Skip to content

From NVIDIA Megatron-LM for visibility#18

Open
RaymondLi0 wants to merge 6615 commits intobigcode-project:multi-query-attentionfrom
NVIDIA:main
Open

From NVIDIA Megatron-LM for visibility#18
RaymondLi0 wants to merge 6615 commits intobigcode-project:multi-query-attentionfrom
NVIDIA:main

Conversation

@RaymondLi0
Copy link
Copy Markdown
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
yobibyte and others added 27 commits February 25, 2026 17:50
@yobibyte
Fix default cuda graph persist arg. Add persist to rl common.sh. (#3584)
b359981
@sidsingh-nvidia
Optimize away add request overheads in dummy ep cuda-graphed forward …
60a25aa 
…passes (#3525)
@ko3n1g
ci: Test docs build (#3583)
e7d80bd 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: Fix docs build for release (#3597)
80ef2ae 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: Remove secrets (#3598)
2561a59 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: Define secrets (#3599)
19444b9 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: gh-release-from-tag (#3600)
a08ad25 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
Ko3n1g/ci/remove twine username (#3601)
ac449c4 
Signed-off-by: oliver könig <[email protected]>
@maanug-nv
Add training code to MCore wheel (#3573)
9088d4f 
Signed-off-by: Maanu Grover <[email protected]>
@vasunvidia @yaox12
FP8 attention knob for nvFP4 recipe (#3363)
dd4c0b7 
Co-authored-by: Xin Yao <[email protected]>
@guyueh1 @yaox12
Fix error with --load-main-params-from-ckpt (#3569)
c2782fd 
Signed-off-by: Guyue Huang <[email protected]>
Co-authored-by: Guyue Huang <[email protected]>
Co-authored-by: Xin Yao <[email protected]>
@ko3n1g
ci: Create comment (#3610)
c921891 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: Skip cleanup-taint-node jobs during deployments (#3612)
9100119 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: No comment for release workflow (#3615)
6161f7a 
Signed-off-by: oliver könig <[email protected]>
@ko3n1g
ci: Re-add release tag prefix (#3619)
36d8a9d 
Signed-off-by: oliver könig <[email protected]>
@chtruong814
docs: Fix version picker urls (#3621)
1b1f5c4 
Signed-off-by: Charlie Truong <[email protected]>
@chtruong814
ci: Increase changelog generation max PRs fetched (#3620)
027e0f3 
Signed-off-by: Charlie Truong <[email protected]>
@yobibyte
Add debug info to an assert. (#3588)
18c94d2
@sbak5
fix: async_utils: explicit GC in persistent checkpoint worker loop (#…
e1a9ac9 
…3591)
@CodersAcademy006 @yaox12
Fix: Perform sigmoid calculation in fp32 for aux loss stability (#2765)
5f668c1 
Co-authored-by: Xin Yao <[email protected]>
@saakshigupta2002 @yaox12
fix: forward use_te_activation_func flag in non-MoE GPT layer spec (#…
d3c10df 
…3300)

Co-authored-by: Xin Yao <[email protected]>
@faradawn
Multimodal: Limit transformer version in Dockerfile (#3448)
7e3f670 
Signed-off-by: nvidia <[email protected]>
@tdene
Track and plot per-token off-policy in RL (#3515)
6c0b9c6
@ko3n1g
Revert "Add single-process checkpoint save to avoid forked multiproce… (
a7c207f 
#3630)
@faradawn
Multimodal: fix VQA dataset selection (#3464)
6503bf8 
Signed-off-by: Faradawn Yang <[email protected]>
@faradawn
Multimodal: Fix multimodal training example - tokenizer, Triton Cache…
2f549e5 
… Manager patch, docs (#3507)

Signed-off-by: Faradawn Yang <[email protected]>
@santhnm2
Support TP > GQA for inference (#3627)
afbce84 
Signed-off-by: Keshav Santhanam <[email protected]>
nanz-nv and others added 30 commits March 24, 2026 02:44
@nanz-nv @Victarry @Phlip79
Forced load imbalance (#3380)
7da5b28 
Co-authored-by: Dennis(Zhenhuan) Liu <[email protected]>
Co-authored-by: Philip Petrakian <[email protected]>
@Phlip79 @claude
Add /claude copy command (#3978)
9e28104 
Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>
@yashaswikarnati @dimapihtar @claude
Add multi-module heterogeneous parallelism support for MIMO model (#3211
70a89af 
)

