From NVIDIA Megatron-LM for visibility #18

RaymondLi0 · 2023-01-24T20:01:13Z

No description provided.

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

Co-authored-by: Robert Kirby <rkirby@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Co-authored-by: oliver könig <okoenig@nvidia.com>

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

Co-authored-by: Siddharth Singh <136645615+sidsingh-nvidia@users.noreply.github.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Co-authored-by: oliver könig <okoenig@nvidia.com>

This reverts commit a652e2c.

Signed-off-by: Charlie Truong <chtruong@nvidia.com>

Co-authored-by: Mcore Bot <mcore-bot@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

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

This reverts commit 7487c53.

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

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

Co-authored-by: Mcore Bot <mcore-bot@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Signed-off-by: Keith Wyss <kwyss@nvidia.com> Co-authored-by: Cyril Meurillion <cmeurillon@nvidia.com>

… to a multiple of TP (#2574) Co-authored-by: Jon Barker <19699370+jon-barker@users.noreply.github.com>

This reverts commit bcf07a2.

…r sequence parallelism (#1997)

Signed-off-by: Li Tao <lit@nvidia.com> Co-authored-by: Charlie Truong <chtruong@nvidia.com>

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>

…-sharding (#2565) Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com> Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>

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

…#1755) Co-authored-by: Li Ruixiao <cgruixiao@outlook.com> Co-authored-by: oliver könig <okoenig@nvidia.com> Co-authored-by: Eric Harper <eharper@nvidia.com>

Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>

This reverts commit 32ebde7.

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

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com> Co-authored-by: Venmugil Elango <velango@nvidia.com> Co-authored-by: Venmugil Elango <498703+venmugil@users.noreply.github.com> Co-authored-by: Eric Harper <eharper@nvidia.com>

Signed-off-by: Pablo Garay <pagaray@nvidia.com>

Co-authored-by: drakezhang <drakezhang@tencent.com>

Co-authored-by: Xin Yao <xiny@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

… (#2571) Co-authored-by: Xin Yao <xiny@nvidia.com>

Co-authored-by: Xin Yao <xiny@nvidia.com>

Signed-off-by: gkollu <gkollu@nvidia.com> Co-authored-by: Xin Yao <xiny@nvidia.com>

Co-authored-by: Xin Yao <xiny@nvidia.com>

Co-authored-by: root <root@gpu-h100-0435.cm.cluster> Co-authored-by: root <root@gpu-h100-0012.cm.cluster> Co-authored-by: root <root@gpu-h100-0426.cm.cluster> Co-authored-by: root <root@gpu-h100-0188.cm.cluster> Co-authored-by: root <root@gpu-h100-0013.cm.cluster> Co-authored-by: root <root@gpu-h100-0032.cm.cluster> Co-authored-by: root <root@gpu-h100-0240.cm.cluster> Co-authored-by: root <root@gpu-h100-0089.cm.cluster>

…2595) Signed-off-by: Pablo Garay <pagaray@nvidia.com>

This reverts commit ef12f16.

#2084) Co-authored-by: Xin Yao <xiny@nvidia.com>

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 and others added 28 commits October 30, 2025 20:23

ci(hotfix): Repair codeowners file

4c2e1c9

ci(hotfix): Set docs allowed to fail

a07e00b

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

Ko3n1g/ci/test iteration time (#2067)

f559059

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

ci(hotfix): Remove performance for ckpt-resume

818e072

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

Allow inference test throughput to vary by 10% (#2070)

f248fcb

ci(hotfix): Inference test pipeline

e715d2f

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

chore: Fix autoformatter (#2073)

aad8761

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

ci(hotfix): Remove iteration-time from t5

e3ae351

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

ci(hotfix): disable inference test

87cbe76

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

ci(hotfix): Disable inference test

d0d00b3

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

ci(hotfix): Bypass approvalbot in merge-queue (#2082)

88e3a8a

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

ci(hotfix): Enable merge-group for approval bot

53305bc

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

chore: Update local tooling (#2066)

7c16ca0

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

Add extra RL files (#2077)

dc7a0ca

Co-authored-by: Robert Kirby <rkirby@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Prevent summary jobs from running in forks (#2083)

5cfad7b

Co-authored-by: oliver könig <okoenig@nvidia.com>

ci: Fix test scope (#2091)

ba21b69

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

ci(hotfix): Remove publish workflows

7ca2890

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

Refactor the attention metadata into separate classes (#2001)

a652e2c

Co-authored-by: Siddharth Singh <136645615+sidsingh-nvidia@users.noreply.github.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Guard against incorrectly using MoE prefill graphs (#2030)

65cd27c

Co-authored-by: oliver könig <okoenig@nvidia.com>

Revert "Refactor the attention metadata into separate classes (#2001)"

d3f1af4

This reverts commit a652e2c.

Run mr-slim tests in lightweight-mode (#2106)

5671e3a

Signed-off-by: Charlie Truong <chtruong@nvidia.com>

Inference | Lazy compile UVM allocator. (#1977)

7487c53

Co-authored-by: Mcore Bot <mcore-bot@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

chore: Reenable trustees (#2108)

1307f87

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

Revert "Inference | Lazy compile UVM allocator. (#1977)"

282b74c

This reverts commit 7487c53.

ci(fix): Changeset of copyright checker (#2110)

2cab46f

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

Ko3n1g/chore/update release settings (#2097)

d4194b7

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

Remove unnecessary check on rotary_pos_cos (#2003)

5dee638

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

(Reverted) Inference | Lazy compile UVM allocator. (#2125)

aecce9e

Co-authored-by: Mcore Bot <mcore-bot@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

kwyss-nvidia and others added 30 commits December 5, 2025 20:37

Make check_large_grads non-fatal (#2307)

bcf07a2

Signed-off-by: Keith Wyss <kwyss@nvidia.com> Co-authored-by: Cyril Meurillion <cmeurillon@nvidia.com>

fix for sequence packing plus sequence parallel: padding the sequence…

416687f

… to a multiple of TP (#2574) Co-authored-by: Jon Barker <19699370+jon-barker@users.noreply.github.com>

Revert "Make check_large_grads non-fatal (#2307)"

32ebde7

This reverts commit bcf07a2.

Torch symmetric - new latency optimized NVLS communication kernels fo…

972d9b6

…r sequence parallelism (#1997)

[Main] Support MTP packed-seq in main branch (#2173)

8c4df6b

Signed-off-by: Li Tao <lit@nvidia.com> Co-authored-by: Charlie Truong <chtruong@nvidia.com>

Various quality-of-life improvements in training loop (#2580)

8a5f379

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>

Support TP greater than num_kv_heads by supporting QKV activation sub…

f7dfb99

…-sharding (#2565) Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com> Co-authored-by: Roger Waleffe <rwaleffe@nvidia.com>

Fix FA3 import (#2577)

dfc3913

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

Fix runaway Etpt in straggler detector by resetting FLOPs accumulator (…

5232820

…#1755) Co-authored-by: Li Ruixiao <cgruixiao@outlook.com> Co-authored-by: oliver könig <okoenig@nvidia.com> Co-authored-by: Eric Harper <eharper@nvidia.com>

Rename TensorRT Model Optimizer to Model Optimizer (#2373)

4cf809c

Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com>

Reapply "Make check_large_grads non-fatal (#2307)"

e2199af

This reverts commit 32ebde7.

Fix aux loss scale when CP is enabled. (#2237)

f6e0d42

Save memory using main_param for moe in param_l2_norm (#2249)

01aad93

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

Changes to support latent MoEs (#2296)

b51db3e

Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com> Co-authored-by: Venmugil Elango <velango@nvidia.com> Co-authored-by: Venmugil Elango <498703+venmugil@users.noreply.github.com> Co-authored-by: Eric Harper <eharper@nvidia.com>

update API compat check baseline to b51db3e (#2588)

03b6d31

Signed-off-by: Pablo Garay <pagaray@nvidia.com>

Fix invalid argument failing tests on main (#2589)

f4957d1

Add openmathinstruct config. (#2586)

2bc35c5

Move model configs to github. (#2587)

e2cf81c

fix: Assign tokenizer to Encoder.tokenizer in legacy mode (#2498)

8d18afd

Co-authored-by: drakezhang <drakezhang@tencent.com>

Delete redundant import in yaml_arguments.py (#2139)

c21bf6e

Co-authored-by: Xin Yao <xiny@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>

Fix world size mismatch causing distributed init deadlock (Issue #2458)…

dfb78dc

… (#2571) Co-authored-by: Xin Yao <xiny@nvidia.com>

Improve performance of request_metadata logic (#2378)

4a4f23a

Fix broken Table of Contents links in README.md (#1954)

7b11553

Co-authored-by: Xin Yao <xiny@nvidia.com>

Add minor log update (#2080)

9fba363

Signed-off-by: gkollu <gkollu@nvidia.com> Co-authored-by: Xin Yao <xiny@nvidia.com>

Fix link to NeMo performance summary documentation (#2190)

bd32927

Co-authored-by: Xin Yao <xiny@nvidia.com>

feat: API compat: ignore ParameterMovedBreakage for __init__ methods (#…

a2aafe3

…2595) Signed-off-by: Pablo Garay <pagaray@nvidia.com>

Revert "Prep for refit (#2590)"

5c54bb6

This reverts commit ef12f16.

Fix NameError in pretrain_retro.py (add import_module), remove unused… (

be4baad

#2084) Co-authored-by: Xin Yao <xiny@nvidia.com>

Use the latest Hybrid-EP (#2479)

d2b500f

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

From NVIDIA Megatron-LM for visibility #18

From NVIDIA Megatron-LM for visibility #18

Uh oh!

RaymondLi0 commented Jan 24, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

From NVIDIA Megatron-LM for visibility #18

Are you sure you want to change the base?

From NVIDIA Megatron-LM for visibility #18

Uh oh!

Conversation

RaymondLi0 commented Jan 24, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants