From NVIDIA Megatron-LM for visibility#18

Open

RaymondLi0 wants to merge 7371 commits into

bigcode-project:multi-query-attentionfrom

RaymondLi0 commented Jan 24, 2023

Collaborator

No description provided.

RaymondLi0 changed the base branch from multi-query-attention to before-merge

June 20, 2023 20:12

RaymondLi0 changed the base branch from before-merge to multi-query-attention

June 20, 2023 20:12

ko3n1g force-pushed the main branch from ad5a627 to cd71d4c Compare

February 18, 2026 09:18

tdene and others added 27 commits

June 5, 2026 16:37


          Fix error in deepseek parser (#5136)

5a2c6fa


          Fix logprob slicing for 0 generated token case (#5167)

9d33299

Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>


          [Perf] Fold frozen linear dgrad matmul (#5092)

0f891a1

Signed-off-by: Chen Cui <chcui@nvidia.com>


          Clamp max_new_tokens in MInf to mirror vllm (#5181)

53d9ba0


          build: add managed = true to [tool.uv] (#5190)

dbf5228

Signed-off-by: Kajal Jain <kajalj@nvidia.com>


          Stabilize GB200 inference perf tests against cold-start noise (#5171)

6204b92

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          nvidia style guide audit for getting started folder (#5168)

947c0b3

Signed-off-by: meg miranda <mmiranda@nvidia.com>


          AI aided audit for Nvidia Style guidance (#5141)

05c76af

Signed-off-by: meg miranda <mmiranda@nvidia.com>
Co-authored-by: Philip Petrakian <pgpetrak@gmail.com>


          Enable selective recompute for norm_out in GDN layers (#4715)

ff5264c


          fix(elastification): align with get_batch + utils refactors (#5194)

adcdf16

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>


          fix(combined-1f1b): release loss-node input storage after combined ba…

5da641c

…ckward (#4909)


          chore(beep boop 🤖): Bump (main) (2026-06-08)

dbf719b


          Avoid stat syscall in rerun result validation (#5107)

17a5a80

Signed-off-by: dimapihtar <dpykhtar@nvidia.com>


          docs: Update Latest News in README.md (#3790)

71e2246


          Fix bug with Megatron-FSDP zero counter not working with decoupled gr…

3e6e32b

…adients. (#4802)

Signed-off-by: Cory Ye <cye@nvidia.com>


          ci: add smoke tests (#5143)

1829fd9


          Add mtp_detach_heads config to detach MTP head inputs (#3456)

71e418e

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          docs: fix install guide NGC container anchor (#5224)

a95d866


          Fuse per-sequence AlltoAll into a unified one in GDN forward (#4913)

48032d7


          Apply MIMO SP/CP sharding with explicit groups and enable THD in non-…

…colocated path (#5150)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          Fix CUDA IMA in fsdp_double_buffer when an FSDP unit's bucket doesn't…

d199bb9

… fit the pool (#4810)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>


          Add named layouts to HyperCommGrid for heterogeneous parallelism (#5148)

ba71ec2

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          Fix wgrad race condition when using double buffers. (#5222)

55638bc

Signed-off-by: Cory Ye <cye@nvidia.com>


          Move uneven DTensor distributed fixture to conftest (#5237)

504b02f


          Route bridge communicator cross-grid P2P through a dedicated process …

b180d2c

…group (#5234)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          Fix test_split_tensor_along_last_dim to actually assert correctness (#…

e03878b

…4710)

Co-authored-by: peibli <lipeibao@126.com>
Co-authored-by: Dmytro Pykhtar <37850217+dimapihtar@users.noreply.github.com>


          chore: rotate oncall schedule

2065b5a

shanhaoli and others added 30 commits

July 15, 2026 18:41


          Fix configured norm epsilon in MambaLayer (#5750)

Signed-off-by: Shanhao <5197744+shanhaoli@users.noreply.github.com>


          [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>


          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>


          Set num_splits to 0 for FA4 inference (#5804)

8bf7365

Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>


          ci: Enhance nightly/mr/weekly error reporting (#5831)

06bb399

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


          feat(docker): Add NCCL installation script and install NCCL 2.30.4 (#…

3927d9b

…5815)

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


          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>


          Add GPTModel to HybridModel migration guide (#5698)

802da56

Signed-off-by: Philip Petrakian <ppetrakian@nvidia.com>


          Pair frozen FSDP backward hooks (#5710)

38626c2

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          Clarify NVIDIA email signing guidance (#5699)

c2f0df5

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          Overlap FSDP communication with compute (#5719)

d981f66

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          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>


          chore(skills): remove Open Plugin manifest (superseded) (#5842)

2aa3645

Signed-off-by: oliver könig <okoenig@nvidia.com>


          return prefix cache hits data from the chat completions api (#5609)

edd4562

Signed-off-by: Siddharth Singh <sidsingh@nvidia.com>


          Refactor data parallel coordinator to enable modular handlers (#5550)

f8e1ac6

Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>


          Pass device IDs to cleanup barrier (#5702)

407ea4f

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          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>


          test(mfsdp): AUT-881 mark test_overlaps_communication_and_compute fla…

9a67ea0

…ky (#5848)

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>


          Inference: Extend default cuda-graph coverage to 512 tokens (#5797)

740c16e


          Fix for sequence-level aux MoE loss being dependent on batch size (#5798

e3fe550

)

Signed-off-by: Oleg Sudakov <osudakov@nvidia.com>
Signed-off-by: Oleg Sudakov <oleg.sudakov@outlook.com>
Co-authored-by: Fei Wu <33940270+YangFei1990@users.noreply.github.com>


          Allow parameterless FSDP root modules (#5711)

61f3114

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          Fix issue where parameter groups with different min/max LRs get overr…

27e312c

…idden at checkpoint load time (#4705)

Signed-off-by: John St John <jstjohn@nvidia.com>
Signed-off-by: John St. John <jstjohn@nvidia.com>


          fix bug where Gemma4 is not working with recompute_granularity = "ful…

acc7e64

…l" (#5324)

Signed-off-by: Yury Parfenov <4665475+warpuv@users.noreply.github.com>
Co-authored-by: Guihong Li <guihongl@nvidia.com>


          Avoid extra MFSDP v2 model-weight sync memcpy (#5834)

4a1f743

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          [experimental] Add experimental/agent_compose placeholder with previe…

167d51d

…w pointer (#5639)

Signed-off-by: Yan Bai <bayan@nvidia.com>


          Stabilize mfsdp_v2 overlap test by enlarging the model (#5846)

53492eb

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>


          Print important dependencies (#5814)

dc57c2b

Signed-off-by: Gautham Kollu <gkollu@nvidia.com>


          Test zero-CTA copy-engine all-gather (#5858)

f258d4f

Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>


          Route Lion through DistributedOptimizer and support single-moment che…

b4ad280

…ckpointing (#5742)

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>


          Fix broken remove_sharded_tensors public API and re-enable its unit t…

3219f38

…est (#5759)

Signed-off-by: Rui Zhu <rui.zhu.rz399@yale.edu>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Dmytro Pykhtar <37850217+dimapihtar@users.noreply.github.com>

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet