Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
130 commits
Select commit Hold shift + click to select a range
a976754
[fix] Bug fix for offloading in evaluate() (#3043)
lhb8125 Jan 30, 2026
991c38f
Update copy-pr-bot.yaml [skip ci]
github-actions[bot] Jan 31, 2026
5d0a7fd
cp: `Fix: nccl-ub in ddp path (3181)` into `main` (#3182)
ko3n1g Jan 31, 2026
ffbc43f
Miscellaneous inference cleanup (#2955)
santhnm2 Jan 31, 2026
0fe3232
Revert "Miscellaneous inference cleanup (#2955)"
ko3n1g Jan 31, 2026
69a5c63
ci: Fix DSv3 (#3188)
ko3n1g Jan 31, 2026
2fadde8
Fix missing argument in MoELayer.forward() (#3133)
jiemingz Feb 1, 2026
ae67076
Fix H2D stream synchronization in optimizer offload (#3140)
tgkyrie Feb 1, 2026
300d1b6
Add MTP support for hybrid models (#2363)
rkarimimahab Feb 1, 2026
dceb1fb
docs: improve Megatron-LM and Megatron Core descriptions (#3115)
sbhavani Feb 2, 2026
f4502eb
Handle `step` key correctly in checkpoint save with `--optimizer-cpu-…
ahmadki Feb 2, 2026
70719cd
mRoPE for MTP (#3114)
BestJuly Feb 2, 2026
e836e62
Fix two minor bugs in MTP implementation for hybrid models (#3194)
deepakn94 Feb 2, 2026
1362e4a
Update README.md (#2111)
mvirts Feb 2, 2026
31d0c87
Revert "Fix two minor bugs in MTP implementation for hybrid models (#…
ko3n1g Feb 2, 2026
a0cc8ca
Revert "Add MTP support for hybrid models (#2363)"
ko3n1g Feb 2, 2026
50546da
Fix bug in SFTDataset (#3185)
duncanriach Feb 2, 2026
dff4189
Fix several syntax error (#3004)
HollowMan6 Feb 2, 2026
c4bea0a
Fix for RL Test (#3148)
wdykas Feb 3, 2026
a4008d0
Fix latent moe flops and backward_dw (#2977)
buptzyb Feb 3, 2026
afe443b
Use global user buffer when the bucket size does not fit FixedPoolAll…
shengf-nv Feb 3, 2026
78475fe
ci: Checkpoint retention (#3205)
ko3n1g Feb 3, 2026
7080697
Add unit test for LatentMoE (#2892)
venmugil Feb 3, 2026
0028273
ci: Enable unit tests on merge-queue (#3186)
ko3n1g Feb 3, 2026
94c9eae
Fix seq pack flag in `get_logprobs` (#3206)
mathemakitten Feb 3, 2026
b477d12
ci(fix): Parse unit tests in merge-queue (#3224)
ko3n1g Feb 3, 2026
1a61b77
Fix TE 2.12 AllGather CI failure (#3101)
BestJuly Feb 3, 2026
79e7bfe
ci(hotfix): Pin uv (#3233)
ko3n1g Feb 3, 2026
18d69f1
Add a unit test to check that RL `get_logprobs` will reuse training c…
mathemakitten Feb 3, 2026
27a5f83
Do not offload grad buffers when training graphs are enabled (#3231)
mathemakitten Feb 3, 2026
bc2eb9a
Fix missing PackedSeqParams import (#3214)
parthmannan Feb 3, 2026
1fdb29f
Synchronize the request counts for EP inference with strict matching …
santhnm2 Feb 3, 2026
e02344e
Do not let requests fail silently inside inference engine (#3228)
tdene Feb 3, 2026
4c48248
Update copy-pr-bot.yaml [skip ci]
github-actions[bot] Feb 4, 2026
9050d5b
Fix coordinator address collision check in flask (#3208)
tdene Feb 3, 2026
cd5ed74
torch saver inference model offload (#3170)
wdykas Feb 4, 2026
982ca5d
enable cuda graph ut (#3197)
Autumn1998 Feb 4, 2026
473e283
Support EP with HSDP (#2840)
wplf Feb 4, 2026
4a23972
[Main] Add the missing part to support 1F1B overlap for Qwen3-Next (#…
BestJuly Feb 4, 2026
c036e77
Missing import fix (#3241)
parthmannan Feb 4, 2026
43db8c1
Miscellaneous inference cleanup (Replay of !2955) (#3232)
santhnm2 Feb 4, 2026
adce147
Add DistributedInitConfig (#3173)
maanug-nv Feb 4, 2026
f3e6cc8
Fix checkpoint converter missing parallel group initialization (#3217)
yashaswikarnati Feb 4, 2026
d558b5f
Skip empty sequences and chunks in MTP tensor roll (#3035)
BestJuly Feb 4, 2026
f708b5d
Implement get_parameters for ChainedOptimizer (#3201)
nschank Feb 4, 2026
66c432a
ci(fix): Create main/dev image tags (#3252)
ko3n1g Feb 4, 2026
e24767f
ci(hotfix): Skopeo copy
ko3n1g Feb 4, 2026
d959620
ci(hotfix): Add skopeo
ko3n1g Feb 4, 2026
9d71cb1
Reapply "Add MTP support for hybrid models (#2363)" (#3207)
sancha Feb 4, 2026
b043863
Fix uv install for GH actions (#3259)
Phlip79 Feb 4, 2026
dd7d141
Update the project structure in README (#3251)
janEbert Feb 5, 2026
1f6d8c2
chore: rotate oncall schedule
github-actions[bot] Feb 5, 2026
1b11076
Cherry-pick: Fix mtp_num_layers and clip_qk issues (#2581, #2776) (#3…
BestJuly Feb 5, 2026
111a2a0
RL: training cudagraphs functional test (#3235)
mathemakitten Feb 5, 2026
1934391
[Main] fix cg missing wgrad hook (#3074)
Wohox Feb 5, 2026
801f12f
Avoid .cuda call on meta device in LanguageModel (#3202)
nschank Feb 5, 2026
347ad21
Nano QAT/D fix with sft tokenizer and datasets (#3254)
ChenhanYu Feb 5, 2026
3c0a4f3
Update copy-pr-bot.yaml [skip ci]
github-actions[bot] Feb 6, 2026
0434f87
fix checkpointing error message (#3203)
dimapihtar Feb 5, 2026
8379d43
Revert "fix checkpointing error message (#3203)" (#3283)
ko3n1g Feb 6, 2026
e2e5a6a
Reapply "fix checkpointing error message (#3203)" (#3283) (#3285)
ko3n1g Feb 6, 2026
a116ce3
docs: Add changelog for 0.15.3 (#3286)
ko3n1g Feb 6, 2026
4376cc5
ci: Set throughput tests as flaky (#3301)
chtruong814 Feb 6, 2026
f92460b
chore: Move GB200 tests to nightly (#3302)
ko3n1g Feb 6, 2026
cfbe9b5
Ensure type-checker understands use of Submodules in bert_model (#3256)
nschank Feb 6, 2026
a63d045
Override extra_repr instead of __repr__ (#3200)
nschank Feb 7, 2026
f68c7c1
Replace ModuleSpec with Protocols for LayerNorm submodules (#3090)
nschank Feb 7, 2026
2f99ee8
chore: Remove gpt_grpo_tp2tp1_pp4pp2_dp8_583m_throughputtest
ko3n1g Feb 7, 2026
e3ae6e4
Non colocated refit (#3213)
wdykas Feb 7, 2026
554ce49
Fuse permute+pad and unpermute+unpad ops for FP8/FP4 training (#2763)
xiaoxi-wangfj Feb 7, 2026
7cbbba2
Add check to prevent MFSDP from numeric issue in gradient accumulate …
shjwudp Feb 7, 2026
c99c962
update get_embedding_ranks and get_position_embedding_ranks docstring…
c1lovez1 Feb 7, 2026
6d81e3d
ci: Add secrets detector (#3180)
chtruong814 Feb 7, 2026
a3ec4b0
Param offset in _ParamAndGradBucket should be aligned (#3007)
skydoorkai Feb 7, 2026
916301a
updates to support modelopt EAGLE training with CP (#3147)
yeyu-nvidia Feb 9, 2026
6103cb5
Ensure type-checker understands use of Submodules in llava_model (#3257)
nschank Feb 9, 2026
4ff7686
M-FSDP: Remove redundant stream waits in HSDP to prevent CG fail (#2941)
shjwudp Feb 9, 2026
3257093
fully remove legacy tokenizer system (#2946)
dimapihtar Feb 9, 2026
3069591
General README and pyproject fixes (#2907)
ahmadki Feb 9, 2026
3bb539e
chore: More aggressive checkpointing (#3315)
ko3n1g Feb 9, 2026
c072f89
ci: Pin down setuptools to lt 82 (#3313)
ko3n1g Feb 9, 2026
9ddbce3
fix: T5 dataset (#3307)
ko3n1g Feb 9, 2026
f14d161
fix: numpy overflow (#3306)
ko3n1g Feb 9, 2026
8d79987
ci: Revert "ci: Add secrets detector (#3180)" (#3330)
chtruong814 Feb 10, 2026
55c3e63
ci: Add more tests, run on merge-queue (#3317)
ko3n1g Feb 10, 2026
ba76934
ci: Remove merge-gate environment check (#3331)
chtruong814 Feb 10, 2026
ab5e277
Use FP4 context for mamba (#2604)
kwyss-nvidia Feb 10, 2026
fc557ec
ci: Ensure we run all functional tests in merge group (#3332)
chtruong814 Feb 10, 2026
55198ba
Replace ModuleSpec with Protocols for inputs to MLP (#3084)
nschank Feb 10, 2026
5eb20b8
ci: Fix merge queue functional tests (#3337)
chtruong814 Feb 10, 2026
367f0b8
ci: skip queue in merge-gate (#3343)
ko3n1g Feb 10, 2026
3fb6006
ci: Timeout for functional tests (#3346)
ko3n1g Feb 10, 2026
76cf11e
update checkpointing documentation (#3347)
dimapihtar Feb 10, 2026
836d473
Update golden values to reflect improvements (#3350)
tdene Feb 10, 2026
2451508
BUGFIX: gpt vs hybrid model mtp naming mismatch (#3334)
sancha Feb 10, 2026
8da949e
Disable flaky test (#3354)
tdene Feb 10, 2026
bb97791
re-enable gpt grpo tests (#3348)
jon-barker Feb 10, 2026
4bce841
Fix SFT Pipeline when TP>1 (#3268)
asolergi-nv Feb 10, 2026
f5238ba
Fixes for KD mode (#3342)
AAnoosheh Feb 10, 2026
c1169ea
chore: rotate oncall schedule
github-actions[bot] Feb 11, 2026
4f025a1
chore: Update codeowners file (#3365)
ko3n1g Feb 11, 2026
66ec17e
Siddharth/fix inference functional tests (#3357)
sidsingh-nvidia Feb 11, 2026
6a9da99
Switch oncall (#3360)
janEbert Feb 11, 2026
b6e883b
Add missing RMSNorm to llama train script (#3314)
AAnoosheh Feb 11, 2026
cd14090
Fix inference for MTP models (#3297)
tdene Feb 11, 2026
6f5de16
Add a logprobs test with real gpt model. (#2870)
yobibyte Feb 11, 2026
b5d50cb
Add simple GRPO functional test (#3323)
tdene Feb 11, 2026
1c245c7
ci: Concurrency control for merge-queue (#3353)
ko3n1g Feb 11, 2026
d9f075c
ci: Update golden value download script to work with Github (#3335)
chtruong814 Feb 11, 2026
d0b768f
Removing etc from main index page, shifted name of discussions (#3271)
megnvidia Feb 11, 2026
7d1acf6
fix: correct typos 'seperated' and 'recieved' (#3305)
thecaptain789 Feb 11, 2026
2807a4e
Improved PyTorch profiler and added PyTorch execution trace (#3273)
shengf-nv Feb 11, 2026
1959739
build: Bump TE on 2.12 (#3371)
ko3n1g Feb 11, 2026
f06e669
ci(hotfix): job conditions (#3376)
ko3n1g Feb 11, 2026
6467cbc
Update copy-pr-bot.yaml [skip ci]
github-actions[bot] Feb 12, 2026
faced51
Record moe routing decisions during inference. (#3034)
sidsingh-nvidia Feb 11, 2026
dedb6dd
[Main] Fix EP Overlap Bugs for Full-Iter CG (#3164)
Wohox Feb 12, 2026
11a4659
Avoid direct pickle import (#3375)
maanug-nv Feb 12, 2026
fe9279e
Delete old pretrain_* files (#3359)
Phlip79 Feb 12, 2026
7df15bd
Add Qwen3-VL support with Megatron-FSDP (#2841)
xuwchen Feb 12, 2026
c65fb25
Refactor Mamba chunked prefill (#3265)
santhnm2 Feb 12, 2026
47938af
Improved parallel logging of learning rate (#3319)
jstjohn Feb 12, 2026
a51c1c8
Add enhanced event tracking with TTFT measurement and compact seriali…
lmcafee-nvidia Feb 12, 2026
dbc444f
Add assertion that max_requests is divisible by tp_size (#3304)
santhnm2 Feb 12, 2026
1123fc0
Move to using the Inference OpenAI API server (#3107)
ArEsKay3 Feb 12, 2026
4184bfa
Revert "Move to using the Inference OpenAI API server (#3107)"
ko3n1g Feb 13, 2026
9119fae
Update moe github test cases. (#3077)
Feb 13, 2026
e0aa16b
Revert "Update moe github test cases. (#3077)"
ko3n1g Feb 13, 2026
28ccdaa
Split layer_specs to return Submodules instead of ModuleSpecs (#3255)
nschank Feb 13, 2026
9d75ec4
Use copy_signature to preserve typing of pass-through methods
nschank Jan 30, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions .github/CODEOWNERS
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,8 @@ megatron/core/ssm/ @NVIDIA/core-adlr @NVIDIA/core-nemo @NVIDIA/hybrid-mamba

