Skip to content
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.

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

Set weights_only=False in torch.load in EMA callback and AdapterMixin #12198

Merged
merged 12 commits into from
Feb 15, 2025

Conversation

chtruong814
Copy link
Collaborator

What does this PR do ?

Set weights_only=False in torch.load in EMA callback and AdapterMixin because the default for torch.load changed to weights_only=True.

Changelog

  • Set weights_only=False in torch.load in EMA callback and AdapterMixin

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

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 add or update any necessary documentation?
  • Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

  • New Feature
  • Bugfix
  • Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

  • Related to # (issue)

@github-actions github-actions bot added core Changes to NeMo Core common labels Feb 14, 2025
@chtruong814
Copy link
Collaborator Author

Tests passed for EMA here. Checking others.
https://github.com/NVIDIA/NeMo/actions/runs/13335164772/job/37252997159

Copy link
Collaborator

@maanug-nv maanug-nv left a comment

Choose a reason for hiding this comment

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

LGTM

Copy link
Collaborator

@titu1994 titu1994 left a comment

Choose a reason for hiding this comment

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

Thanks !

@chtruong814
Copy link
Collaborator Author

adapter mixin tests are passing now.
https://github.com/NVIDIA/NeMo/actions/runs/13338432490/job/37260251059

Looks like another location we need to pass in weights_only for rest of the expmanager tests. Will hunt for that to just get it all in for this PR.

@chtruong814
Copy link
Collaborator Author

All the tests passed here
https://github.com/NVIDIA/NeMo/actions/runs/13339479867/job/37262385957

Will go ahead and merge this now into the other "bump" branch we have. We'll open a PR for the accumulated fixes next week into main branch and cherry-pick to release branch.

@chtruong814 chtruong814 merged commit f28e055 into ko3n1g/build/bump-pyt-25.01 Feb 15, 2025
212 of 213 checks passed
@chtruong814 chtruong814 deleted the fix-ema-test branch February 15, 2025 01:26
nithinraok pushed a commit that referenced this pull request Feb 17, 2025
…#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
@nithinraok nithinraok mentioned this pull request Feb 17, 2025
8 tasks
nithinraok added a commit that referenced this pull request Feb 18, 2025
* fix_5089931

Signed-off-by: Nithin Rao Koluguri <nithinraok>

* Set weights_only=False in torch.load in EMA callback and AdapterMixin (#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>

* removed Confidense ensembles notebook

Signed-off-by: Nithin Rao Koluguri <nithinraok>

---------

Signed-off-by: Nithin Rao Koluguri <nithinraok>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: Nithin Rao Koluguri <nithinraok>
Co-authored-by: Charlie Truong <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
chtruong814 added a commit that referenced this pull request Feb 21, 2025
* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* try to fix checkpoint loading 

Signed-off-by: Chen Cui <[email protected]>

* update ci checkpoint

Signed-off-by: Chen Cui <[email protected]>

* fix ckpt loading for nemo1 and t5 tests

Signed-off-by: Chen Cui <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* bump

Signed-off-by: oliver könig <[email protected]>

* Rename neva datamodule (#12121)

* Rename dataset

Signed-off-by: yaoyu-33 <[email protected]>

* Apply isort and black reformatting

Signed-off-by: yaoyu-33 <[email protected]>

* Update

Signed-off-by: yaoyu-33 <[email protected]>

* pylink

Signed-off-by: yaoyu-33 <[email protected]>

* fix f string

Signed-off-by: yaoyu-33 <[email protected]>

* fix intern vit default factory

Signed-off-by: yaoyu-33 <[email protected]>

---------

Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>

* tests: Run FSDP2 on dual-gpu (#12145)

Signed-off-by: oliver könig <[email protected]>

* fix cmd

Signed-off-by: Chen Cui <[email protected]>

* Revert "build: Force re-install VCS dependencies (#12155)" (#12163)

This reverts commit 4b19ade.

Signed-off-by: Charlie Truong <[email protected]>

* fix tests

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* remove unsed variable

Signed-off-by: dimapihtar <[email protected]>

* Fix nsys callback tests (#12177)

* Fix nsys callback tests

Signed-off-by: Hemil Desai <[email protected]>

* Simplify tests

Signed-off-by: Hemil Desai <[email protected]>

* Fix

Signed-off-by: Hemil Desai <[email protected]>

---------

Signed-off-by: Hemil Desai <[email protected]>

* Fixes for bumping pyt to 25.01 (#12165)

* fix attention impl config

Signed-off-by: Maanu Grover <[email protected]>

* safeguard against empty key added by mcore

Signed-off-by: Maanu Grover <[email protected]>

* docu

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove unused import

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

---------

Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: akoumpa <[email protected]>

* Use pip --no-deps --force-reinstall when building the test container (#12175)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load in EMA callback and AdapterMixin (#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>

* Fix nemo-run stdin exception (#12197)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

* mock invoke context

Signed-off-by: Maanu Grover <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests"

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* tweak

Signed-off-by: Maanu Grover <[email protected]>

* more patches needed

Signed-off-by: Maanu Grover <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>

* Add legacy_ckpt arg to scripts/llm/gpt_distillation.py

Signed-off-by: Charlie Truong <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests" (#12202)

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* Ckpt fixes pytorch update (#12228)

* Fix checkpoint loading for ptq tests

Signed-off-by: Charlie Truong <[email protected]>

* Pass --ckpt_load_strictness log_all to L2_NeMo_2_SSM_Finetuning test

Signed-off-by: Charlie Truong <[email protected]>

* Fix tests in L2_NMT_Attention_is_All_You_Need_Finetuning by using default_factory

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* bump modelopt

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12 (#12265)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12

Signed-off-by: Piotr Żelasko <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Co-authored-by: oliver könig <[email protected]>

* remove retro test

Signed-off-by: oliver könig <[email protected]>

* L2_VLM_HF_Transformer_SFT_FSDP2

Signed-off-by: oliver könig <[email protected]>

* Fix distillation state-dict loading bug (#12270)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Add default param dtype in mistral configs (#12186)

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* ci: Fix pypi link of dry-run (#12267)

Signed-off-by: oliver könig <[email protected]>

* Fix tiny missing asterisk bug

Signed-off-by: Asha Anoosheh <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Co-authored-by: oliver könig <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>

* Add optimizer fix (#12253)

* Add optimizer fix

Signed-off-by: Boxiang Wang <[email protected]>

* Update hf_auto_model_for_causal_lm.py

Signed-off-by: BoxiangW <[email protected]>

---------

Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>

* num gpus

Signed-off-by: oliver könig <[email protected]>

* Build bitsandbytes (#12279)

Co-authored-by: Charlie Truong <[email protected]>

* Fixing error when loading T5 checkpoint created with TE<1.13 (#12264)

* fix error when loading checkpoint created with TE<1.13

* fix formatting

* fix formatting

* fix formatting

* Apply isort and black reformatting

Signed-off-by: huvunvidia <[email protected]>

---------

Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>

* no cancel for manual

Signed-off-by: oliver könig <[email protected]>

* Set L2_Speech_Batch_Size_OOMptimizer_Canary to be optional (#12299)

Signed-off-by: Charlie Truong <[email protected]>

* make expressions r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove parallelize_fn

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* rstring

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Chen Cui <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: Hemil Desai <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>
Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: oliver könig <[email protected]>
Co-authored-by: Chen Cui <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: Yu Yao <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: Hemil Desai <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
Co-authored-by: akoumpa <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
Co-authored-by: Piotr Żelasko <[email protected]>
Co-authored-by: Asha Anoosheh <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: BoxiangW <[email protected]>
Co-authored-by: Huy Vu <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>
ko3n1g added a commit that referenced this pull request Feb 22, 2025
* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* try to fix checkpoint loading 

Signed-off-by: Chen Cui <[email protected]>

* update ci checkpoint

Signed-off-by: Chen Cui <[email protected]>

* fix ckpt loading for nemo1 and t5 tests

Signed-off-by: Chen Cui <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* bump

Signed-off-by: oliver könig <[email protected]>

* Rename neva datamodule (#12121)

* Rename dataset

Signed-off-by: yaoyu-33 <[email protected]>

* Apply isort and black reformatting

Signed-off-by: yaoyu-33 <[email protected]>

* Update

Signed-off-by: yaoyu-33 <[email protected]>

* pylink

Signed-off-by: yaoyu-33 <[email protected]>

* fix f string

Signed-off-by: yaoyu-33 <[email protected]>

* fix intern vit default factory

Signed-off-by: yaoyu-33 <[email protected]>

---------

Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>

* tests: Run FSDP2 on dual-gpu (#12145)

Signed-off-by: oliver könig <[email protected]>

* fix cmd

Signed-off-by: Chen Cui <[email protected]>

* Revert "build: Force re-install VCS dependencies (#12155)" (#12163)

This reverts commit 4b19ade.

Signed-off-by: Charlie Truong <[email protected]>

* fix tests

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* remove unsed variable

Signed-off-by: dimapihtar <[email protected]>

* Fix nsys callback tests (#12177)

* Fix nsys callback tests

Signed-off-by: Hemil Desai <[email protected]>

* Simplify tests

Signed-off-by: Hemil Desai <[email protected]>

* Fix

Signed-off-by: Hemil Desai <[email protected]>

---------

Signed-off-by: Hemil Desai <[email protected]>

* Fixes for bumping pyt to 25.01 (#12165)

* fix attention impl config

Signed-off-by: Maanu Grover <[email protected]>

* safeguard against empty key added by mcore

Signed-off-by: Maanu Grover <[email protected]>

* docu

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove unused import

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

---------

Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: akoumpa <[email protected]>

* Use pip --no-deps --force-reinstall when building the test container (#12175)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load in EMA callback and AdapterMixin (#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>

* Fix nemo-run stdin exception (#12197)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

* mock invoke context

Signed-off-by: Maanu Grover <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests"

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* tweak

Signed-off-by: Maanu Grover <[email protected]>

* more patches needed

Signed-off-by: Maanu Grover <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>

* Add legacy_ckpt arg to scripts/llm/gpt_distillation.py

Signed-off-by: Charlie Truong <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests" (#12202)

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* Ckpt fixes pytorch update (#12228)

* Fix checkpoint loading for ptq tests

Signed-off-by: Charlie Truong <[email protected]>

* Pass --ckpt_load_strictness log_all to L2_NeMo_2_SSM_Finetuning test

Signed-off-by: Charlie Truong <[email protected]>

* Fix tests in L2_NMT_Attention_is_All_You_Need_Finetuning by using default_factory

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* bump modelopt

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12 (#12265)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12

Signed-off-by: Piotr Żelasko <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Co-authored-by: oliver könig <[email protected]>

* remove retro test

Signed-off-by: oliver könig <[email protected]>

* L2_VLM_HF_Transformer_SFT_FSDP2

Signed-off-by: oliver könig <[email protected]>

* Fix distillation state-dict loading bug (#12270)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Add default param dtype in mistral configs (#12186)

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* ci: Fix pypi link of dry-run (#12267)

Signed-off-by: oliver könig <[email protected]>

* Fix tiny missing asterisk bug

Signed-off-by: Asha Anoosheh <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Co-authored-by: oliver könig <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>

* Add optimizer fix (#12253)

* Add optimizer fix

Signed-off-by: Boxiang Wang <[email protected]>

* Update hf_auto_model_for_causal_lm.py

Signed-off-by: BoxiangW <[email protected]>

---------

Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>

* num gpus

Signed-off-by: oliver könig <[email protected]>

* Build bitsandbytes (#12279)

Co-authored-by: Charlie Truong <[email protected]>

* Fixing error when loading T5 checkpoint created with TE<1.13 (#12264)

* fix error when loading checkpoint created with TE<1.13

* fix formatting

* fix formatting

* fix formatting

* Apply isort and black reformatting

Signed-off-by: huvunvidia <[email protected]>

---------

Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>

* no cancel for manual

Signed-off-by: oliver könig <[email protected]>

* Set L2_Speech_Batch_Size_OOMptimizer_Canary to be optional (#12299)

Signed-off-by: Charlie Truong <[email protected]>

* make expressions r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove parallelize_fn

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* rstring

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* reduce grad-acc-steps

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update tests

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* force-switch from MegatronCheckpointIO to HFCheckpointIO

