-
Notifications
You must be signed in to change notification settings - Fork 485
Pull requests: NVIDIA-NeMo/Megatron-Bridge
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[data] fix: Resume MegatronMIMO validation samples
area:data
Dataset builders, preprocessing, and samplers
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5987
opened Sep 6, 2026 by
yaoyu-33
Contributor
Loading…
[training] fix: Use global DP for DistTrain sample accounting
area:training
Training loop, callbacks, and runtime integration
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5986
opened Sep 6, 2026 by
yaoyu-33
Contributor
Loading…
[inference] fix: Forward derived DeepSeek-V4 pipeline layouts
area:misc
Cross-cutting utilities, logging, helpers, and other changes
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5985
opened Sep 6, 2026 by
yaoyu-33
Contributor
Loading…
[data] fix: Release inherited GPU descriptors in workers
area:data
Dataset builders, preprocessing, and samplers
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5984
opened Sep 6, 2026 by
yaoyu-33
Contributor
Loading…
[training] fix: Enable NVRx detection for MegatronMIMO
area:training
Training loop, callbacks, and runtime integration
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5983
opened Sep 6, 2026 by
yaoyu-33
Contributor
Loading…
Perf(recipe): enable CuTe DSL fused grouped GEMM for Qwen3.5-VL GB300
area:perf
Performance optimizations and benchmarking
feature
New capabilities, enhancements, or enablement work
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5969
opened Sep 4, 2026 by
tomlifu
Contributor
Loading…
5 tasks
perf(qwen): enable Qwen3 235B NVFP4 full-iteration optimizations
area:perf
Performance optimizations and benchmarking
feature
New capabilities, enhancements, or enablement work
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5965
opened Sep 4, 2026 by
rhmukundan
Contributor
Loading…
[data] fix: Preserve RNG state during synchronous loading
area:data
Dataset builders, preprocessing, and samplers
bug
Something isn't working
needs-review
PR is ready for code review and waiting on a reviewer
#5960
opened Sep 4, 2026 by
yaoyu-33
Contributor
Loading…
[ckpt] fix: Preserve ModelOpt state in low-memory saves
area:ckpt
Checkpoint conversion, loading, export, and save paths
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5956
opened Sep 4, 2026 by
yaoyu-33
Contributor
Loading…
fix(model): honor explicit meta-device disable
area:model
Model implementations and HF bridge logic
bug
Something isn't working
community-request
needs-review
PR is ready for code review and waiting on a reviewer
#5952
opened Sep 3, 2026 by
ZenAlexa
Contributor
Loading…
fix(model): fix(model): export correct MTP layer count for Nemotron-H
area:model
Model implementations and HF bridge logic
bug
Something isn't working
community-request
ready-to-merge
PR is approved, current, and only waiting for CI to pass before merge
#5948
opened Sep 3, 2026 by
odedovadia
Loading…
5 tasks
fix(ckpt): initialize model parallel for builder loads
area:ckpt
Checkpoint conversion, loading, export, and save paths
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5942
opened Sep 3, 2026 by
yaoyu-33
Contributor
Loading…
Support 3D-MoE LoRA export for shared-outer routed-expert adapters
area:peft
Parameter-efficient fine-tuning (LoRA, adapters)
community-request
feature
New capabilities, enhancements, or enablement work
needs-more-tests
Requires additional L0 and L1 test coverage before merge
ready-to-merge
PR is approved, current, and only waiting for CI to pass before merge
#5939
opened Sep 2, 2026 by
HollowMan6
Contributor
Loading…
2 of 5 tasks
[data] fix: Allow small datasets with cyclic loaders
area:data
Dataset builders, preprocessing, and samplers
bug
Something isn't working
needs-review
PR is ready for code review and waiting on a reviewer
#5937
opened Sep 2, 2026 by
yaoyu-33
Contributor
Loading…
cp: restrict HybridEP padding to THD on r0.6.0
area:training
Training loop, callbacks, and runtime integration
bug
Something isn't working
cherry-pick
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
r0.6.0
Auto-cherrypick to release branch. Apply before merge; cherrypick happens after merge.
#5935
opened Sep 1, 2026 by
svcnemo-autobot
Contributor
Loading…
fix(nemotron_omni): read and write exported modeling.py as UTF-8
area:ckpt
Checkpoint conversion, loading, export, and save paths
bug
Something isn't working
community-request
needs-review
PR is ready for code review and waiting on a reviewer
#5934
opened Sep 1, 2026 by
MohammedAlkindi
Loading…
fix(fsdp): drop stale MFSDP v2 guard rails for clip_grad, precision-aware optimizer, HSDP, and optimizer CUDA graphs
area:training
Training loop, callbacks, and runtime integration
bug
Something isn't working
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5933
opened Sep 1, 2026 by
wujingyue
Contributor
Loading…
1 of 2 tasks
feat(conversion): build native MXFP8 export tasks
area:ckpt
Checkpoint conversion, loading, export, and save paths
feature
New capabilities, enhancements, or enablement work
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5917
opened Sep 1, 2026 by
seonjinn
Loading…
[peft] fix: scale grouped-expert LoRA at the bottleneck, not the output
area:peft
Parameter-efficient fine-tuning (LoRA, adapters)
bug
Something isn't working
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5913
opened Aug 31, 2026 by
cuichenx
Contributor
Loading…
chore(docker): update base image to nvcr.io/nvidia/pytorch:26.08-py3
area:build
Dependencies, packaging, images, and environment setup
ci
CI, automation, test queue, or workflow infrastructure work
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5911
opened Aug 31, 2026 by
balasaajay
Contributor
Loading…
5 tasks
chore: AUT-2155 bump Megatron Core nightly pins
area:build
Dependencies, packaging, images, and environment setup
ci
CI, automation, test queue, or workflow infrastructure work
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
needs-review
PR is ready for code review and waiting on a reviewer
#5897
opened Aug 30, 2026 by
svcnemo-autobot
Contributor
Loading…
test: AUT-2155 validate Megatron Core dev compatibility
area:ci
ci
CI, automation, test queue, or workflow infrastructure work
dummy-pr
Filed this PR to run tests, not going to merge
full-test-suite
needs-more-tests
Requires additional L0 and L1 test coverage before merge
#5896
opened Aug 30, 2026 by
svcnemo-autobot
Contributor
Loading…
Previous Next
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.