megatron/core/datasets/ @NVIDIA/core-adlr @NVIDIA/core-nemo @NVIDIA/datasets

megatron/core/tokenizers/ @NVIDIA/core-adlr @NVIDIA/core-nemo @NVIDIA/tokenizers

megatron/core/distributed/fsdp/ @NVIDIA/core-adlr @NVIDIA/core-nemo @NVIDIA/megatron-fsdp

megatron/core/transformer/fsdp_dtensor_checkpoint.py @NVIDIA/core-adlr @NVIDIA/core-nemo @NVIDIA/megatron-fsdp
Expand Down
23 changes: 17 additions & 6 deletions .github/actions/action.yml
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,9 @@ inputs:
is_ci_workload:
description: "Is CI workload"
required: true

is_merge_group:
description: "Is merge group"
required: true
runs:
using: "composite"
steps:
Expand Down Expand Up @@ -117,8 +119,10 @@ runs:
export PYTHONPATH=$(pwd)
export NEMORUN_HOME=$(pwd)
export NCCL_DEBUG=INFO
pip install --no-cache-dir uv
uv sync --only-group test
pip install --no-cache-dir "uv<0.9.29"
uv venv .venv
uv cache clean
uv sync --no-cache --only-group test
uv run python tests/test_utils/python_scripts/launch_nemo_run_workload.py \
--scope unit-tests \
--model unit-tests \
Expand Down Expand Up @@ -177,7 +181,12 @@ runs:
#!/bin/bash
set -euxo pipefail

if [ "${{ steps.has-run-tests-label.outputs.main }}" == "true" ]; then
if [ "${{ inputs.is_merge_group }}" == "true" ]; then
ARGS=(
--scope mr-github
--n-repeat 1
)
elif [ "${{ steps.has-run-tests-label.outputs.main }}" == "true" ]; then
ARGS=(
--scope mr-github
--enable-lightweight-mode
Expand All @@ -197,8 +206,10 @@ runs:

