g6.4xlarge.experimental.nvidia.gpu` | `mt-l-x86aavx2-11-41-l4` | | `linux.12xlarge` | `mt-l-x86iavx512-48-384` | | `linux.4xlarge` | `mt-l-x86iavx512-16-128` | Each pod sits on the same instance the label it replaces resolved to, so GPU model and count are unchanged and only the Kubernetes overhead comes off. `linux_job_v2` -> `v3` comes with it, per job: OSDC rejects a job with no container. The lint jobs set no `runner:` at all, so they were quietly taking v2's `linux.2xlarge` default. They now pin an OSDC pod explicitly. On OSDC the checkout runs inside the job container and no stock nvidia/cuda image ships git, so 42 jobs move to the images test-infra publishes for this -- the same bases plus git: nvidia/cuda:12.4.0-devel-ubuntu22.04 -> osdc-cuda:cuda12.4.1-cudnn-devel-ubuntu22.04 nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04 -> osdc-cuda:cuda13.0.3-cudnn-devel-ubuntu24.04 nvidia/cuda:12.8.x-*-ubuntu22.04 -> osdc-cuda:cuda12.8.1-cudnn-devel-ubuntu24.04 nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 -> osdc-cuda:cuda11.8.0-cudnn8-devel-ubuntu22.04 pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel -> osdc-cuda:cuda12.9.2-cudnn-devel-ubuntu24.04 unittests-robohive keeps nvidia/cudagl:11.4.0-base -- it is the only image providing the EGL stack that suite renders with -- so it runs standalone with an explicit git install instead of through v3. test-linux-libs.yml's `unittests-isaaclab` stays on EC2: it calls a fork's reusable workflow that cannot be confirmed to set a container. docs.yml is handled in #4524. Wheel builds are in #4487. Authored with Claude Code. ghstack-source-id: b70316e Pull-Request: #4527
huydhn
added a commit
that referenced
this pull request
Oct 9, 2026
Moves the Linux test and lint jobs across 11 workflows onto OSDC pods. | before | after | |---|---| | `linux.g5.4xlarge.nvidia.gpu` | `mt-l-x86aavx2-11-41-a10g` | | `linux.g6.4xlarge.experimental.nvidia.gpu` | `mt-l-x86aavx2-11-41-l4` | | `linux.12xlarge` | `mt-l-x86iavx512-48-384` | | `linux.4xlarge` | `mt-l-x86iavx512-16-128` | Each pod sits on the same instance the label it replaces resolved to, so GPU model and count are unchanged and only the Kubernetes overhead comes off. `linux_job_v2` -> `v3` comes with it, per job: OSDC rejects a job with no container. The lint jobs set no `runner:` at all, so they were quietly taking v2's `linux.2xlarge` default. They now pin an OSDC pod explicitly. On OSDC the checkout runs inside the job container and no stock nvidia/cuda image ships git, so 42 jobs move to the images test-infra publishes for this -- the same bases plus git: nvidia/cuda:12.4.0-devel-ubuntu22.04 -> osdc-cuda:cuda12.4.1-cudnn-devel-ubuntu22.04 nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04 -> osdc-cuda:cuda13.0.3-cudnn-devel-ubuntu24.04 nvidia/cuda:12.8.x-*-ubuntu22.04 -> osdc-cuda:cuda12.8.1-cudnn-devel-ubuntu24.04 nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 -> osdc-cuda:cuda11.8.0-cudnn8-devel-ubuntu22.04 pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel -> osdc-cuda:cuda12.9.2-cudnn-devel-ubuntu24.04 unittests-robohive keeps nvidia/cudagl:11.4.0-base -- it is the only image providing the EGL stack that suite renders with -- so it runs standalone with an explicit git install instead of through v3. test-linux-libs.yml's `unittests-isaaclab` stays on EC2: it calls a fork's reusable workflow that cannot be confirmed to set a container. docs.yml is handled in #4524. Wheel builds are in #4487. Authored with Claude Code. ghstack-source-id: b70316e Pull-Request: #4527
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Stack from ghstack (oldest at bottom):
Moves the Linux test and lint jobs across 11 workflows onto OSDC pods.
linux.g5.4xlarge.nvidia.gpumt-l-x86aavx2-11-41-a10glinux.g6.4xlarge.experimental.nvidia.gpumt-l-x86aavx2-11-41-l4linux.12xlargemt-l-x86iavx512-48-384linux.4xlargemt-l-x86iavx512-16-128Each pod sits on the same instance the label it replaces resolved to, so GPU
model and count are unchanged and only the Kubernetes overhead comes off.
linux_job_v2->v3comes with it, per job: OSDC rejects a job with nocontainer.
The lint jobs set no
runner:at all, so they were quietly taking v2'slinux.2xlargedefault. They now pin an OSDC pod explicitly.On OSDC the checkout runs inside the job container and no stock nvidia/cuda
image ships git, so 42 jobs move to the images test-infra publishes for
this -- the same bases plus git:
nvidia/cuda:12.4.0-devel-ubuntu22.04 -> osdc-cuda:cuda12.4.1-cudnn-devel-ubuntu22.04
nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04 -> osdc-cuda:cuda13.0.3-cudnn-devel-ubuntu24.04
nvidia/cuda:12.8.x-*-ubuntu22.04 -> osdc-cuda:cuda12.8.1-cudnn-devel-ubuntu24.04
nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 -> osdc-cuda:cuda11.8.0-cudnn8-devel-ubuntu22.04
pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel -> osdc-cuda:cuda12.9.2-cudnn-devel-ubuntu24.04
unittests-robohive keeps nvidia/cudagl:11.4.0-base -- it is the only image
providing the EGL stack that suite renders with -- so it runs standalone
with an explicit git install instead of through v3.
test-linux-libs.yml's
unittests-isaaclabstays on EC2: it calls a fork'sreusable workflow that cannot be confirmed to set a container.
docs.yml is handled in #4524. Wheel builds are in #4487.
Authored with Claude Code.