Skip to content

GRPO MCORE Path: Improve MoE Training MFU #985

@guyueh1

Description

@guyueh1

Opening this issue to compare and align GRPO mcore-training MFU in RL with pre-training MFU in Megatron-LM, for MoE models: DeepSeek V3, Qwen 30B, Qwen 235 B.

Metadata

Metadata

Assignees

Labels

PerformanceRelated to improving performancedeepseekRelated to deepseek 671bqa_rcca_donewhen RCCA finished for the issue, the qa will mark with this label .t-mcore

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions