Bumping the remaining stack for transformers v5 by terrykong · Pull Request #2058 · NVIDIA-NeMo/RL

terrykong · 2026-03-03T23:39:29Z

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Issues

List issues that this PR closes (syntax):

Usage

You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

Make sure you read and followed Contributor guidelines
Did you write any new necessary tests?
Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

...

github-actions · 2026-03-03T23:41:31Z

❌ Submodule Fast-Forward Check Failed

Check based on commit: 97c40d2 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

❌ Submodules that need attention:

Megatron-LM: ❌ PR branch is BEHIND hemil/automodel-transformers-v5 branch
TARGET (hemil/automodel-transformers-v5 branch): https://github.com/yaoyu-33/Megatron-LM/commits/193463c4f8414e6906a40dd527a450bca50706b1/
CURRENT (PR #2058 from tk/vllm-v5): https://github.com/yaoyu-33/Megatron-LM/commits/bbbedbb9f53343762e4dc70abc771b813a83d817/

Please ensure all submodule commits are fast-forwards of the hemil/automodel-transformers-v5 branch before merging.

hemildesai · 2026-03-03T23:48:28Z

nemo_rl/distributed/virtual_cluster.py


    # Use NeMo-RL direct dependencies and nemo-automodel.
-    AUTOMODEL = f"uv run --locked --no-cache --extra automodel --no-group transformers-v4 --directory {git_root}"
+    AUTOMODEL = f"uv run --locked --no-cache --extra automodel --directory {git_root}"


--no-cache shouldn't be required with TE 2.12 I believe

github-actions · 2026-03-04T07:50:18Z

❌ Submodule Fast-Forward Check Failed

Check based on commit: caed2de (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

❌ Submodules that need attention:

Megatron-LM: ❌ Commits have DIVERGED from a common ancestor
TARGET (hemil/automodel-transformers-v5 branch): https://github.com/yaoyu-33/Megatron-LM/commits/193463c4f8414e6906a40dd527a450bca50706b1/
CURRENT (PR #2058 from tk/vllm-v5): https://github.com/yaoyu-33/Megatron-LM/commits/23dd639cf3de30f3b9d8d0fae71ee31180be9ddd/

Please ensure all submodule commits are fast-forwards of the hemil/automodel-transformers-v5 branch before merging.

github-actions · 2026-03-04T08:21:22Z

❌ Submodule Fast-Forward Check Failed

Check based on commit: 3eb8634 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

❌ Submodules that need attention:

Megatron-LM: ❌ Commits have DIVERGED from a common ancestor
TARGET (hemil/automodel-transformers-v5 branch): https://github.com/yaoyu-33/Megatron-LM/commits/193463c4f8414e6906a40dd527a450bca50706b1/
CURRENT (PR #2058 from tk/vllm-v5): https://github.com/yaoyu-33/Megatron-LM/commits/23dd639cf3de30f3b9d8d0fae71ee31180be9ddd/

Please ensure all submodule commits are fast-forwards of the hemil/automodel-transformers-v5 branch before merging.

github-actions · 2026-03-04T08:53:37Z

ℹ️ File Consistency Check

Check based on commit: d10dfee (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T08:54:33Z

❌ Submodule Fast-Forward Check Failed

Check based on commit: d10dfee (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

❌ Submodules that need attention:

Megatron-LM: ❌ Commits have DIVERGED from a common ancestor
TARGET (hemil/automodel-transformers-v5 branch): https://github.com/yaoyu-33/Megatron-LM/commits/193463c4f8414e6906a40dd527a450bca50706b1/
CURRENT (PR #2058 from tk/vllm-v5): https://github.com/yaoyu-33/Megatron-LM/commits/23dd639cf3de30f3b9d8d0fae71ee31180be9ddd/

Please ensure all submodule commits are fast-forwards of the hemil/automodel-transformers-v5 branch before merging.

github-actions · 2026-03-04T09:16:59Z

ℹ️ File Consistency Check

Check based on commit: f40891a (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T09:17:32Z

❌ Submodule Fast-Forward Check Failed

Check based on commit: f40891a (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

❌ Submodules that need attention:

Megatron-LM: ❌ Commits have DIVERGED from a common ancestor
TARGET (hemil/automodel-transformers-v5 branch): https://github.com/yaoyu-33/Megatron-LM/commits/193463c4f8414e6906a40dd527a450bca50706b1/
CURRENT (PR #2058 from tk/vllm-v5): https://github.com/yaoyu-33/Megatron-LM/commits/23dd639cf3de30f3b9d8d0fae71ee31180be9ddd/

Please ensure all submodule commits are fast-forwards of the hemil/automodel-transformers-v5 branch before merging.

…sor policy v2 Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com> fix Signed-off-by: Hemil Desai <hemild@nvidia.com>

Signed-off-by: Terry Kong <terryk@nvidia.com>

github-actions · 2026-03-04T18:35:45Z

ℹ️ File Consistency Check

Check based on commit: 06fac10 (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T18:36:18Z

✅ Submodule Fast-Forward Check Results

Check based on commit: 06fac10 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

All submodule changes look good! ✨

Signed-off-by: Terry Kong <terryk@nvidia.com>

github-actions · 2026-03-04T18:39:34Z

ℹ️ File Consistency Check

Check based on commit: 567fb80 (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T18:40:10Z

✅ Submodule Fast-Forward Check Results

Check based on commit: 567fb80 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

All submodule changes look good! ✨

Signed-off-by: Terry Kong <terryk@nvidia.com>

github-actions · 2026-03-04T19:13:29Z

ℹ️ File Consistency Check

Check based on commit: 06f86b2 (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T19:14:03Z

✅ Submodule Fast-Forward Check Results

Check based on commit: 06f86b2 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

All submodule changes look good! ✨

Signed-off-by: Terry Kong <terryk@nvidia.com>

github-actions · 2026-03-04T20:06:56Z

ℹ️ File Consistency Check

Check based on commit: d437cdb (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T20:07:52Z

✅ Submodule Fast-Forward Check Results

Check based on commit: d437cdb (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

All submodule changes look good! ✨

Signed-off-by: Terry Kong <terryk@nvidia.com>

copy-pr-bot · 2026-03-04T22:51:29Z

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.

github-actions · 2026-03-04T22:51:53Z

ℹ️ File Consistency Check

Check based on commit: c19f6d8 (PR #2058 from tk/vllm-v5)

✅ DTensor Policy Worker Synchronization Check

Both DTensor policy worker files were modified in this PR:

nemo_rl/models/policy/workers/dtensor_policy_worker.py
nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py

Please ensure that the changes are consistent between both files where applicable.

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

github-actions · 2026-03-04T22:52:26Z

✅ Submodule Fast-Forward Check Results

Check based on commit: c19f6d8 (PR #2058 from tk/vllm-v5)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of hemil/automodel-transformers-v5 branch (fast-forward)

All submodule changes look good! ✨

terrykong requested review from a team as code owners March 3, 2026 23:39

hemildesai reviewed Mar 3, 2026

View reviewed changes

terrykong requested a review from a team as a code owner March 4, 2026 08:20

terrykong requested a review from a team as a code owner March 4, 2026 08:53

terrykong force-pushed the hemil/automodel-transformers-v5 branch from 23b7ccb to 0d2d3cd Compare March 4, 2026 17:54

terrykong requested review from a team as code owners March 4, 2026 17:54

terrykong force-pushed the hemil/automodel-transformers-v5 branch from 0d2d3cd to 3a2629c Compare March 4, 2026 18:17

terrykong force-pushed the tk/vllm-v5 branch from f40891a to 139e91b Compare March 4, 2026 18:18

github-actions bot added the CI Relating to CI label Mar 4, 2026

hemildesai and others added 7 commits March 4, 2026 10:28

vllm v5/sglang v5/mbridge v5

3b9c3a8

Signed-off-by: Terry Kong <terryk@nvidia.com>

fix vllm needing setuptools_scm at build

3936dcc

Signed-off-by: Terry Kong <terryk@nvidia.com>

let's try another way to setuptools_scm (put it in default deps)

81dfc42

Signed-off-by: Terry Kong <terryk@nvidia.com>

TE 2.12, MLM correct dependencies and remove flash attn vit VLLM patch

543703b

Signed-off-by: Terry Kong <terryk@nvidia.com>

remove the torch 2.9 attn patch

6246f4c

Signed-off-by: Terry Kong <terryk@nvidia.com>

fix vllm's dependencies

06fac10

Signed-off-by: Terry Kong <terryk@nvidia.com>

terrykong force-pushed the tk/vllm-v5 branch from 139e91b to 06fac10 Compare March 4, 2026 18:35

github-actions bot removed the CI Relating to CI label Mar 4, 2026

fix conflict

567fb80

Signed-off-by: Terry Kong <terryk@nvidia.com>

fix up sglang dependencies

06f86b2

Signed-off-by: Terry Kong <terryk@nvidia.com>

fixed up comments + add issue for cleanup + use nightly pinned commit

d437cdb

Signed-off-by: Terry Kong <terryk@nvidia.com>

fix flash infer + sglang disable piecewise cuda graphs since it crashes

c19f6d8

Signed-off-by: Terry Kong <terryk@nvidia.com>

Conversation

terrykong commented Mar 3, 2026

What does this PR do ?

Issues

Usage

Before your PR is "Ready for review"

Additional Information

Uh oh!

github-actions bot commented Mar 3, 2026

❌ Submodule Fast-Forward Check Failed

✅ Submodules that are properly updated:

❌ Submodules that need attention:

Uh oh!

hemildesai Mar 3, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Mar 4, 2026

❌ Submodule Fast-Forward Check Failed

✅ Submodules that are properly updated:

❌ Submodules that need attention:

Uh oh!

github-actions bot commented Mar 4, 2026

❌ Submodule Fast-Forward Check Failed

✅ Submodules that are properly updated:

❌ Submodules that need attention:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

❌ Submodule Fast-Forward Check Failed

✅ Submodules that are properly updated:

❌ Submodules that need attention:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

❌ Submodule Fast-Forward Check Failed

✅ Submodules that are properly updated:

❌ Submodules that need attention:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

✅ Submodule Fast-Forward Check Results

✅ Submodules that are properly updated:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

✅ Submodule Fast-Forward Check Results

✅ Submodules that are properly updated:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

✅ Submodule Fast-Forward Check Results

✅ Submodules that are properly updated:

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check

✅ DTensor Policy Worker Synchronization Check

Uh oh!

github-actions bot commented Mar 4, 2026

✅ Submodule Fast-Forward Check Results

✅ Submodules that are properly updated:

Uh oh!

copy-pr-bot bot commented Mar 4, 2026

Uh oh!

github-actions bot commented Mar 4, 2026

ℹ️ File Consistency Check