Skip to content

Fetch from nvidia Megatron-LM #5

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.

Sign up for GitHub

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

Jump to bottom
Open
RaymondLi0 wants to merge 6,225 commits into
base: load-iter
Choose a base branch
from
Open

Fetch from nvidia Megatron-LM #5

RaymondLi0 wants to merge 6,225 commits into from

Conversation

@RaymondLi0
Copy link

@RaymondLi0 RaymondLi0 commented Aug 3, 2022

No description provided.

deepakn94 and others added 30 commits November 13, 2025 21:45
@deepakn94 @aklife97
Fix aux-loss logging for hybrid models (#2197)
b7ef391 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Abhinav Khattar <akhattar@nvidia.com>
@deepakn94 @rogerwaleffe
Update flops calculation (for throughput) for hybrid MoEs (#2198)
610a75e 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>
@deepakn94 @rogerwaleffe
Add MoE layer type to hybrid models (#2196)
2751749 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>
@tdene
Tensorize dynamic inference mixed sampling (bis) (#2231)
9be7c7b
@ko3n1g
Revert "Add MoE layer type to hybrid models (#2196)"
c4f83f0 
This reverts commit 2751749.
@ko3n1g
ci(hotfix): Checkout repo before install check
41eecc4 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
chore: Fix codeowners (#2264)
c4ba666 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ananthsub @jaredcasper
Allow loading checkpoint from iteration 0 (#2199)
4696d42 
Signed-off-by: Ananth Subramaniam <ansubramania@nvidia.com>
Co-authored-by: Jared Casper <155158+jaredcasper@users.noreply.github.com>
@chtruong814
ci: Skip install test in merge queue (#2281)
a2d8519 
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
@deepakn94 @rogerwaleffe
Add MoE layer type to hybrid models (#2259)
9a1c0d0 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>
@Autumn1998
Add the Hybrid-EP backend to the Flex Dispatcher (#2176)
3df2009
@zhongbozhu @xrennvidia @yaox12
[MAIN][NVFP4] Support NVFP4 MOE with Proper Padding (#1985)
e8b9df1 
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>
Signed-off-by: Xiaowei Ren <xren@nvidia.com>
Co-authored-by: Xiaowei Ren <xren@nvidia.com>
Co-authored-by: Xin Yao <xiny@nvidia.com>
@kevalmorabia97
Update ModelOpt example readmes and advanced usage (#2273)
a755887 
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
@lmcafee-nvidia
Fix UVM compatibility with CUDA 13. (#2243)
dcd3b39
@ko3n1g
ci: Add flaky marker to LTS tests (#2290)
5e3fa28 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@lmcafee-nvidia @tdene
Dynamic engine suspend/resume via prefill. (#1982)
29eed5d 
Co-authored-by: Teodor-Dumitru Ene <tene@nvidia.com>
Co-authored-by: Mcore Bot <mcore-bot@nvidia.com>
Co-authored-by: Teodor-Dumitru Ene <teodord.ene@gmail.com>
@ko3n1g
Revert "Dynamic engine suspend/resume via prefill. (#1982)"
3b83c3f 
This reverts commit 29eed5d.
@yanring @yaox12
fix: Pass the timeout argument for the EP group (#2268)
19d0422 
Co-authored-by: Xin Yao <xiny@nvidia.com>
@yaox12 @yanring
JIT for MoE router and preprocess (#1919)
efdc681 
Signed-off-by: Xin Yao <xiny@nvidia.com>
Co-authored-by: Zijie Yan <zijiey@nvidia.com>
@tdene
Hotfix to CI, until the fix gets reviewed (#2298)
00884a8
@tdene
Add functional test for DP coordinator throughput (#2189)
f885d9c 
Co-authored-by: root <root@pool0-01101.cm.cluster>
@tdene
Add asyncio Queue like in Python 3.13 (#2224)
70db86a
@lmcafee-nvidia @tdene
Fixes for PR#1982 (#2303)
744505e 
Co-authored-by: Mcore Bot <mcore-bot@nvidia.com>
Co-authored-by: Teodor-Dumitru Ene <tene@nvidia.com>
Co-authored-by: Teodor-Dumitru Ene <teodord.ene@gmail.com>
@santhnm2 @tdene
Fix PP KV cache allocation and enable multi-node PP inference (#2182)
314a378 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Co-authored-by: Teodor-Dumitru Ene <teodord.ene@gmail.com>
@lmcafee-nvidia @ko3n1g
Revert active-buffer-size-gb arg name. (#2257)
21968ea 
Co-authored-by: oliver könig <okoenig@nvidia.com>
@pablo-garay
feat: check: api backwards compatibility (#2251)
712dff8 
Signed-off-by: Pablo Garay <pagaray@nvidia.com>
@santhnm2
Add MambaInferenceStateConfig dataclass (#2265)
6c8cdd5 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@santhnm2
Fix typo in inference example (#2311)
dc473f9 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@pablo-garay
feat: initialization of API backward compatibility verification (#2310)
7dec856 
Signed-off-by: Pablo Garay <pagaray@nvidia.com>
@jaredcasper @rogerwaleffe @santhnm2
Fix Mamba TP and remove confusing legacy initialization (#2202)
e4b7259 
Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>
Co-authored-by: Keshav Santhanam <ksanthanam@nvidia.com>
Phlip79 and others added 30 commits December 17, 2025 23:15
@Phlip79
Use pull_request_target and don't use uv (#2702)
d321026
@buptzyb @jaredcasper
Optimize TE cudagraph input memory (#2392)
94b4759 
Signed-off-by: Robin Zhang <robinz@nvidia.com>
Co-authored-by: Jared Casper <155158+jaredcasper@users.noreply.github.com>
@ko3n1g
ci(fix): Pin gojq to stable version (#2480)
c7e5489 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@sidsingh-nvidia
NVLS - fused reduce-scatter + residual + rms-norm + all-gather kernel (…
f19b59e 
…#2599)
@lmcafee-nvidia
Default UVM level to 0. (#2450)
0170e70
@sbhavani
docs: improve documentation organization and add additional guides (#…
0b13c98 
…2671)
@chtruong814
Revert "Default UVM level to 0. (#2450)" (#2713)
d81b37b 
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
@maanug-nv
Add missing imports in no-triton fallback (#2711)
a6f822c 
Signed-off-by: Maanu Grover <maanug@nvidia.com>
@lmcafee-nvidia
Fixes for #2450. (#2714)
1503c33
@tdene
Add RL parameter to set parallel generation tasks (#2712)
1a2257b
@wdykas
Refit prep 3 (#2708)
30694e0 
Co-authored-by: William Dykas <wdykas@cw-pdx-cs-001-vscode-02.cm.cluster>
Co-authored-by: root <root@gpu-h100-0371.cm.cluster>
Co-authored-by: root <root@gpu-h100-0159.cm.cluster>
@ko3n1g
chore: Add cudagraph codeowners (#2720)
000c2e2
@Phlip79
[docs] Add developer section to docs (#2717)
703bc36
@tdene
Fix UVM argument for RL (#2722)
7f471d7
@Phlip79
[dcos] Update docs title to Megatron Core (#2729)
ddf691d
@HaochenYuan @Phlip79
remove fp16 assert in moe_grouped_gemm & EP (#2495)
4193f3a 
Co-authored-by: Philip Petrakian <pgpetrak@gmail.com>
@jenchen13 @jaredcasper
Improve ModelOpt paths & add more Nemotron/hybrid model support (#2131)
a057662 
Signed-off-by: Jennifer Chen <jennifchen@nvidia.com>
Co-authored-by: Jared Casper <155158+jaredcasper@users.noreply.github.com>
@asolergi-nv @deepakn94
Add options to improve data loader initialization time, especially at…
cfd980b 
… scale (#2445)

Signed-off-by: Antoni-Joan Solergibert <asolergibert@nvidia.com>
Co-authored-by: Deepak Narayanan <2724038+deepakn94@users.noreply.github.com>
@github-actions
Update copy-pr-bot.yaml [skip ci]
1c67e7e
@ko3n1g
ci: Fix copy-pr-bot update (#2736)
8ea3b8d 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@Phlip79
Add oncall to all new PRs (#2734)
5b1ef07
@tomlifu
Hsdp register submesh fix lifuz mirror (#2467)
cc1b0b5 
Signed-off-by: Lifu Zhang <lifuz@login-lyris01.lyris.clusters.nvidia.com>
Co-authored-by: Lifu Zhang <lifuz@login-lyris01.lyris.clusters.nvidia.com>
@shanmugamr1992
Adding stop word support (#2685)
1febe9f 
Co-authored-by: Shanmugam Ramasamy <shanmugamr@cw-dfw-cs-001-login-01.cm.cluster>
@Phlip79
Fix oncall assign (#2737)
a477766
@santhnm2 @kanz-nv
Add support for non-decode CUDA graphs for Mamba models (#2474)
f5d4c3a 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Co-authored-by: Kan Zhu <kanz@nvidia.com>
@mathemakitten
Update sequence packing case when dummy PackedSeqParams are used (#2743)
0a77122
@youngeunkwon0405
feat: manual registration mode for nccl-ub option when using megatron…
876a046 
…-fsdp (#2661)

Signed-off-by: Youngeun Kwon <youngeunk@nvidia.com>
@github-actions
chore: rotate oncall schedule
ede9ae4
@Phlip79
Update oncall for next few weeks (#2748)
3cf7a63
@nschank
Prep work for migrating to types from ModuleSpec (#2668)
dd7c9f4
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.