Skip to content

Cherry pick 1216 r0.2.0 - #1318

Closed
adityavavreNVDA wants to merge 63 commits into
mainfrom
cherry-pick-1216-r0.2.0
Closed

Cherry pick 1216 r0.2.0#1318
adityavavreNVDA wants to merge 63 commits into
mainfrom
cherry-pick-1216-r0.2.0

Conversation

@adityavavreNVDA

@adityavavreNVDA adityavavreNVDA commented Nov 12, 2025

Copy link
Copy Markdown
Contributor

What does this PR do ?

Cherry pick #1216 into r0.2.0

Changelog

  • Add specific line by line info of high level changes in this PR.

GitHub Actions CI

See the CI sectionin the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

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?

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

Additional Information

  • Related to # (issue)

chtruong814 and others added 30 commits October 21, 2025 23:47
@chtruong814 @ko3n1g
@ananthsub @maanug-nv
chore(beep boop 🤖): Bump uv.lock (main) (2025-10-16) (#969) (#1024)
27b7c46 
* chore(beep boop 🤖): Bump `uv.lock` (main) (2025-10-16)



* expose eos property on tokenizers



* dev branch fix for linear attention type spec arg (#987)

* chore(beep boop 🤖): Bump `uv.lock` (main) (2025-10-15)



* explicitly set kwargs in mlm compat gpt specs



* support existing checkpoint without linear attn arg



---------





* fix tests for mlm compat to use kwargs



---------

Signed-off-by: Oliver Koenig <okoenig@nvidia.com>
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: Maanu Grover <maanug@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: ko3n1g <16716991+ko3n1g@users.noreply.github.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
Co-authored-by: Maanu Grover <109391026+maanug-nv@users.noreply.github.com>
@chtruong814 @ko3n1g
chore(beep boop 🤖): Bump uv.lock (main) (2025-10-21) (#1018) (#1032)
44d021b 
Signed-off-by: Oliver Koenig <okoenig@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: ko3n1g <16716991+ko3n1g@users.noreply.github.com>
@chtruong814 @maanug-nv
Enable overriding TokenizerConfig when instantiating from checkpoint (#…
469bbd2 
…1030) (#1034)

* add override kwargs



* add test



---------

Signed-off-by: Maanu Grover <maanug@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Maanu Grover <109391026+maanug-nv@users.noreply.github.com>
@chtruong814 @ananthsub
Qwen3 finetune recipes (#975) (#1042)
317052a 
* qwen3 finetuning configs



* qwen3 finetuning configs



* qwen3 finetune



* qwen3 finetune recipes



* updates



* revert changes to gpt step



* address feedback



* add load dir



---------

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
remove lru_cache decorator from finetune dataset building (#1012) (#1021
be1e5bb 
)

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
update mamba recipe imports (#1038) (#1049)
ca491a4 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
Make nemotronh recipe imports consistent with codebase (#1006) (#1019)
5b66497 
* make nemotronh recipe imports consistent



* rm extra



---------

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
support loading checkpoints from a given step (#1040) (#1059)
580fca5 
* support loading checkpoints from a given step



* support loading checkpoints from a given step



* add config test



* fix tests



---------

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @cuichenx
@yaoyu-33 @ananthsub
Add gpt-oss recipes (#898) (#1055)
258dc0d 
* enable ep and etp in bridge code



* fix embeddings_are_tied issues



* fix tp + pp, expert number issue



* lint



* add gpt-oss provider



* add gpt oss bridge placeholder



* unit test fix



* naming fix



* lint



* Update src/megatron/bridge/models/conversion/param_mapping.py




* address comments



* fix for hf to megatron to run



* finish hf import

* push current export code



* fix export



* fix export



* address comment



* fix row parallel bias



* fix expert weight export on multi gpus



* fix assertion



* fix config and clean up



* fix for ep and tp



* remove strict flag



* add pipeline dtype to ModelParallelKwargs



* fix ep=1



* pipeline dtype



* revert



* remove quantization option (import can handle both, export is always de-quantized)



* fix etp



* clean up



* add tests



* fix conversion test



* linting



* address comments



* fix tensor shape for ep=1 for other moe models



* add recipes



* clean up

* lint

* add recipe test

* clean up



* address comment



* update test



---------

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Signed-off-by: Chen Cui <cxcui@alumni.cmu.edu>
Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <cxcui@alumni.cmu.edu>
Co-authored-by: yaoyu-33 <yaoyu.094@gmail.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@ko3n1g
ci: Track M-LM's release branch (#1066)
171acd5 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@chtruong814 @ko3n1g
cp: build: Dependency bump (1037) into r0.2.0 (#1094)
ccf9924 
Signed-off-by: oliver könig <okoenig@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: oliver könig <okoenig@nvidia.com>
@chtruong814 @ko3n1g
cp: docs: Update contrib guide (1022) into r0.2.0 (#1102)
edda462 
Signed-off-by: oliver könig <okoenig@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: oliver könig <okoenig@nvidia.com>
@chtruong814 @ananthsub
cp: [docs] Add supported models page (980) into r0.2.0 (#1104)
b0a635e 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: link perf summary from readme (1041) into r0.2.0 (#1108)
875bc4b 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @maanug-nv
cp: Use TE quantization modes for first/last bf16 layers (715) into…
d589b41 
… `r0.2.0` (#1109)

Signed-off-by: Maanu Grover <maanug@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Maanu Grover <109391026+maanug-nv@users.noreply.github.com>
@chtruong814 @ananthsub
cp: Clean up fusion checks (938) into r0.2.0 (#1117)
d0eb9fc 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: reduce peak memory for qwen3 moe recipes (1054) into r0.2.0 (#…
b686c9f 
…1112)

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: Add settings/validation for deterministic mode (968) into `r0.2…
0015c0f 
….0` (#1107)

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@ko3n1g
build: Bump MCore (#1097)
dcc81b1 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@chtruong814 @ananthsub
cp: Fix load-hf-weights order (1096) into r0.2.0 (#1103)
1df31c8 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: `Remove sequence length validation check for custom dataset provi…
a68e0a0 
…ders (1004)` into `r0.2.0` (#1118)

Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: fix throughput logging when resuming (1110) into r0.2.0 (#1124)
c31e1b8 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @ananthsub
cp: fix Nemotron h provider and bridge (1053) into r0.2.0 (#1125)
007917a 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @cuichenx
cp: Documentation for Nemotron Nano V2 VL (1121) into r0.2.0 (#1122)
1ab74b5 
Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@chtruong814 @ananthsub
cp: add __init__.py for gemma3 vl recipes (1139) into r0.2.0 (#1142)
32a94eb 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ananth Subramaniam <ansubramania@nvidia.com>
@chtruong814 @yaoyu-33 @yuzhongw-nvidia
cp: [Model Support] Qwen3-Next (1068) into r0.2.0 (#1126)
0628f3b 
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: Yuzhong Wang <yuzhongw@nvidia.com>
@chtruong814 @rhmukundan
cp: Bug fix for Warmup Iters (1135) into r0.2.0 (#1137)
00073e5 
Signed-off-by: Raghav Hrishikeshan Mukundan <rmukundan@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Raghav Hrishikeshan Mukundan <102543536+rhmukundan@users.noreply.github.com>
@chtruong814 @suiyoubi
cp: Add moonlight 16B recipe (1133) into r0.2.0 (#1158)
a96b61c 
Signed-off-by: Ao Tang <aot@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ao Tang <aot@nvidia.com>
@adityavavreNVDA
Cherry Pick Qwen3-Next Model Provider + Pretraining Support (#1027) (#…
f872e23 
…1173)

Signed-off-by: adityavavreNVDA <avavre@nvidia.com>
@chtruong814 @AAnoosheh
cp: ModelOpt Knowledge-Distillation support (880) into r0.2.0 (#1171
a3d88b7 
)

Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Asha Anoosheh <aanoosheh@nvidia.com>
chtruong814 and others added 4 commits November 11, 2025 13:15
@chtruong814 @malay-nagda
cp: CUDA_DEVICE_MAX_CONNECTIONS update for Blackwell (1287) into `r…
f6bd2e7 
…0.2.0` (#1293)

Signed-off-by: Malay Nagda <malayn@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: malay-nagda <malayn@nvidia.com>
@chtruong814 @github-actions
chore(beep boop 🤖): Bump uv.lock (r0.2.0) (2025-11-11) (#1282)
0299f54 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
@chtruong814 @ko3n1g @malay-nagda
cp: chore: Merge fp8 args (1279) into r0.2.0 (#1306)
e9f7f8c 
Signed-off-by: Malay Nagda <malayn@nvidia.com>
Signed-off-by: oliver könig <okoenig@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: oliver könig <okoenig@nvidia.com>
Co-authored-by: Malay Nagda <malayn@nvidia.com>
@chtruong814 @adityavavreNVDA Aditya Vavre
cp: Low Precision Recipes for LLama3-8B (1178) into r0.2.0 (#1300)
dccbd83 
Signed-off-by: Aditya Vavre <avavre@nvidia.com>
Signed-off-by: adityavavreNVDA <avavre@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: adityavavreNVDA <avavre@nvidia.com>
Co-authored-by: Aditya Vavre <avavre@gcp-nrt-cs-001-login-001.cm.cluster>
@adityavavreNVDA
adityavavreNVDA requested a review from a team as a code owner November 12, 2025 18:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@adityavavreNVDA @chtruong814 @ko3n1g @ananthsub @malay-nagda