Skip to content

[FEA] Optimize online request scheduling for HSTU inference #461

Description

@shijieliu

Is your feature request related to a problem? Please describe.
Online HSTU requests have variable history length, candidate count, cache residency, and compute cost. FIFO or static batching can combine poorly matched requests, increase padding and cache movement, and create high tail latency. The repository needs an HSTU-aware scheduling policy evaluated against a realistic online benchmark.

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context, code examples, or references to existing implementations about the feature request here.


By submitting this issue, you agree to follow our code of conduct and our contributing guidelines.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions