Fetch from nvidia Megatron-LM #5

RaymondLi0 · 2022-08-03T20:32:25Z

No description provided.

Signed-off-by: Asha Anoosheh <[email protected]>

…ate (#2183) Signed-off-by: Keshav Santhanam <[email protected]>

Signed-off-by: dimapihtar <[email protected]>

Check-breaking was force merged given special circumstances/special need (requested to me directly)

…erence (#2008)" (#2360) Signed-off-by: Charlie Truong <[email protected]>

Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Jorge Albericio <[email protected]>

Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Jorge Albericio <[email protected]> Co-authored-by: Jon Barker <[email protected]>

Signed-off-by: Charlie Truong <[email protected]>

Signed-off-by: Ananth Subramaniam <[email protected]>

Signed-off-by: Chen Cui <[email protected]>

Signed-off-by: Charlie Truong <[email protected]>

… input

Signed-off-by: Pablo Garay <[email protected]>

Signed-off-by: Charlie Truong <[email protected]>

Signed-off-by: Pablo Garay <[email protected]>

Signed-off-by: oliver könig <[email protected]>

…efill (#2358) Co-authored-by: Siddharth Singh <[email protected]>

Co-authored-by: Xin Yao <[email protected]>

Signed-off-by: Xin Yao <[email protected]>

…erlap (#2236) Signed-off-by: Hongbin Liu <[email protected]> Signed-off-by: Pingtian Li <[email protected]> Co-authored-by: Pingtian Li <[email protected]>

Co-authored-by: William Dykas <[email protected]>

Signed-off-by: Keshav Santhanam <[email protected]>

…o empty parameters in DDP (#2652) Signed-off-by: John St. John <[email protected]>

#2773)

…ice correctly for CUDA context creation / hypothetical memory allocations (#2710) Co-authored-by: Deepak Narayanan <[email protected]>

…#2723) Signed-off-by: John St John <[email protected]> Signed-off-by: John St. John <[email protected]>

Signed-off-by: oliver könig <[email protected]>

Signed-off-by: John St. John <[email protected]>

Signed-off-by: Robin Zhang <[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: William Dykas <[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]> 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]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]>

Signed-off-by: oliver könig <[email protected]>

Signed-off-by: Robin Zhang <[email protected]>

…2752)

…#2751) Co-authored-by: Philip Petrakian <[email protected]>

…2738)" (#2884) Signed-off-by: Charlie Truong <[email protected]>

Signed-off-by: kunlunl <[email protected]> Signed-off-by: jianbinc <[email protected]> Co-authored-by: jianbinc <[email protected]> Co-authored-by: Cory Ye <[email protected]>

Co-authored-by: Xin Yao <[email protected]>

AAnoosheh and others added 30 commits November 21, 2025 22:55

Refactor KD to use ModelOpt plugins file (v2) (#2355)

63d4e7d

Signed-off-by: Asha Anoosheh <[email protected]>

Prevent unnecessarily overwriting the default Hugging Face chat templ…

29a810e

…ate (#2183) Signed-off-by: Keshav Santhanam <[email protected]>

add FIM dataset support (#2291)

7994405

Signed-off-by: dimapihtar <[email protected]>

Update DEFAULT_BASELINE in workflow configuration

e35495d

Check-breaking was force merged given special circumstances/special need (requested to me directly)

Revert "Explicitly zero out padding token activations for dynamic inf…

233b5b0

…erence (#2008)" (#2360) Signed-off-by: Charlie Truong <[email protected]>

Clean up DP coord code & unit test (#2277)

90c8536

[4/4] Merge Megatron-RL into LM (#2002)

8daf046

Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Jorge Albericio <[email protected]>

Update coordinator control logic to be compatible with RL (#2227)

53bbf7a

Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Robert Kirby <[email protected]> Co-authored-by: Jorge Albericio <[email protected]> Co-authored-by: Jon Barker <[email protected]>

ci: Update backwards compat check baseline to 53bbf7a (#2361)

8954e04

Signed-off-by: Charlie Truong <[email protected]>

Account for test regression caused by prints (#2354)

d313c6d

Remove dependency on megatron.training within megatron.core (#2274)

14464d1

Signed-off-by: Ananth Subramaniam <[email protected]>

Fixes for gpt-oss (#2038)

9873958

Signed-off-by: Chen Cui <[email protected]>

update

26b2e72

[HOT FIX] Fix bug of hybrid-ep backend in flex-dispatcher (#2286)

326ec8c

ci: Remove nemo-ci environment (#2364)

17cd106

Signed-off-by: Charlie Truong <[email protected]>

ci: Pass COMMUNITY_PROJECT_ID to community bot (#2366)

278e058

Signed-off-by: Charlie Truong <[email protected]>

ci: Remove environment from community-bot (#2376)

d61029f

Signed-off-by: Charlie Truong <[email protected]>

monitoring & results in mcore

9269dda

Add mbridge_ref input to select MBridge branch

77b65ed

Fix: Use correct repo NVIDIA-NeMo/Megatron-Bridge and add mbridge_ref…

aa7a564

… input

gha action

7f70e22

Signed-off-by: Pablo Garay <[email protected]>

ci: Bump commit for api check to d61029f (#2386)

c28b84e

Signed-off-by: Charlie Truong <[email protected]>

tidy / synthesize / enhance

ab1e26e

Signed-off-by: Pablo Garay <[email protected]>

Merge branch 'main' of https://github.com/NVIDIA/Megatron-LM

56e8810

Revert: trigger_mbridge_tests.yml‎ file change (#2389)

bc242d9

build: Upgrade deps (#2289)

49eef58

Signed-off-by: oliver könig <[email protected]>

Change KV cache init to empty to speedup graph recording and first pr…

2a51d86

…efill (#2358) Co-authored-by: Siddharth Singh <[email protected]>

Handle UVM compile lock issues (#2299)

4c7d3d6

Remove experimental tags for fused kernels. (#2233)

14b791b

Co-authored-by: Xin Yao <[email protected]>

Reduce Overhead in Timers (#2210)

ffb8c35

Signed-off-by: Xin Yao <[email protected]>

lhb8125 and others added 30 commits January 6, 2026 03:27

fix(moe): Support HybridEP and reduce memory overhead for 1F1B A2A ov…

a56a0b0

…erlap (#2236) Signed-off-by: Hongbin Liu <[email protected]> Signed-off-by: Pingtian Li <[email protected]> Co-authored-by: Pingtian Li <[email protected]>

Inference memory test (#2724)

4a584cb

Co-authored-by: William Dykas <[email protected]>

Update copy-pr-bot.yaml [skip ci]

44e5efd

Move batch invariance mode init to initialize.py (#2832)

10561f9

Signed-off-by: Keshav Santhanam <[email protected]>

Move full model init to cuda stream to avoid race condition leading t…

de56227

…o empty parameters in DDP (#2652) Signed-off-by: John St. John <[email protected]>

[docs] Cleanup homepage (#2823)

5af715b

[docs] Update oncall doc (#2822)

8a59fb5

Make default for rerun_mode=disabled not terminate with non-fatal rer… (

c2327e7

#2773)

Bugfix: ensure spawned persistent checkpoint worker sets its CUDA dev…

5950971

…ice correctly for CUDA context creation / hypothetical memory allocations (#2710) Co-authored-by: Deepak Narayanan <[email protected]>

Implementation of a more flexible optimizer/scheduler override system (…

41cedd6

…#2723) Signed-off-by: John St John <[email protected]> Signed-off-by: John St. John <[email protected]>

chore: rotate oncall schedule

64bc482

ci(fix): PyPI upload (#2843)

144049d

Signed-off-by: oliver könig <[email protected]>

ci(fix): Don't fail on empty var (#2850)

ed5bc5c

Signed-off-by: oliver könig <[email protected]>

Add RL support for MOEs (#2742)

c64f227

ci(fix): GH release version tag (#2854)

ae1a2f1

Signed-off-by: oliver könig <[email protected]>

ci(hotfix): Disable flaky test

7e5e16b

Signed-off-by: oliver könig <[email protected]>

Reduce the scope of the side stream around DDP initialization (#2852)

4ba9f46

Signed-off-by: John St. John <[email protected]>

Manually update first oncall rotation (#2855)

49dfee2

Remove flaky iteration time functional test (#2862)

5fa42ec

Signed-off-by: Robin Zhang <[email protected]>

build: Bump jet-client (#2876)

8d6c604

Signed-off-by: oliver könig <[email protected]>

Change oncall team name (#2861)

8b10a64

Dynamic Inference | Evict and re-compute context requests. (#2738)

c8ac1fe

Fix CUDA RNG Tracker (#2641)

965cfd3

Signed-off-by: Robin Zhang <[email protected]>

[main] feat(moe): Support attention output gate for Qwen3-Next (3/4) (#…

65dccab

…2752)

[main] feat(moe): Support moe shared expert gate for Qwen3-Next (2/4) (…

bddd0a8

…#2751) Co-authored-by: Philip Petrakian <[email protected]>

[docs] Fix docs and add generation doc (#2882)

0b5be24

Revert "Dynamic Inference | Evict and re-compute context requests. (#…

43b4471

…2738)" (#2884) Signed-off-by: Charlie Truong <[email protected]>

FP8 params support for megatron-fsdp (MXFP8/Blockwise) (#2239)

8f2f700

Signed-off-by: kunlunl <[email protected]> Signed-off-by: jianbinc <[email protected]> Co-authored-by: jianbinc <[email protected]> Co-authored-by: Cory Ye <[email protected]>

docs: fix broken images, links, and typos across documentation (#2794)

ed29157

Co-authored-by: Xin Yao <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fetch from nvidia Megatron-LM #5

Fetch from nvidia Megatron-LM #5

Uh oh!

RaymondLi0 commented Aug 3, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Fetch from nvidia Megatron-LM #5

Are you sure you want to change the base?

Fetch from nvidia Megatron-LM #5

Uh oh!

Conversation

RaymondLi0 commented Aug 3, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants