Skip to content

Run the optional matched MTP tuning sweep for OrcaRouter Qwen 3.8 #2

Description

@francip

Imported from PLAN.md for initial project tracking.

  • Compare matched MTP depths and disabled mode
  • Record throughput, quality, stability, and memory results
  • Keep or change the production depth based on evidence

Keep the source document current when this work is completed or changed.

Metadata

Metadata

Assignees

No one assigned

    Labels

    backlogPlanned work not yet scheduledenhancementNew feature or requestmodels.serverWork for models.server

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions