Skip to content

[FEAT][distributed]: implement overlapping rollout and training pipeline #18

Closed

Description

@Flink-ddd

Construct an advanced asynchronous execution pipeline to overlap computing cycles. Allow vLLM rollout workers to sample data for iteration N+1 concurrently while DeepSpeed training workers process gradients for iteration N.

Activity

  1. added
    enhancementNew feature or request
    component: distributedTasks involving Ray actor management, cross-node scheduling, and communication synchronization.
    type: performancePerformance optimization tasks aimed at increasing throughput and reducing latency etc.
    priority: highSevere congestion issues require the highest priority for resolution.
    on May 31, 2026
  2. inaniloquentee commented on Jun 2, 2026

    @inaniloquentee
    Collaborator
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

component: distributedTasks involving Ray actor management, cross-node scheduling, and communication synchronization.enhancementNew feature or requestpriority: highSevere congestion issues require the highest priority for resolution.type: performancePerformance optimization tasks aimed at increasing throughput and reducing latency etc.

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions