Search results

torch.compile is incompatible with tensor parallelism

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#2598 In NVIDIA/Megatron-LM;

· robotsp opened on Dec 9, 2025

Qwen3-Next-80B-A3B with MTP NaN issue on Blackwell

bugSomething isn't workingSomething isn't working

module: moe

Status: Open.

#2597 In NVIDIA/Megatron-LM;

· yuzhongw-nvidia opened on Dec 9, 2025

[QUESTION] How to perform the RoPE in LLama3 attention?

community-request

Status: Open.

#2596 In NVIDIA/Megatron-LM;

· xianyi11 opened on Dec 9, 2025

What are the recommended arguments for using Muon

community-request

questionFurther information is requestedFurther information is requested

Status: Open.

#2583 In NVIDIA/Megatron-LM;

· Cccei000 opened on Dec 8, 2025

fp8-param-gather for mxfp8

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#2582 In NVIDIA/Megatron-LM;

· bhaktatejas922 opened on Dec 7, 2025

softmax_type != vanilla support thd format

community-request

enhancementNew feature or requestNew feature or request

Status: Open.

#2566 In NVIDIA/Megatron-LM;

· nicosouth1 opened on Dec 5, 2025

Cannot install megatron-core[dev] - nv-grouped-gemm build error

bugSomething isn't workingSomething isn't working

Status: Open.

#2541 In NVIDIA/Megatron-LM;

· sbhavani opened on Dec 4, 2025

Improve Megatron-RL testing

enhancementNew feature or requestNew feature or request

module: rl

Status: Open.

Enhancement

#2460 In NVIDIA/Megatron-LM;

· jon-barker opened on Dec 2, 2025

TP=1 embedding path assumes in-range token IDs while MockGPTDataset ignores config

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#2459 In NVIDIA/Megatron-LM;

· vineash opened on Dec 2, 2025

Feature Request: Kimi Delta Attention (KDA)

enhancementNew feature or requestNew feature or request

Status: Open.

#2446 In NVIDIA/Megatron-LM;

· sbhavani opened on Dec 1, 2025

[QUESTION] Why doesn't theoretical memory take into account the expert parallelism and the partitioning of experts?

community-request

questionFurther information is requestedFurther information is requested

Status: Open.

#2439 In NVIDIA/Megatron-LM;

· 1195343015 opened on Dec 1, 2025

Feature Request: adopt ant's AMem as an optional API to offload NCCL memory in RL scenario

enhancementNew feature or requestNew feature or request

Status: Open.

#2423 In NVIDIA/Megatron-LM;

· ISEEKYAN opened on Nov 28, 2025