Expand description
Portable execution plans, capabilities, and telemetry. Portable execution-plan and telemetry schemas.
Structs§
- Backend
Id - Stable, extensible identity of an execution backend.
- Device
Plan - Backend and device selected for one complete model session.
- Execution
Plan - A concrete, backend-neutral set of model/session execution choices.
- Expert
Cache Plan - Optional independent routed-expert cache selected by a plan.
Enums§
- Draft
Placement Plan - External assistant placement selected by an execution plan.
- Drafting
Plan - Speculative decoding selected by an execution plan.
- Execution
Plan Error - Execution plan validation error.
- Residency
Plan - Static weight placement selected by an execution plan.
- Weight
Transformation Plan - Optional load-time transformation applied to checkpoint weights.
Constants§
- DEFAULT_
MAX_ CACHED_ SHARDS - Default bound for simultaneously open checkpoint payload sources.
- EXECUTION_
PLAN_ SCHEMA_ VERSION - Schema version shared by execution-plan documents.