Search results

[dev] Broken GB200 nightly tests

bugSomething isn't workingSomething isn't working

Status: Open.

#4211 In NVIDIA/Megatron-LM;

· ko3n1g opened on Apr 8, 2026

Bug of save_checkpoint with ckpt_format is torch.

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#4200 In NVIDIA/Megatron-LM;

· LiJunscs opened on Apr 8, 2026

GDN packed sequence (THD format) + CP + SP causes shape mismatch at residual connection.

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#4194 In NVIDIA/Megatron-LM;

· yxs opened on Apr 8, 2026

Prefix-Cache Routing and Coordinator Robustness: Active-Rank-Aware Scheduling, Bounded Metadata and Protocol Hardening

community-request

enhancementNew feature or requestNew feature or request

Status: Open.

#4176 In NVIDIA/Megatron-LM;

· DhineshPonnarasan opened on Apr 7, 2026

Support for MXFP8 on 12.0+ architectures

community-request

enhancementNew feature or requestNew feature or request

Status: Open.

#4175 In NVIDIA/Megatron-LM;

· yuqie opened on Apr 7, 2026

Feature Request: FlashOptim - optimizer memory reduction

enhancementNew feature or requestNew feature or request

Status: Open.

#4171 In NVIDIA/Megatron-LM;

· sbhavani opened on Apr 6, 2026

Feature Request: DenseMixer - dense forward pass for MoE router gradient estimation during RL

enhancementNew feature or requestNew feature or request

module: rl

Status: Open.

#4169 In NVIDIA/Megatron-LM;

· sbhavani opened on Apr 6, 2026

Feature Request: Router Replay (R3) for stable RL with MoE models

enhancementNew feature or requestNew feature or request

module: rl

Status: Open.

#4168 In NVIDIA/Megatron-LM;

· sbhavani opened on Apr 6, 2026

Feature Request: ScatterMoE (Triton-based Sparse MoE with fused scatter/gather GEMMs)

enhancementNew feature or requestNew feature or request

Status: Open.

#4167 In NVIDIA/Megatron-LM;

· sbhavani opened on Apr 6, 2026

Add functional tests for NVFP4 precision

enhancementNew feature or requestNew feature or request

Status: Open.

#4161 In NVIDIA/Megatron-LM;

· WanZzzzzz opened on Apr 6, 2026

Improve prefix-cache coordinator routing for engine churn and long-running performance

community-request

enhancementNew feature or requestNew feature or request

Status: Open.

#4145 In NVIDIA/Megatron-LM;

· DhineshPonnarasan opened on Apr 4, 2026

Fix output path argument bug in build_sequences_per_dataset script

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#4143 In NVIDIA/Megatron-LM;

· DhineshPonnarasan opened on Apr 4, 2026