Skip to content

From NVIDIA Megatron-LM for visibility #18

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 6,277 commits into
base: multi-query-attention
Choose a base branch
from
Open

From NVIDIA Megatron-LM for visibility #18

RaymondLi0 wants to merge 6,277 commits into from

Conversation

@RaymondLi0
Copy link
Collaborator

@RaymondLi0 RaymondLi0 commented Jan 24, 2023

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
ko3n1g and others added 28 commits January 16, 2026 15:53
@ko3n1g
ci(fix): CI_COMMIT_BRANCH on forks (#2982)
b96cadb 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@yuzhongw-nvidia @ko3n1g
[main] feat(moe): Support gated delta net for Qwen3-Next (1/4) (#1989)
20d66d5 
Signed-off-by: oliver könig <okoenig@nvidia.com>
Co-authored-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci: Add more gb200 nightly tests (#2981)
92f052b 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci(hotfix): gpt3_mcore_te_tp1_pp2_resume_torch_dist_rope_embeddings
4ebd1ad 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci(hotfix): move test to nightly
b50da25 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@yuzhongw-nvidia @jstjohn @deepakn94
[main] feat(moe): Support apply wd to qk layernorm for Qwen3-Next (4/…
622a06a 
…4) (#2753)

Signed-off-by: John St. John <jstjohn@nvidia.com>
Co-authored-by: John St. John <jstjohn@nvidia.com>
Co-authored-by: Deepak Narayanan <2724038+deepakn94@users.noreply.github.com>
@asolergi-nv @ko3n1g
Set token_dtype_code init value in GPTDatasetConfig to fix CI (#2912
b305422 
)

Co-authored-by: oliver könig <okoenig@nvidia.com>
@hexinw-nvidia @rhewett-nv @chtruong814
Re-submit "Various fixes to in-job restarter and better time accounti…
c40b6ea 
…ng of startup operations" (#2954)

Co-authored-by: Russell Hewett <rhewett@nvidia.com>
Co-authored-by: Charlie Truong <chtruong@nvidia.com>
@Phlip79
Use slack-sdk in a different manner (#2950)
100e11e
@lmcafee-nvidia
Inference | Move assert active_request_count > 0. (#2958)
b705148
@parthmannan @xiaoyao0115 @kunlunl
Hybrid Context Parallel Feature (#2282)
98d8c56 
Signed-off-by: tailaim <tailaim@nvidia.com>
Signed-off-by: Parth Mannan <pmannan@nvidia.com>
Co-authored-by: Mcore Bot <mcore-bot@nvidia.com>
Co-authored-by: tailaim <tailaim@nvidia.com>
Co-authored-by: kunlunl <kunlunl@nvidia.com>
Co-authored-by: Kunlun Li <94586211+kunlunl@users.noreply.github.com>
@yuzhongw-nvidia
[main] ci(moe): Add --apply-wd-to-qk-layernorm flag to the gdn test…
829e461 
… case (#2995)
@ko3n1g
ci: Disable step time on `gpt3_moe_mcore_te_tp2_pp2_ep4_etp1_no_mtp_n… (
a191791 
#2991)

Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci: Fix workflows on main (#2990)
b49d810 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@shjwudp @cspades
Make Megatron-FSDP torch.compile compatible (#2425)
35129e7 
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
@cspades
[Megatron-FSDP] Test FP8 activations + parameter sharding with Megatr…
517dfd4 
…on-FSDP fully-shard. Update README. (#2894)

Signed-off-by: Cory Ye <cye@nvidia.com>
@ko3n1g
chore: Escape special chars (#3014)
2f3fa0f 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@deepakn94
Improve memory logging (#2839)
82ea022 
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
@github-actions
chore: rotate oncall schedule
bcdd405
@santhnm2
Add a wrapper function for FA3 _flash_attn_forward call (#2933)
a3615d7 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@ko3n1g
ci(hotfix): Fix unit tests on main
a1ba844 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci(hotfix): Tests on main
e4b18f7 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
chore: Set umask 0002 (#3027)
ba876ef 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@mathemakitten
Make attn mask inversion in-place instead of allocating it again (#3019)
28c7221
@cspades
[Megatron-FSDP] Fix incorrect gradient scaling target. (#3023)
ba456fd 
Signed-off-by: Cory Ye <cye@nvidia.com>
@santhnm2
Ensure that last prefill chunk is handled correctly by Mamba models (#…
096dbeb 
…2897)

Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@nschank
Replaces ModuleSpec with Protocols for some of the inputs to SelfAtte…
90e685b 
…ntion/CrossAttention (#2761)
@Phlip79
Update oncall schedule (#3017)
dd72aee
dimapihtar and others added 30 commits February 10, 2026 16:32
@dimapihtar
update checkpointing documentation (#3347)
76cf11e 
Signed-off-by: dimapihtar <dpihtar@gmail.com>
@tdene
Update golden values to reflect improvements (#3350)
836d473
@sancha @BestJuly
BUGFIX: gpt vs hybrid model mtp naming mismatch (#3334)
2451508 
Co-authored-by: Li Tao <lit@nvidia.com>
@tdene
Disable flaky test (#3354)
8da949e
@jon-barker
re-enable gpt grpo tests (#3348)
bb97791
@asolergi-nv
Fix SFT Pipeline when TP>1 (#3268)
4bce841
@AAnoosheh
Fixes for KD mode (#3342)
f5238ba 
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
@github-actions
chore: rotate oncall schedule
c1169ea
@ko3n1g
chore: Update codeowners file (#3365)
4f025a1 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@sidsingh-nvidia @ko3n1g
Siddharth/fix inference functional tests (#3357)
66ec17e 
Co-authored-by: oliver könig <okoenig@nvidia.com>
@janEbert @ko3n1g
Switch oncall (#3360)
6a9da99 
Co-authored-by: oliver könig <okoenig@nvidia.com>
@AAnoosheh
Add missing RMSNorm to llama train script (#3314)
b6e883b 
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>
@tdene
Fix inference for MTP models (#3297)
cd14090
@yobibyte
Add a logprobs test with real gpt model. (#2870)
6f5de16
@tdene
Add simple GRPO functional test (#3323)
b5d50cb
@ko3n1g
ci: Concurrency control for merge-queue (#3353)
1c245c7 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@chtruong814
ci: Update golden value download script to work with Github (#3335)
d9f075c 
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
@megnvidia
Removing etc from main index page, shifted name of discussions (#3271)
d0b768f 
Signed-off-by: meg miranda <mmiranda@nvidia.com>
@thecaptain789 @Phlip79
fix: correct typos 'seperated' and 'recieved' (#3305)
7d1acf6 
Co-authored-by: thecaptain789 <thecaptain789@users.noreply.github.com>
Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>
@shengf-nv
Improved PyTorch profiler and added PyTorch execution trace (#3273)
2807a4e
@ko3n1g
build: Bump TE on 2.12 (#3371)
1959739 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ci(hotfix): job conditions (#3376)
f06e669 
Signed-off-by: oliver könig <okoenig@nvidia.com>
@github-actions
Update copy-pr-bot.yaml [skip ci]
6467cbc
@sidsingh-nvidia @ISEEKYAN
Record moe routing decisions during inference. (#3034)
faced51 
Co-authored-by: litianjian <litianjian@bytedance.com>
Co-authored-by: Yan Bai <baiyan1996@icloud.com>
@Wohox @dimapihtar
[Main] Fix EP Overlap Bugs for Full-Iter CG (#3164)
dedb6dd 
Co-authored-by: Dmytro Pykhtar <37850217+dimapihtar@users.noreply.github.com>
@maanug-nv
Avoid direct pickle import (#3375)
11a4659 
Signed-off-by: Maanu Grover <maanug@nvidia.com>
@Phlip79
Delete old pretrain_* files (#3359)
fe9279e
@xuwchen
Add Qwen3-VL support with Megatron-FSDP (#2841)
7df15bd
@santhnm2
Refactor Mamba chunked prefill (#3265)
c65fb25 
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
@jstjohn @ko3n1g
Improved parallel logging of learning rate (#3319)
47938af 
Co-authored-by: oliver könig <okoenig@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.