Search results

🧠 [Research] Fundamental Equation of Consciousness: Ψ = argmax H(p) s.t. Φ > Φ_min

community-request

Status: Open.

#4049 In NVIDIA/Megatron-LM;

· dancinlife opened on Mar 29, 2026

Update megatron-core[dev] to support megatron-energon 7.x

bugSomething isn't workingSomething isn't working

community-request

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

Status: Open.

#4045 In NVIDIA/Megatron-LM;

· psycherun-kaiko opened on Mar 27, 2026

[Question] Enable GDN Packed Sequence Support for Context Parallelism in Qwen 3.5 series

community-request

questionFurther information is requestedFurther information is requested

Status: Open.

#4043 In NVIDIA/Megatron-LM;

· plmsmile opened on Mar 27, 2026

[Bug] HybridDeviceOptimizer KeyError with MoE models when using CPU offload

bugSomething isn't workingSomething isn't working

community-request

module: moe

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

Status: Open.

#4042 In NVIDIA/Megatron-LM;

· khazic opened on Mar 27, 2026

Improve docstrings and inline comments in example scripts

community-request

enhancementNew feature or requestNew feature or request

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

Status: Open.

#4040 In NVIDIA/Megatron-LM;

· DhineshPonnarasan opened on Mar 27, 2026

[QUESTION] Does Megatron-LM support torch 2.10.0

community-request

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

questionFurther information is requestedFurther information is requested

Status: Open.

#4034 In NVIDIA/Megatron-LM;

· longnguyen1802 opened on Mar 26, 2026

gpt3_mcore_te_tp2_zp_z3_resume_torch_dist_te_8experts2parallel_top2router broken after TE2.14

bugSomething isn't workingSomething isn't working

Status: Open.

#4033 In NVIDIA/Megatron-LM;

· ko3n1g opened on Mar 26, 2026

Feature Request: Attention Residuals

enhancementNew feature or requestNew feature or request

Status: Open.

#4016 In NVIDIA/Megatron-LM;

· sbhavani opened on Mar 24, 2026

[QUESTION] How to use Megatron-FSDP with mixed precision?

community-request

module: megatron-fsdp

questionFurther information is requestedFurther information is requested

Status: Open.

#4013 In NVIDIA/Megatron-LM;

· pavelgein opened on Mar 24, 2026

[ROADMAP] Megatron Core Roadmap

call for contribution

Status: Open.

#4003 In NVIDIA/Megatron-LM;

· sbhavani opened on Mar 23, 2026

[Bug] HybridEP dispatcher passes incorrect max_num_of_tokens_per_rank to DeepEP, causing RDMA QP assertion failure

bugSomething isn't workingSomething isn't working

community-request

module: moe

Status: Open.

#3999 In NVIDIA/Megatron-LM;

· tingmingzhong opened on Mar 23, 2026

[ENHANCEMENT] Add a minimal "quickstart" example for single-GPU / CPU local runs

community-request

enhancementNew feature or requestNew feature or request

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

Status: Open.

#3994 In NVIDIA/Megatron-LM;

· nathon-lee opened on Mar 23, 2026