List view

[Dev] feat(moe): Adding context parallel support to eager attention implementation

enhancementNew feature or requestNew feature or request

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1859

#1859 In NVIDIA/Megatron-LM;

· nrailg opened on Oct 13, 2025

[DEV] speed and memory improvements for layerwise distopt

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1965

#1965 In NVIDIA/Megatron-LM;

· FDecaYed opened on Oct 27, 2025

Inference linear layer

Expert ReviewApply this label to indicate that your PR is ready for expert review.Apply this label to indicate that your PR is ready for expert review.

module: inference

Run tests

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1908

#1908 In NVIDIA/Megatron-LM;

· sidsingh-nvidia opened on Oct 23, 2025

[Dev] Adding context parallel support to eager attention implementation

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1991

#1991 In NVIDIA/Megatron-LM;

· yuzhongw-nvidia opened on Oct 28, 2025

Torch symmetric - new latency optimized NVLS communication kernels for sequence parallelism

module: inference

Run tests

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1997

#1997 In NVIDIA/Megatron-LM;

· sidsingh-nvidia opened on Oct 28, 2025

[4/4] Merge Megatron-RL into LM

Status: Draft (not ready).

NVIDIA/Megatron-LMnumber 2002

#2002 In NVIDIA/Megatron-LM;

· tdene opened on Oct 28, 2025

Async task and event loop safety in Megatron Core

Expert ReviewApply this label to indicate that your PR is ready for expert review.Apply this label to indicate that your PR is ready for expert review.

Run tests

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2025

#2025 In NVIDIA/Megatron-LM;

· tdene opened on Oct 29, 2025

Adds hybrid model support for dynamic inference engine

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 1907

#1907 In NVIDIA/Megatron-LM;

· santhnm2 opened on Oct 23, 2025

Fix FP8 inference with sequence parallelism

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2009

#2009 In NVIDIA/Megatron-LM;

· santhnm2 opened on Oct 28, 2025

Remove unnecessary check on rotary_pos_cos

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2003

#2003 In NVIDIA/Megatron-LM;

· santhnm2 opened on Oct 28, 2025

[DEV][NVFP4] Fix NVFP4 Selective Activation Recompute

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2036

#2036 In NVIDIA/Megatron-LM;

· zhongbozhu opened on Oct 29, 2025

Optimizer refactor

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2047

#2047 In NVIDIA/Megatron-LM;

· deepakn94 opened on Oct 30, 2025

Siddharth/torch symm fused

Status: Draft (not ready).

NVIDIA/Megatron-LMnumber 2048

#2048 In NVIDIA/Megatron-LM;

· sidsingh-nvidia opened on Oct 30, 2025

ci: POC: Gridify test cases

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2049

#2049 In NVIDIA/Megatron-LM;

· ko3n1g opened on Oct 30, 2025

Fix param init

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2033

#2033 In NVIDIA/Megatron-LM;

· cuichenx opened on Oct 29, 2025

Fixes for gpt-oss

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2038

#2038 In NVIDIA/Megatron-LM;

· cuichenx opened on Oct 29, 2025

Add flash_attn_3 as first option for FA3 import

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2010

#2010 In NVIDIA/Megatron-LM;

· santhnm2 opened on Oct 28, 2025

Enable TE custom quantization recipe

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2005

#2005 In NVIDIA/Megatron-LM;

· negvet opened on Oct 28, 2025

Hybrid Data x Context Parallelism Feature

enhancementNew feature or requestNew feature or request

Expert ReviewApply this label to indicate that your PR is ready for expert review.Apply this label to indicate that your PR is ready for expert review.

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2054

#2054 In NVIDIA/Megatron-LM;

· parthmannan opened on Oct 30, 2025

Make <code>PipelineParallelLayout</code> always return str from <code> __repr__</code>

Final ReviewApply this label to indicate that your PR is ready for final review.Apply this label to indicate that your PR is ready for final review.

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2055

#2055 In NVIDIA/Megatron-LM;

· ananthsub opened on Oct 30, 2025

Fix Qwen3-Next hang on Blackwell, add a flag to control torch.compile

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2058

#2058 In NVIDIA/Megatron-LM;

· yuzhongw-nvidia opened on Oct 31, 2025

[Dev] Support packed seq in MTP

core_dev_r0.15.0

module: moe

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2043

#2043 In NVIDIA/Megatron-LM;

· BestJuly opened on Oct 30, 2025

Move <code>data_samplers.py</code> from <code>legacy</code> to <code>core.datasets</code> &amp; add <code>DistributedSignalHandler</code> to DataLoader workers

Expert ReviewApply this label to indicate that your PR is ready for expert review.Apply this label to indicate that your PR is ready for expert review.

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2068

#2068 In NVIDIA/Megatron-LM;

· asolergi-nv opened on Oct 31, 2025

Inference throughput tests: refactor goldens to be in list format

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2072

#2072 In NVIDIA/Megatron-LM;

· mathemakitten opened on Oct 31, 2025

wandb Inference stats logging

Status: Open (in progress).

NVIDIA/Megatron-LMnumber 2026

#2026 In NVIDIA/Megatron-LM;

· wdykas opened on Oct 29, 2025