Search results

Feature Request: Dual Chunk Attention for Long Context

enhancementNew feature or requestNew feature or request

Status: Open.

#2797 In NVIDIA/Megatron-LM;

· sbhavani opened on Jan 2, 2026

Feature Request: Video Generation Model Support (Wan, DiT architectures)

enhancementNew feature or requestNew feature or request

module: multimodal

Status: Open.

#2796 In NVIDIA/Megatron-LM;

· sbhavani opened on Jan 2, 2026

Feature Request: Enhanced Cross-Datacenter Training User Experience

enhancementNew feature or requestNew feature or request

Status: Open.

#2795 In NVIDIA/Megatron-LM;

· sbhavani opened on Jan 2, 2026

[BUG] model_param_group_index_map becomes stale due to parameter reordering in DistributedOptimizer

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#2777 In NVIDIA/Megatron-LM;

· HelloWorld686 opened on Dec 30, 2025

[BUG] Save checkpoint bug when using --optimizer-cpu-offload and --use-precision-aware-optimizer

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#2775 In NVIDIA/Megatron-LM;

· wplf opened on Dec 30, 2025

[QUESTION] inprocess restart

community-request

questionFurther information is requestedFurther information is requested

Status: Open.

#2774 In NVIDIA/Megatron-LM;

· tendar opened on Dec 30, 2025

Using Megatron-FSDP + EP consumes more GPU memory than using EP alone

bugSomething isn't workingSomething isn't working

community-request

module: megatron-fsdp

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#2772 In NVIDIA/Megatron-LM;

· cailun01 opened on Dec 29, 2025

<code>--use-te-activation-func</code> Flag Ignored for Non-MoE GPT Models

bugSomething isn't workingSomething isn't working

community-request

module: transformer engine

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#2770 In NVIDIA/Megatron-LM;

· vineash opened on Dec 28, 2025

Correctness Issue with Torch FSDP2

bugSomething isn't workingSomething isn't working

community-request

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#2769 In NVIDIA/Megatron-LM;

· haok1402 opened on Dec 28, 2025

[QUESTION] MoE score for aux loss

community-request

module: moe

needs-follow-upIssue needs follow-upIssue needs follow-up

questionFurther information is requestedFurther information is requested

Status: Open.

#2767 In NVIDIA/Megatron-LM;

· ysj1173886760 opened on Dec 27, 2025

[QUESTION] Low TFLOPS on Qwen3-30B-A3B (120 vs 277) &amp; OOM issues compared to NeMo AutoModel benchmarks

community-request

module: moe

questionFurther information is requestedFurther information is requested

Status: Open.

#2762 In NVIDIA/Megatron-LM;

· Bettinum opened on Dec 26, 2025

Checkpointing is not compatible with .grad(), please use .backward() if possible

bugSomething isn't workingSomething isn't working

community-request

module: moe

Status: Open.

#2758 In NVIDIA/Megatron-LM;

· chengmengli06 opened on Dec 25, 2025