Expand description
Model trait and configuration system Core Model Abstraction
This module provides the foundational model abstraction that all model implementations build upon. It defines clean interfaces that hide implementation complexity.
Structs§
- Generation
Config - Generation configuration
- Memory
Requirements - Memory requirements for a model
- MoEConfig
- Configuration for MoE layer
- MoEExpert
- Helper struct for a standard MoE expert (simple MLP)
- MoELayer
- Mixture of Experts layer Routes tokens to top-k experts and combines their outputs
- Model
Registry - Model registry for managing all supported architectures
- Model
Weights - Model weights container
- Weight
Metadata - Weight metadata
Enums§
- Model
Inputs - Unified model inputs - supports all model types
- Model
Outputs - Unified model outputs - supports all model types
- Router
Type - Router types for MoE
- Weight
Format - Supported weight formats
Traits§
- MLPLayer
- MLP trait for expert networks in MoE layers
- Model
- Core model trait - the single interface all models implement
- Model
Config - Model configuration trait
- Model
Factory - Model factory trait for creating models by name
Functions§
- registry
- Get global model registry