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix setup_environment call

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update test

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update verify_sft_checkpoint_structure to handle sharded checkpoints

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* make io_bytes.pt optional

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* set broken tests to optional

Signed-off-by: oliver könig <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Chen Cui <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: Hemil Desai <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>
Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Chen Cui <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: Yu Yao <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: Hemil Desai <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
Co-authored-by: akoumpa <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
Co-authored-by: Piotr Żelasko <[email protected]>
Co-authored-by: Asha Anoosheh <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: BoxiangW <[email protected]>
Co-authored-by: Huy Vu <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>
ko3n1g added a commit that referenced this pull request Feb 22, 2025
* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* try to fix checkpoint loading

Signed-off-by: Chen Cui <[email protected]>

* update ci checkpoint

Signed-off-by: Chen Cui <[email protected]>

* fix ckpt loading for nemo1 and t5 tests

Signed-off-by: Chen Cui <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* bump

Signed-off-by: oliver könig <[email protected]>

* Rename neva datamodule (#12121)

* Rename dataset

Signed-off-by: yaoyu-33 <[email protected]>

* Apply isort and black reformatting

Signed-off-by: yaoyu-33 <[email protected]>

* Update

Signed-off-by: yaoyu-33 <[email protected]>

* pylink

Signed-off-by: yaoyu-33 <[email protected]>

* fix f string

Signed-off-by: yaoyu-33 <[email protected]>

* fix intern vit default factory

Signed-off-by: yaoyu-33 <[email protected]>

---------

Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>

* tests: Run FSDP2 on dual-gpu (#12145)

Signed-off-by: oliver könig <[email protected]>

* fix cmd

Signed-off-by: Chen Cui <[email protected]>

* Revert "build: Force re-install VCS dependencies (#12155)" (#12163)

This reverts commit 4b19ade.

Signed-off-by: Charlie Truong <[email protected]>

* fix tests

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* remove unsed variable

Signed-off-by: dimapihtar <[email protected]>

* Fix nsys callback tests (#12177)

* Fix nsys callback tests

Signed-off-by: Hemil Desai <[email protected]>

* Simplify tests

Signed-off-by: Hemil Desai <[email protected]>

* Fix

Signed-off-by: Hemil Desai <[email protected]>

---------

Signed-off-by: Hemil Desai <[email protected]>

* Fixes for bumping pyt to 25.01 (#12165)

* fix attention impl config

Signed-off-by: Maanu Grover <[email protected]>

* safeguard against empty key added by mcore

Signed-off-by: Maanu Grover <[email protected]>

* docu

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove unused import

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

---------

Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: akoumpa <[email protected]>

* Use pip --no-deps --force-reinstall when building the test container (#12175)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load in EMA callback and AdapterMixin (#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>

* Fix nemo-run stdin exception (#12197)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

* mock invoke context

Signed-off-by: Maanu Grover <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests"

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* tweak

Signed-off-by: Maanu Grover <[email protected]>

* more patches needed

Signed-off-by: Maanu Grover <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>

* Add legacy_ckpt arg to scripts/llm/gpt_distillation.py

Signed-off-by: Charlie Truong <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests" (#12202)

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* Ckpt fixes pytorch update (#12228)

* Fix checkpoint loading for ptq tests

Signed-off-by: Charlie Truong <[email protected]>

* Pass --ckpt_load_strictness log_all to L2_NeMo_2_SSM_Finetuning test

Signed-off-by: Charlie Truong <[email protected]>

* Fix tests in L2_NMT_Attention_is_All_You_Need_Finetuning by using default_factory

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* bump modelopt

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12 (#12265)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12

Signed-off-by: Piotr Żelasko <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Co-authored-by: oliver könig <[email protected]>

* remove retro test

Signed-off-by: oliver könig <[email protected]>

* L2_VLM_HF_Transformer_SFT_FSDP2

Signed-off-by: oliver könig <[email protected]>

* Fix distillation state-dict loading bug (#12270)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Add default param dtype in mistral configs (#12186)

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* ci: Fix pypi link of dry-run (#12267)

Signed-off-by: oliver könig <[email protected]>

* Fix tiny missing asterisk bug

Signed-off-by: Asha Anoosheh <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Co-authored-by: oliver könig <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>

* Add optimizer fix (#12253)

* Add optimizer fix

Signed-off-by: Boxiang Wang <[email protected]>

* Update hf_auto_model_for_causal_lm.py

Signed-off-by: BoxiangW <[email protected]>

---------

Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>

* num gpus

Signed-off-by: oliver könig <[email protected]>

* Build bitsandbytes (#12279)

Co-authored-by: Charlie Truong <[email protected]>

* Fixing error when loading T5 checkpoint created with TE<1.13 (#12264)

* fix error when loading checkpoint created with TE<1.13

* fix formatting

* fix formatting

* fix formatting

* Apply isort and black reformatting

Signed-off-by: huvunvidia <[email protected]>

---------

Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>

* no cancel for manual

Signed-off-by: oliver könig <[email protected]>

* Set L2_Speech_Batch_Size_OOMptimizer_Canary to be optional (#12299)

Signed-off-by: Charlie Truong <[email protected]>

* make expressions r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove parallelize_fn

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* rstring

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* reduce grad-acc-steps

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update tests

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* force-switch from MegatronCheckpointIO to HFCheckpointIO

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix setup_environment call

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update test

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update verify_sft_checkpoint_structure to handle sharded checkpoints

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* make io_bytes.pt optional

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* set broken tests to optional

Signed-off-by: oliver könig <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Chen Cui <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: Hemil Desai <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>
Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Chen Cui <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: Yu Yao <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: Hemil Desai <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
Co-authored-by: akoumpa <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
Co-authored-by: Piotr Żelasko <[email protected]>
Co-authored-by: Asha Anoosheh <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: BoxiangW <[email protected]>
Co-authored-by: Huy Vu <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>
Signed-off-by: oliver könig <[email protected]>
ko3n1g added a commit that referenced this pull request Feb 22, 2025
* build: Bump PyT to 25.01 (#11973)

* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* try to fix checkpoint loading

Signed-off-by: Chen Cui <[email protected]>

* update ci checkpoint

Signed-off-by: Chen Cui <[email protected]>

* fix ckpt loading for nemo1 and t5 tests

Signed-off-by: Chen Cui <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* bump

Signed-off-by: oliver könig <[email protected]>

* Rename neva datamodule (#12121)

* Rename dataset

Signed-off-by: yaoyu-33 <[email protected]>

* Apply isort and black reformatting

Signed-off-by: yaoyu-33 <[email protected]>

* Update

Signed-off-by: yaoyu-33 <[email protected]>

* pylink

Signed-off-by: yaoyu-33 <[email protected]>

* fix f string

Signed-off-by: yaoyu-33 <[email protected]>

* fix intern vit default factory

Signed-off-by: yaoyu-33 <[email protected]>

---------

Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>

* tests: Run FSDP2 on dual-gpu (#12145)

Signed-off-by: oliver könig <[email protected]>

* fix cmd

Signed-off-by: Chen Cui <[email protected]>

* Revert "build: Force re-install VCS dependencies (#12155)" (#12163)

This reverts commit 4b19ade.

Signed-off-by: Charlie Truong <[email protected]>

* fix tests

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* Apply isort and black reformatting

Signed-off-by: dimapihtar <[email protected]>

* fix style

Signed-off-by: dimapihtar <[email protected]>

* remove unsed variable

Signed-off-by: dimapihtar <[email protected]>

* Fix nsys callback tests (#12177)

* Fix nsys callback tests

Signed-off-by: Hemil Desai <[email protected]>

* Simplify tests

Signed-off-by: Hemil Desai <[email protected]>

* Fix

Signed-off-by: Hemil Desai <[email protected]>

---------

Signed-off-by: Hemil Desai <[email protected]>

* Fixes for bumping pyt to 25.01 (#12165)

* fix attention impl config

Signed-off-by: Maanu Grover <[email protected]>

* safeguard against empty key added by mcore

Signed-off-by: Maanu Grover <[email protected]>

* docu

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* noqa?

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove unused import

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

---------

Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: akoumpa <[email protected]>

* Use pip --no-deps --force-reinstall when building the test container (#12175)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load in EMA callback and AdapterMixin (#12198)

* Set weights_only=False in torch.load within EMA callbck

Signed-off-by: Charlie Truong <[email protected]>

* Set weights_only=False in torch.load call in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix lint errors in ema and adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Fix undefined name error in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Remove unnecessary import of __futures__ in adapter_mixins

Signed-off-by: Charlie Truong <[email protected]>

* Fix more locations where weights_only=False needs to be passed to torch.load

Signed-off-by: Charlie Truong <[email protected]>

* Fix lint issues in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

* Apply isort and black reformatting

Signed-off-by: chtruong814 <[email protected]>

* Correct noqa F821 line in nemo_model_checkpoint

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Co-authored-by: chtruong814 <[email protected]>

* Fix nemo-run stdin exception (#12197)

* Use pip --no-deps --force-reinstall when building the test container

Signed-off-by: Charlie Truong <[email protected]>

* Explicitly reinstall nvidia-resiliency

Signed-off-by: Charlie Truong <[email protected]>

* Do not have pytest capture output in lightning unit tests

Signed-off-by: Charlie Truong <[email protected]>

* mock invoke context

Signed-off-by: Maanu Grover <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests"

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* tweak

Signed-off-by: Maanu Grover <[email protected]>

* more patches needed

Signed-off-by: Maanu Grover <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>

* Add legacy_ckpt arg to scripts/llm/gpt_distillation.py

Signed-off-by: Charlie Truong <[email protected]>

* Revert "Do not have pytest capture output in lightning unit tests" (#12202)

This reverts commit 162c730.

Signed-off-by: Maanu Grover <[email protected]>

* Ckpt fixes pytorch update (#12228)

* Fix checkpoint loading for ptq tests

Signed-off-by: Charlie Truong <[email protected]>

* Pass --ckpt_load_strictness log_all to L2_NeMo_2_SSM_Finetuning test

Signed-off-by: Charlie Truong <[email protected]>

* Fix tests in L2_NMT_Attention_is_All_You_Need_Finetuning by using default_factory

Signed-off-by: Charlie Truong <[email protected]>

---------

Signed-off-by: Charlie Truong <[email protected]>

* bump modelopt

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12 (#12265)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Fix 2D bucketing test on Python 3.12

Signed-off-by: Piotr Żelasko <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Co-authored-by: oliver könig <[email protected]>

* remove retro test

Signed-off-by: oliver könig <[email protected]>

* L2_VLM_HF_Transformer_SFT_FSDP2

Signed-off-by: oliver könig <[email protected]>

* Fix distillation state-dict loading bug (#12270)

* ci: Bump release workflows (#12259)

Signed-off-by: oliver könig <[email protected]>

* Add default param dtype in mistral configs (#12186)

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* ci: Fix pypi link of dry-run (#12267)

Signed-off-by: oliver könig <[email protected]>

* Fix tiny missing asterisk bug

Signed-off-by: Asha Anoosheh <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Co-authored-by: oliver könig <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>

* Add optimizer fix (#12253)

* Add optimizer fix

Signed-off-by: Boxiang Wang <[email protected]>

* Update hf_auto_model_for_causal_lm.py

Signed-off-by: BoxiangW <[email protected]>

---------

Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>

* num gpus

Signed-off-by: oliver könig <[email protected]>

* Build bitsandbytes (#12279)

Co-authored-by: Charlie Truong <[email protected]>

* Fixing error when loading T5 checkpoint created with TE<1.13 (#12264)

* fix error when loading checkpoint created with TE<1.13

* fix formatting

* fix formatting

* fix formatting

* Apply isort and black reformatting

Signed-off-by: huvunvidia <[email protected]>

---------

Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>

* no cancel for manual

Signed-off-by: oliver könig <[email protected]>

* Set L2_Speech_Batch_Size_OOMptimizer_Canary to be optional (#12299)

Signed-off-by: Charlie Truong <[email protected]>

* make expressions r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* remove parallelize_fn

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* rstring

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* r-strings

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* reduce grad-acc-steps

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* r-string

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update tests

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* force-switch from MegatronCheckpointIO to HFCheckpointIO

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* fix setup_environment call

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update test

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* update verify_sft_checkpoint_structure to handle sharded checkpoints

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* make io_bytes.pt optional

Signed-off-by: Alexandros Koumparoulis <[email protected]>

* Apply isort and black reformatting

Signed-off-by: akoumpa <[email protected]>

* set broken tests to optional

Signed-off-by: oliver könig <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Chen Cui <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: Hemil Desai <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>
Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Chen Cui <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: Yu Yao <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: Hemil Desai <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
Co-authored-by: akoumpa <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
Co-authored-by: Piotr Żelasko <[email protected]>
Co-authored-by: Asha Anoosheh <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: BoxiangW <[email protected]>
Co-authored-by: Huy Vu <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>
Signed-off-by: oliver könig <[email protected]>

* Update reinstall.sh

Signed-off-by: oliver könig <[email protected]>

* fix

Signed-off-by: oliver könig <[email protected]>

* remove retro

Signed-off-by: oliver könig <[email protected]>

* weights_only=False

Signed-off-by: oliver könig <[email protected]>

* weights_only=False

Signed-off-by: oliver könig <[email protected]>

* bump mcore

Signed-off-by: oliver könig <[email protected]>

* weights_only=False

Signed-off-by: oliver könig <[email protected]>

* fixes

Signed-off-by: oliver könig <[email protected]>

* ++model.dist_ckpt_load_strictness=log_all

Signed-off-by: oliver könig <[email protected]>

* f

Signed-off-by: oliver könig <[email protected]>

* Distillation NeMo run entrypoint and recipe (#12143)

* Remove duplicate test script

Signed-off-by: Asha Anoosheh <[email protected]>

* Minor renamings

Signed-off-by: Asha Anoosheh <[email protected]>

* Create entrypoint + recipe

Signed-off-by: Asha Anoosheh <[email protected]>

* Appease the linter

Signed-off-by: Asha Anoosheh <[email protected]>

* Fix circular import

Signed-off-by: Asha Anoosheh <[email protected]>

* Update interface

Signed-off-by: Asha Anoosheh <[email protected]>

* Appease linter

Signed-off-by: Asha Anoosheh <[email protected]>

* Appease even more linter

Signed-off-by: Asha Anoosheh <[email protected]>

* Make TP8 default recipe

Signed-off-by: Asha Anoosheh <[email protected]>

* Fix tokenizer override

Signed-off-by: Asha Anoosheh <[email protected]>

---------

Signed-off-by: Asha Anoosheh <[email protected]>

* set starcoder to optional

Signed-off-by: oliver könig <[email protected]>

---------

Signed-off-by: oliver könig <[email protected]>
Signed-off-by: Chen Cui <[email protected]>
Signed-off-by: Alexandros Koumparoulis <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Charlie Truong <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: dimapihtar <[email protected]>
Signed-off-by: Hemil Desai <[email protected]>
Signed-off-by: Maanu Grover <[email protected]>
Signed-off-by: akoumpa <[email protected]>
Signed-off-by: chtruong814 <[email protected]>
Signed-off-by: Piotr Żelasko <[email protected]>
Signed-off-by: Asha Anoosheh <[email protected]>
Signed-off-by: Boxiang Wang <[email protected]>
Signed-off-by: BoxiangW <[email protected]>
Signed-off-by: huvunvidia <[email protected]>
Co-authored-by: Chen Cui <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: Yu Yao <[email protected]>
Co-authored-by: yaoyu-33 <[email protected]>
Co-authored-by: Charlie Truong <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: dimapihtar <[email protected]>
Co-authored-by: Hemil Desai <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
Co-authored-by: akoumpa <[email protected]>
Co-authored-by: chtruong814 <[email protected]>
Co-authored-by: Piotr Żelasko <[email protected]>
Co-authored-by: Asha Anoosheh <[email protected]>
Co-authored-by: Alexandros Koumparoulis <[email protected]>
Co-authored-by: BoxiangW <[email protected]>
Co-authored-by: Huy Vu <[email protected]>
Co-authored-by: Huy Vu2 <[email protected]>
Co-authored-by: huvunvidia <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants