Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
368 Open 2,361 Closed
368 Open 2,361 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Make param_index_map always use unpacked (full numel) offsets
#4328 opened Apr 15, 2026 by deepakn94 Contributor • Draft Core 0.16
3
Fix Megatron initialization with extra_args_provider complexity: low
#4327 opened Apr 15, 2026 by santhnm2 Contributor Loading…
5 tasks
Core 0.16
4
Load MTP and latent MoE layers in bf16 for mxfp8 inference
#4326 opened Apr 15, 2026 by santhnm2 Contributor • Draft
5 tasks
1
Fix activation_func check and MLP sharded_state_dict
#4325 opened Apr 15, 2026 by gdengk Contributor • Draft
5 tasks
1
Fix checkpoint loading with load_main_params_from_ckpt=True for grouped weight
#4324 opened Apr 15, 2026 by ksivaman Member Loading…
5 tasks
1
Fix inference graph override in RL flow complexity: low
#4323 opened Apr 15, 2026 by tdene Contributor Loading…
5 tasks
Core 0.16
1
remove legacy GPT code complexity: high Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review.
#4322 opened Apr 15, 2026 by dimapihtar Contributor Loading…
5 tasks
Core 0.16
2
Inference: enable flex dispatcher (DeepEP/HybridEP) for prefill complexity: low
#4321 opened Apr 15, 2026 by mathemakitten Contributor Loading…
5 tasks
Core 0.16
2
Overlap engine bookkeeping with step complexity: high
#4320 opened Apr 15, 2026 by tdene Contributor Loading…
5 tasks
Core 0.16
1
SafeUnpickler class for safe pickle usage complexity: low Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. Final Review PR is in the "final review" stage Run functional tests
#4319 opened Apr 15, 2026 by dimapihtar Contributor Loading…
5 tasks
Core 0.16
7
Add TEFusedDenseMLP for Dense+Grouped GEMM fusion on SM100+ complexity: medium
#4318 opened Apr 15, 2026 by sraman-rgb Loading…
5 tasks
1
Eliminate QPs during checkpoint loading, no need for review or merge
#4317 opened Apr 15, 2026 by CarlosGomes98 Contributor • Draft
2
fix: bisect --second-run flag and preserve exit code in generated test scripts
#4315 opened Apr 15, 2026 by ko3n1g Contributor • Draft
3 tasks
Core 0.16
2
fix: enforce correct pass thresholds for deterministic and approximate tests (core_r0.17.0) Run functional tests
#4314 opened Apr 15, 2026 by ko3n1g Contributor • Draft
1 task
Core 0.16
3
fix: correct 'Seperate'/'Seperated' typo in comments community-request
#4313 opened Apr 15, 2026 by MukundaKatta • Draft
2
feat: Optimize memory footprint of long-context training via fused kernel and chunking community-request
#4312 opened Apr 15, 2026 by terminator123 • Draft
2
Fix fused grouped MLP wgrad hooks for DDP reduce-scatter complexity: low
#4311 opened Apr 15, 2026 by gdengk Contributor Loading…
5 tasks
2
[training migration] Add serialization features to config container
#4309 opened Apr 15, 2026 by maanug-nv Contributor • Draft
5 tasks
Core 0.16
4
Add quantization type debug logging.
#4308 opened Apr 14, 2026 by kwyss-nvidia Contributor • Draft
5 tasks
2
Move inference context bookkeeping to CPU with ContextGPUView
#4306 opened Apr 14, 2026 by lmcafee-nvidia Contributor • Draft
8 tasks
@lmcafee-nvidia
4
checkpoint integrity verification complexity: medium Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. Run functional tests
#4305 opened Apr 14, 2026 by dimapihtar Contributor Loading…
5 tasks
Core 0.16
6
ci: add sync-skills workflow, rename CLAUDE.md → AGENTS.md, move .claude/skills → skills/ Approved All necessary approvals have been made complexity: low
#4303 opened Apr 14, 2026 by ko3n1g Contributor Loading… Core 0.16
3
Fix aux loss computation with per-token loss and dynamic-cp
#4302 opened Apr 14, 2026 by xiaoyao0115 Contributor • Draft
5 tasks
1
Inference | Per-block MoE routing storage for prefix caching complexity: medium Final Review PR is in the "final review" stage
#4301 opened Apr 14, 2026 by lmcafee-nvidia Contributor Loading…
2 of 3 tasks
@lmcafee-nvidia
6
Allow inference to asynchronously await GPU work completion complexity: medium
#4295 opened Apr 14, 2026 by tdene Contributor • Draft
5 tasks
Core 0.16
1
ProTip! What’s not been updated in a month: updated:<2026-03-15.