Support packed sequences in Gated Delta Net by yapdianang · Pull Request #85 · radixark/Megatron-LM

yapdianang · 2026-08-17T13:48:15Z

Summary

port [main] feat(moe): Support packed sequence for gated delta net (GDN) NVIDIA/Megatron-LM#2645 onto miles-main
propagate packed-sequence metadata through GPT and MTP layers
reset causal convolution and gated-delta state at packed sequence boundaries
preserve Miles-specific MTP behavior while adding padding-mask propagation

Motivation

Qwen3.8-27B varlen training currently reaches the explicit GDN does not support packed sequence for now guard. Dense Qwen jobs do not exercise this path, which is why the same B300 stack worked for those models.

Testing

ruff check on all changed files
Black 24.4.2 check on all changed files
Python compile check on all changed files
focused pytest collection attempted; this development host does not have Transformer Engine installed, so GPU test collection stops at that dependency
B300 end-to-end validation pending a Miles image built from this commit

Upstream implementation: NVIDIA#2645

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

…rk#4) Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Yueming Yuan <yym022502@gmail.com>

…adixark#5) Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

- Detach output layer params to prevent MTP gradient flowing to output layer - Add mtp_kwargs interface for flexible MTP label/loss_mask passing - Roll mtp_labels and loss_mask for RL training compatibility Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Yueming Yuan <yym022502@gmail.com>

…yers (radixark#10) - Add is_mtp flag to MoE layers and multi_token_prediction module - Bypass routing replay for MTP layers (MTP uses fresh routing) - Replace rdxa/dev's built-in RouterReplay with miles.utils.routing_replay: - moe_utils.py: use get_routing_replay_compute_topk() wrapper - router.py: use register_routing_replay() for initialization Co-authored-by: Yueming Yuan <yym022502@gmail.com>

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Yueming Yuan <yym022502@gmail.com>

After bumping Megatron (rdxa/dev), colocated IPC weight update fails with torch.AcceleratorError: CUDA error: invalid argument during torch.multiprocessing serialization of CUDA tensors. Root cause: Megatron's new TMS hook (PR NVIDIA#3048) alters allocator behavior in training flow, causing allocations via cuMemCreate/cuMemMap which are incompatible with CUDA IPC (_share_cuda_() fails). Fix: resolve mapping.py and dynamic_context.py conflicts to isolate hook side effects so TMS/allocator state remains IPC-compatible during the weight update phase. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- merge(): truncate dp_reshardable padding on optimizer/param_state path - load_parameter_state_from_dp_reshardable: tolerate missing 'padding' key - ShardedTensor: relax flattened_range to deprecation warning Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

This PR rebases from radixark Megatron fork [miles-20260218](https://github.com/radixark/Megatron-LM/tree/miles-20260218) and resolve conflicts. Upgrade Megatron from Dec 17 (3714d81) to Feb 13 (1dcf0da) PR link: radixark#13 Co-authored-by: Yueming Yuan <yym022502@gmail.com> Made-with: Cursor

…adixark#17)

…se `--disable-weight-backuper` in miles (radixark#18) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

…adixark#20) Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

)

Squash merge of the dense true-on-policy Megatron branch. Co-authored-by: zju-stu-lizheng <lizheng.cs@zju.edu.cn> Co-authored-by: zyxiyy02 <282300612+zyxiyy02@users.noreply.github.com> Co-authored-by: Yi Zhang <1109276519@qq.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…lugins (radixark#50)

Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

…xark#58) Co-authored-by: Zhiyao Jiang <jessicajiang324@gmail.com>

…FP8) (radixark#53)

…xark#60) Co-authored-by: zyzshishui <82826991+zyzshishui@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…rk#63) Co-authored-by: Zhichenzzz <zczeng@uw.edu>

…adixark#68)

Co-authored-by: yueming-yuan <yym022502@gmail.com>

…e host memory instead of pinned memory to avoid SegFault on ROCm (radixark#74)

)

…VIDIA#2645) Signed-off-by: yuzhongw <yuzhongw@nvidia.com> Co-authored-by: kunlunl <kunlunl@nvidia.com> Co-authored-by: Xuesong Ye <xuesongyey@gmail.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com> Co-authored-by: Daisy Gao <daisyg@nvidia.com>

yueming-yuan and others added 30 commits February 25, 2026 19:22

[1/8] fix: misc compatibility fixes for PyTorch and TE (radixark#2)

b4d401c

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

[2/8] feat: support partial checkpoint loading (radixark#3)

6c3fb8c

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

[3/8] feat: add post-attention and post-MLP layernorm support (radixa…

3f0fcac

…rk#4) Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Yueming Yuan <yym022502@gmail.com>

[4/8] fix: MLA RoPE triton kernel head indexing and v_dim=0 support (r…

e82d9ee

…adixark#5) Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

[7/8] feat: add INT4 fake QAT for MoE grouped linear (radixark#9)

a95ad14

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Yueming Yuan <yym022502@gmail.com>

enable tms no cpu backup region for grad/optimizer's param buffers (r…

74f0b38

…adixark#17)

add disable_param_buffers_cpu_backup method to keep cpu backup when u…

5243579

…se `--disable-weight-backuper` in miles (radixark#18) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

[fix] fix mixed precision parameters load from checkpoint (radixark#19)

cf014f9

Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

[feat] distributed optimizer low GPU memory resume from checkpoint (r…

90f69fa

…adixark#20) Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

[fix] Add detach() to fp32 param shard for leaf-tensor consistency

32dbe9f

[fix] attention_output_gate TP slice when num_kv_heads < TP (radixark#22

923b85b

)

[fix] Enforce param dtype before wrap ddp (radixark#24)

d9a5080

Package miles_megatron_plugins with Megatron Core (radixark#49)

3a60706

Broaden megatron-core packages.find to megatron* (namespaces), keep p…

a381c7d

…lugins (radixark#50)

DeepSeek V4 RL support (radixark#28)

ae20f41

Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

Skip missing expert bias updates (radixark#55)

36ceb4f

[AMD] Fix QuantizedTensor import for ROCm TransformerEngine 2.8 (radi…

3a8f0c7

…xark#58) Co-authored-by: Zhiyao Jiang <jessicajiang324@gmail.com>

fix(dist-ckpt): handle bare BytesIO _extra_state in key recovery (TE …

87d1155

…FP8) (radixark#53)

Restore enforce_marked_param_dtypes after model materialization (radi…

79fc089

…xark#60) Co-authored-by: zyzshishui <82826991+zyzshishui@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Wire witness support inside Megatron (radixark#21)

4716f75

feat(optimizer): NVMe streaming of DistributedOptimizer state (radixa…

9fc14d8

…rk#63) Co-authored-by: Zhichenzzz <zczeng@uw.edu>

[tml] fp32 MoE activation and combine for bf16 inference alignment (r…

50ac48e

…adixark#68)

Megatron: Nemotron RL support (radixark#51)

8f83233

Co-authored-by: yueming-yuan <yym022502@gmail.com>

Restore gpt_model imports dropped in the radixark#51 merge (radixark#78)

8d33e68

sreerohi and others added 3 commits August 5, 2026 23:33

RoCM distributed checkpointing: stage GPU checkpoints through pageabl…

c0b2836

…e host memory instead of pinned memory to avoid SegFault on ROCm (radixark#74)

Support inline base64 payloads for te-precision-config-file (radixark#72

37f9161

)

yueming-yuan force-pushed the miles-main branch from 2e64d78 to 235952d Compare August 24, 2026 19:59

Navigation Menu

Uh oh!

Support packed sequences in Gated Delta Net - #85

Support packed sequences in Gated Delta Net#85
yapdianang wants to merge 33 commits into
radixark:miles-mainfrom
yapdianang:dydx/gdn-packed-sequences

yapdianang commented Aug 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

Uh oh!

Conversation

yapdianang commented Aug 17, 2026

Summary

Motivation

Testing

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants