perf: Add DeepEP support to Megatron Policy - #1645

Closed

parthmannan wants to merge 55 commits into

NVIDIA-NeMo:mainfrom

parthmannan:pmannan/deepep_mcore_training

Commits on Jan 15, 2026

Add DeepEP support to Megatron Policy
parthmannan
committed
b82a643
Add example recipe and add to config dict
parthmannan
committed
0488cee
docs: Revise news section for nemotron v3 and DAPO algorithm support (#1640)

snowmanwwg
authored and
parthmannan
committed
32a6794
chore: fix grpo functional test metric (#1643)

RayenTian
authored and
parthmannan
committed
1143d7a
feat: add support from building images using vllm from private repos (#1605)

authored and
parthmannan
committed
d148d8c
feat: Necessary changes for Gym GRPO tutorial (#1630)

authored and
parthmannan
committed
76ddbe1
perf: Add qwen3 30b-a3b async-8-off recipe (#1642)

youngeunkwon0405
authored and
parthmannan
committed
527f37e
feat: Add GPT-OSS support via mcore (#1452)

ashors1
authored and
parthmannan
committed
285a329
chore: Bump vllm to 0.11.2, torch to 2.9, transformers to 4.57.1 (#1563)

authored and
parthmannan
committed
4f2e164
fix: Support datasets saved with save_to_disk in ResponseDataset (#1610)

authored and
parthmannan
committed
789bda4
Recipe update
parthmannan
committed
105a5cc
fix: Handle disabled validation in SFT training (#1611)

sahgerlad
authored and
parthmannan
committed
250f34e
fix: Fix crash when using cp in dtensor path (#1663)

yfw
authored and
parthmannan
committed
afadf3e
fix: Fix Fp8 sequence padding for PP>1 case (#1579)

authored and
parthmannan
committed
87c55e2
test: Perf recipe for v0.5 (#1667)

authored and
parthmannan
committed
56dba8e
fix: Fix fp8 after vllm v0.11.2 bump (#1660)

guyueh1
authored and
parthmannan
committed
b527f27
fix: Fix crash when using activation_checkpointing (#1676)

yfw
authored and
parthmannan
committed
21d75b2
feat: add dapo recipe and test (#1617)

ZhiyuLi-Nvidia
authored and
parthmannan
committed
f952f78
feat: DTensorPolicyV2 GPT-OSS SFT support (#1470)

authored and
parthmannan
committed
8c9ae9f
fix: grad norm calculation for dtensor v2 (#1693)

hemildesai
authored and
parthmannan
committed
d3442b4
feat: Add Nemotron‑3 Nano 30B A3B BF16 SFT nightly tests (FSDP2, +LoRA) (#1648)

RayenTian
authored and
parthmannan
committed
7fbc72e
feat: Support prefetching of specific envs (#1692)

hemildesai
authored and
parthmannan
committed
0419429
Upgrade DeepEP version to match
parthmannan
committed
9108663
Lint fix
parthmannan
committed
78917e0
fix: Fix DTensor slice crash after PyTorch 2.9 bump (#1689)

zpqiu
authored and
parthmannan
committed
78d182c
fix: grad norm check for automodel gpt oss nightly (#1708)

hemildesai
authored and
parthmannan
committed
71a7fa8
fix: relax nanov3 nightly test metrics strict (#1712)

RayenTian
authored and
parthmannan
committed
8c492ff
fix: on GB200 use single-thread checkpoint save to avoid Cpu OOM (#1703)

guyueh1
authored and
parthmannan
committed
4051123
perf: [Perf recipe] Change TP 16->32 for deepseek GB200 sync benchmark (#1715)

guyueh1
authored and
parthmannan
committed
3e4bdcf
docs: Add doc for nano-v3 (#1694)

yfw
authored and
parthmannan
committed
a2580e2
fix: Disable cudnn sdpa backend when using activation checkpointing (#1717)

yfw
authored and
parthmannan
committed
90e14ee
fix: log metrics that can be coerced to scalars (#1723)

terrykong
authored and
parthmannan
committed
121dcf1
fix: use median instead of mean for logprob error for stability in nightlies (#1722)

terrykong
authored and
parthmannan
committed
f63f268
fix: gemma3 27b must now have skip_tokenizer_init=False in vllm (#1721)

terrykong
authored and
parthmannan
committed
6e4c7d3
fix: fix several nightly tests that were flaky (#1724)

terrykong
authored and
parthmannan
committed
d3e219b
fix: apply offloading change from v2 to v1 (#1726)

authored and
parthmannan
committed
0032a2d
fix: mcore generation config restored in nightly test (#1720)

terrykong
authored and
parthmannan
committed
c3ca1d4
feat: Megatron SFT LoRA (#1629)

authored and
parthmannan
committed
512bef9
build: Update aiohttp and urlib3 (#1746)

chtruong814
authored and
parthmannan
committed
d4683ea
fix: patch pytorch aten.alias.default shard strategy (#1728)

RayenTian
authored and
parthmannan
committed
d443965
feat: RL support for custom moe models in dtensor v2 (#1695)

hemildesai
authored and
parthmannan
committed
ff48f85
fix: split dtensorv1 vllm dependency (#1638)

yuki-97
authored and
parthmannan
committed
9b5da01
build: Resolve CVEs for gnupg and aiohttp (#1755)

chtruong814
authored and
parthmannan
committed
b66ec93
build: Bump mamba to d68d16e and causal-conv1d to 67e0a9d (#1759)

chtruong814
authored and
parthmannan
committed
35c7df9
Update uv
parthmannan
committed
6e6c476
ci: Clean up disk space for lint check (#1768)

chtruong814
authored and
parthmannan
committed
0b5d72f
docs: Adding dtensor TP debugging summary (#1767)

authored and
parthmannan
committed
454c304
Fix lock conflict resolution after signoff
parthmannan
committed
1ba733d
Merge branch 'main' of https://github.com/NVIDIA-NeMo/RL into pmannan/deepep_mcore_training
parthmannan
committed
c66bc7a
Merge branch 'main' of https://github.com/NVIDIA-NeMo/RL into pmannan/deepep_mcore_training
parthmannan
committed
a8351c7

Commits on Jan 16, 2026

Make DeepEP related args in all configs
parthmannan
committed
0ff0879
Add deepep args to policy tests
parthmannan
committed
6b34413
Add keys to vllm tests
parthmannan
committed
eb596a4

Commits on Jan 19, 2026