Signed-off-by: ykarnati <[email protected]>
Co-authored-by: Dmytro Pykhtar <[email protected]>
Co-authored-by: Claude Opus 4.5 <[email protected]>
@kinjalpatel27 @dimapihtar
added vllm fakequant export support (#3050)
99638d0 
Co-authored-by: Dmytro Pykhtar <[email protected]>
@jenchen13 @claude
fix(modelopt): use bash array for MLM_EXTRA_ARGS to preserve quoting (#…
a9e4437 
…4002)

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
@sbak5
fix: use dump file prefix for NCCL flight recorder temp files (#3955)
34e1e97
@cluster2600 @maanug-nv
Fix PersistentAsyncCaller.__del__ crash during Python shutdown (#3781)
b44a23c 
Signed-off-by: Maxime Grenu <[email protected]>
Signed-off-by: Maxime Kawawa-Beaudan <[email protected]>
Co-authored-by: Maxime Kawawa-Beaudan <[email protected]>
Co-authored-by: Maanu Grover <[email protected]>
@chtruong814
ci: Run L1 MBridge tests in merge queue (#4009)
0065f9f 
Signed-off-by: Charlie Truong <[email protected]>
@Phlip79
Update Claude review (#3980)
74b0e69
@nschank @yashaswikarnati
Migrate MoeLayer submodules from ModuleSpec to Protocols (#3426)
4f49f2f 
Co-authored-by: Yashaswi Karnati <[email protected]>
@maanug-nv
Guard non-core imports (#3993)
3aa7396 
Signed-off-by: Maanu Grover <[email protected]>
@maanug-nv
Fix config compatibility with Megatron-Core (#3995)
7bca3c8 
Signed-off-by: Maanu Grover <[email protected]>
@yashaswikarnati @claude
Add MimoOptimizer for heterogeneous parallelism (#4019)
d86ba0b 
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
@Wohox
[Main] Support EP Overlap's Dynamic Computation Stream For Full-Iter …
6a7b68e 
…CUDA Graph (#3820)
@sbak5
fix: Handle quantized CUDA tensors in async checkpoint writer (#3845)
a8530db
@CarlosGomes98
accept hooks marked with with_kwargs when using te.ops.sequential (#4000
934edfa 
)
@github-actions
chore: rotate oncall schedule
d31a21f
@nschank
Use GroupedMLPSubmodules for InferenceGroupedMLP (#3743)
1aee2d7
@yashaswikarnati @claude
Fix 2D tensor communication for asymmetric DP in Bridge Communicator (#…
694c3a9 
…4021)

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
@yashaswikarnati @claude
Add distributed checkpoint support for non-colocated MiMo (#4020)
1df5591 
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
@lmcafee-nvidia @claude
CUDA graph support for prefix caching on hybrid models (#3922)
d45fa3e 
Co-authored-by: Claude Opus 4.6 <[email protected]>
@deepakn94 @claude
Add ability to perform local gradient accumulation in FP32 for a subs…
c586f6d 
…et of parameters in the model (#4028)

Co-authored-by: Claude Opus 4.6 <[email protected]>
@santhnm2
Miscellaneous MXFP8 inference fixes (#4017)
09cce75 
Signed-off-by: Keshav Santhanam <[email protected]>
@WanZzzzzz @deepakn94
Use torch.int64 for grad_num_zero accumulation (#4015)
a01a6c5 
Signed-off-by: qiyuw <[email protected]>
Co-authored-by: qiyuw <[email protected]>
Co-authored-by: Deepak Narayanan <[email protected]>
@santhnm2
Make text generation server hostname configurable (#3935)
548028b 
Signed-off-by: Keshav Santhanam <[email protected]>
@mchrzanowski @claude
Add --muon-coefficient-type argument for Muon optimizer (#3927)
0842ca2 
Co-authored-by: root <[email protected]>
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
Co-authored-by: root <[email protected]>
Co-authored-by: root <[email protected]>
Co-authored-by: root <[email protected]>
Co-authored-by: root <[email protected]>
Co-authored-by: root <[email protected]>
Co-authored-by: root <[email protected]>
@i-riyad @santhnm2
Pass gracefully if token_id not found in message (#3862)
606afda 
Signed-off-by: rislam <[email protected]>
Signed-off-by: Keshav Santhanam <[email protected]>
Co-authored-by: Keshav Santhanam <[email protected]>
@santhnm2
Improve load balancing behavior for prefix cache-aware routing (#3930)
0528a40 
Signed-off-by: Keshav Santhanam <[email protected]>
@sakgoyal
Refactor setup.py to use get_pybind_include (#3658)
58e0b85
@ko3n1g
build: Bump TE to 2.14 (#4025)
3758b54 
Signed-off-by: oliver könig <[email protected]>
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.