export PYTHONPATH=$(pwd)
export NEMORUN_HOME=$(pwd)
pip install --no-cache-dir uv
uv sync --only-group test
pip install --no-cache-dir "uv<0.9.29"
uv venv .venv
uv cache clean
uv sync --no-cache --only-group test
uv run python tests/test_utils/python_scripts/launch_nemo_run_workload.py \
${ARGS[@]} \
--model ${{ inputs.model }} \
Expand Down
2 changes: 1 addition & 1 deletion .github/copy-pr-bot.yaml
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
enabled: true
auto_sync_draft: false
auto_sync_ready: true
trustees_override: ["AAnoosheh", "ArEsKay3", "Autumn1998", "BestJuly", "BoxiangW", "ChenhanYu", "FDecaYed", "HaochenYuan", "ISEEKYAN", "JRD971000", "Phlip79", "QiZhangNV", "ShriyaRishab", "Victarry", "Wohox", "ZhiyuLi-Nvidia", "ahmadki", "aklife97", "ananthsub", "asolergi-nv", "buptzyb", "chtruong814", "cspades", "cuichenx", "deepakn94", "dimapihtar", "duncanriach", "erhoo82", "ericharper", "fanshiqing", "frsun-nvda", "gautham-kollu", "gdengk", "guyueh1", "hxbai", "jalbericiola", "janEbert", "jaredcasper", "jenchen13", "jiemingz", "jingqiny-99", "jkamalu", "jon-barker", "jstjohn", "kanz-nv", "kevalmorabia97", "ko3n1g", "kunlunl", "kvareddy", "kwyss-nvidia", "layalir", "lhb8125", "lmcafee-nvidia", "maanug-nv", "mathemakitten", "matthieule", "mehraakash", "mkhona-nvidia", "parthmannan", "prajwal1210", "pthombre", "rogerwaleffe", "sanandaraj5597", "sancha", "santhnm2", "sbak5", "shanmugamr1992", "shengf-nv", "shifangx", "shjwudp", "sidsingh-nvidia", "skyw", "sudhakarsingh27", "tdene", "theothermike", "thomasdhc", "trintamaki", "tylerpoon", "wdykas", "xiaoyao0115", "xuwchen", "yanring", "yaox12", "yaoyu-33", "yashaswikarnati", "yeyu-nvidia", "yobibyte", "youngeunkwon0405", "yuzhongw-nvidia", "zhongbozhu"]
trustees_override: ["AAnoosheh", "ArEsKay3", "Autumn1998", "BestJuly", "BoxiangW", "ChenhanYu", "FDecaYed", "HaochenYuan", "ISEEKYAN", "JRD971000", "Phlip79", "QiZhangNV", "RPrenger", "ShriyaRishab", "Victarry", "Wohox", "ZhiyuLi-Nvidia", "ahmadki", "aklife97", "ananthsub", "asolergi-nv", "buptzyb", "chtruong814", "cspades", "cuichenx", "deepakn94", "dimapihtar", "dingqingy-nv", "duncanriach", "erhoo82", "ericharper", "fanshiqing", "frsun-nvda", "gautham-kollu", "gdengk", "guyueh1", "hxbai", "ilml", "jalbericiola", "janEbert", "jaredcasper", "jenchen13", "jiemingz", "jingqiny-99", "jkamalu", "jon-barker", "jstjohn", "kanz-nv", "kevalmorabia97", "ko3n1g", "kunlunl", "kvareddy", "kwyss-nvidia", "layalir", "lhb8125", "lmcafee-nvidia", "maanug-nv", "mathemakitten", "matthieule", "mehraakash", "mkhona-nvidia", "parthmannan", "prajwal1210", "pthombre", "rogerwaleffe", "sanandaraj5597", "sancha", "santhnm2", "sbak5", "shanmugamr1992", "sharathts", "shengf-nv", "shifangx", "shjwudp", "sidsingh-nvidia", "skyw", "sudhakarsingh27", "tdene", "theothermike", "thomasdhc", "trintamaki", "tylerpoon", "wdykas", "xiaoyao0115", "xuwchen", "yanring", "yaox12", "yaoyu-33", "yashaswikarnati", "yeyu-nvidia", "yobibyte", "youngeunkwon0405", "yuzhongw-nvidia", "zhongbozhu"]
20 changes: 10 additions & 10 deletions .github/oncall_schedule.json
Original file line number Diff line number Diff line change
@@ -1,18 +1,10 @@
[
{
"user": "dimapihtar",
"date": "2026-01-28"
},
{
"user": "gautham-kollu",
"date": "2026-02-04"
},
{
"user": "janEbert",
"user": "Phlip79",
"date": "2026-02-11"
},
{
"user": "Phlip79",
"user": "janEbert",
"date": "2026-02-18"
},
{
Expand Down Expand Up @@ -46,5 +38,13 @@
{
"user": "BoxiangW",
"date": "2026-04-15"
},
{
"user": "Phlip79",
"date": "2026-04-22"
},
{
"user": "asolergi-nv",
"date": "2026-04-29"
}
]
276 changes: 0 additions & 276 deletions .github/workflows/check_api_backwards_compatibility_workflow.yml

This file was deleted.

Loading