Skip to content

From NVIDIA Megatron-LM for visibility - #18

Open
RaymondLi0 wants to merge 7582 commits into
bigcode-project:multi-query-attentionfrom
NVIDIA:main
Open

From NVIDIA Megatron-LM for visibility#18
RaymondLi0 wants to merge 7582 commits into
bigcode-project:multi-query-attentionfrom
NVIDIA:main

Conversation

@RaymondLi0

@RaymondLi0 RaymondLi0 commented Jan 24, 2023

Copy link
Copy Markdown
Collaborator

No description provided.

@RaymondLi0
RaymondLi0 changed the base branch from multi-query-attention to before-merge June 20, 2023 20:12
@RaymondLi0
RaymondLi0 changed the base branch from before-merge to multi-query-attention June 20, 2023 20:12
wujingyue and others added 27 commits July 14, 2026 22:56
@wujingyue
Exercise nested MFSDP CUDA graph capture (#5796)
a327943 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@Mellonta
Implement Quantile Balancing in MoE (#5349)
33706f9 
Signed-off-by: Haoran Zhang <haoranz@nvidia.com>
@wujingyue
Unset NCCL overrides for MFSDP v2 tests (#5794)
f8c9911 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@wujingyue
Fix MegatronFSDP root module hook dispatch (#5808)
4bf7fca 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@balasaajay
Pin cudnn-fe and cuTeDSL version (#5812)
834abc1 
Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
@github-actions
chore: rotate oncall schedule
5abab93
@sidsingh-nvidia
Inference: Add the nemotron_v3 reasoning parser (#5634)
82e9dc6 
Signed-off-by: Siddharth Singh <sidsingh@nvidia.com>
@jinhangchoi @cspades
Missing moe_router_dtype causes unexpected downcast in ModelOpt examp…
ccdfa7d 
…le (#5810)

Signed-off-by: Jinhang Choi <jinhangc@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
@wujingyue @svcnvidia-nemo-ci
Avoid FSDP unit terminology in MFSDP v2 (#5793)
5e45ccf 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Signed-off-by: svcnvidia-nemo-ci <svcnvidia-nemo-ci@nvidia.com>
Co-authored-by: svcnvidia-nemo-ci <svcnvidia-nemo-ci@nvidia.com>
@shanhaoli
Fix configured norm epsilon in MambaLayer (#5750)
7711346 
Signed-off-by: Shanhao <5197744+shanhaoli@users.noreply.github.com>
@Connor-XY @claude
[refactor] Common combined-1F1B schedule-plan base (1/4 of #4798) (#4941
ffbe018 
)

Signed-off-by: Yan Xu <yxu1@nvidia.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@svcnemo-autobot
chore(tests): AUT-851 move NCCL defaults from run_ci_test.sh to conft…
ecf5547 
…est (#5826)

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@santhnm2
Set num_splits to 0 for FA4 inference (#5804)
8bf7365 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@balasaajay
ci: Enhance nightly/mr/weekly error reporting (#5831)
06bb399 
Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
@balasaajay
feat(docker): Add NCCL installation script and install NCCL 2.30.4 (#…
3927d9b 
…5815)

Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
@shanmugamr1992
IMA fix by making the copy of book keeping buffer to GPU blocking (#5715
53a2dd5 
)

Signed-off-by: Shanmugam Ramasamy <111910568+shanmugamr1992@users.noreply.github.com>
@Phlip79
Add GPTModel to HybridModel migration guide (#5698)
802da56 
Signed-off-by: Philip Petrakian <ppetrakian@nvidia.com>
@wujingyue
Pair frozen FSDP backward hooks (#5710)
38626c2 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@wujingyue
Clarify NVIDIA email signing guidance (#5699)
c2f0df5 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@wujingyue
Overlap FSDP communication with compute (#5719)
d981f66 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@ko3n1g @claude
chore(skills): add Regent Open Plugin manifest (#5840)
7b2a97b 
Signed-off-by: oliver könig <okoenig@nvidia.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
@ko3n1g
chore(skills): remove Open Plugin manifest (superseded) (#5842)
2aa3645 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@sidsingh-nvidia
return prefix cache hits data from the chat completions api (#5609)
edd4562 
Signed-off-by: Siddharth Singh <sidsingh@nvidia.com>
@santhnm2
Refactor data parallel coordinator to enable modular handlers (#5550)
f8e1ac6 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@wujingyue
Pass device IDs to cleanup barrier (#5702)
407ea4f 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@Marks101 Markus Schnös
Fix averaging for MoE z-loss metric tracking (#3199)
4abeecc 
Signed-off-by: Markus Schnös <markus.schnoes@deepl.com>
Co-authored-by: Markus Schnös <markus.schnoes@deepl.com>
@svcnemo-autobot
test(mfsdp): AUT-881 mark test_overlaps_communication_and_compute fla…
9a67ea0 
…ky (#5848)

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
xuantengh and others added 30 commits August 11, 2026 00:23
@xuantengh
Bump FLA to 0.5.1 and integrate GDN2 kernels (#5765)
0c1fc4c 
Signed-off-by: Xuanteng Huang <xuantengh@nvidia.com>
@JavaZeroo @guihong-nv
Update docstring for moe_router_topk_scaling_factor (#2178)
67a355e 
Signed-off-by: JavaZeroo <2487163254@qq.com>
Co-authored-by: Guihong Li <guihongl@nvidia.com>
@asolergi-nv
Add --ckpt-drop-redundant-extra-state to skip persisting redundant TE…
105445d 
… _extra_state (#5552)

Signed-off-by: asolergi-nv <asolergibert@nvidia.com>
@wujingyue
Add construction-scoped MFSDP v2 fully shard options (#6127)
78901d8 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@jenchen13
Support explicit Hugging Face finetune data files in finetune.sh (#6413)
5e86dfa 
Signed-off-by: Jennifer Chen <jennifchen@nvidia.com>
@deepakn94
Add megatron.core.telemetry base layer and observability guide (#6415)
d7d7a6a 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
@YangFei1990
[NCCL EP] Support eager mode and overflow handling (#6229)
8971d19 
Signed-off-by: YangFei1990 <feiw@nvidia.com>
@shurkat-nvidia @claude
Gate expensive work behind log levels (#5805)
db57c15 
Signed-off-by: Skand Hurkat <shurkat@nvidia.com>
Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
@santhnm2 @shanmugamr1992 @claude
Fixes for SP + MTP all-gathers (#5748)
8018459 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Signed-off-by: shanmugamr1992 <shanmugamr1992@gmail.com>
Co-authored-by: shanmugamr1992 <shanmugamr1992@gmail.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
@mkhona-nvidia
TE Layernorm dtype guard with fp32 residuals (#6272)
7a76dff 
Signed-off-by: mkhona <mkhona@nvidia.com>
@deepakn94 @claude
Treat LayerWise bucket_size as a soft minimum and fill would-be paddi…
e0ccd26 
…ng with real parameters (#5415)

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
@devnkong @deepakn94
@claude @mkhona-nvidia Mikail Khona root @rogerwaleffe @santhnm2
Gated Delta Product (GDP) implementation (#6074)
8e70f69 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Signed-off-by: Mikail Khona <mkhona@cw-dfw-cs-001-vscode-01.cm.cluster>
Signed-off-by: Mikail Khona (NVIDIA) <mkhona@nvidia.com>
Signed-off-by: Kezhi Kong <kezhik@nvidia.com>
Co-authored-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Mikail Khona (NVIDIA) <mkhona@nvidia.com>
Co-authored-by: Mikail Khona <mkhona@cw-dfw-cs-001-vscode-01.cm.cluster>
Co-authored-by: root <root@nvl72018-T18.cm.cluster>
Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>
Co-authored-by: Keshav Santhanam <ksanthanam@nvidia.com>
@wujingyue @claude
fix(mfsdp): refresh V2 compute weights after optimizer step (#6336)
1d402ad 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@chengcuiping @svcnvidia-nemo-ci @janEbert
[core] Clarify GPT output_processor callback contract (#5739)
c3d44dc 
Signed-off-by: chengcuiping <96756894+chengcuiping@users.noreply.github.com>
Signed-off-by: svcnvidia-nemo-ci <svcnvidia-nemo-ci@nvidia.com>
Signed-off-by: janEbert <janpabloe@nvidia.com>
Co-authored-by: svcnvidia-nemo-ci <svcnvidia-nemo-ci@nvidia.com>
Co-authored-by: janEbert <janpabloe@nvidia.com>
@wujingyue
Validate MFSDP module phase transitions (#6359)
03318a8 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@wujingyue
Refine MFSDP v2 configuration validation (#6333)
80cda5d 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@Phlip79
Add hybrid layer config definitions (#6410)
f3eb0df 
Signed-off-by: Philip Petrakian <ppetrakian@nvidia.com>
@Phlip79
Propagate RL runtime config updates to module configs (#6423)
da18ed6 
Signed-off-by: Philip Petrakian <ppetrakian@nvidia.com>
@fanshiqing
[fix] GTP+recompute: keep adjacent GTP recompute weights off the same…
e22bcb0 
… gather buffer (#6407)

Signed-off-by: Shiqing Fan <shiqingf@nvidia.com>
@github-actions
Update copy-pr-bot.yaml [skip ci]
a0394fe
@wujingyue
Add MFSDP v2 design document (#6431)
49a49a9 
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
@Phlip79
Make core-nemo fallback owner for top-level files (#6358)
8af9b04 
Signed-off-by: Philip Petrakian <ppetrakian@nvidia.com>
@ko3n1g
ci(auth): treat svcnemo-autobot as internal (#6436)
725949a 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@mkhona-nvidia @deepakn94 @santhnm2
Add mixed-precision (FP32) LM output logits (#6252)
897649d 
Signed-off-by: mkhona <mkhona@nvidia.com>
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Co-authored-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Keshav Santhanam <ksanthanam@nvidia.com>
@asolergi-nv
Eliminate the ShardedObject all_gather_object on the FullyParallel lo…
14346b6 
…ad path (#5551)

Signed-off-by: asolergi-nv <asolergibert@nvidia.com>
@github-actions
chore: rotate oncall schedule
15c83d2
@JF-D
GTP + PartialCG: Classify latent projections for partial CUDA graphs (#…
81fe7c7 
…6446)

Signed-off-by: Jiangfei Duan <jiangfeid@nvidia.com>
@deepakn94 @claude
Balance LayerWise optimizer shards by Newton-Schulz cost, not paramet…
8534490 
…er size (#6379)

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
@deepakn94
Suppress noisy external-library log lines on non-rank-0 processes (#5590
fad6111 
)

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
@balasaajay
feat(ci): enhance triage script to assign on-call assignee dynamically (
b2f2888 
#6469)

Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
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.