Mila
Deep Neural Network Library
Loading...
Searching...
No Matches
Dnn.Components.MixtureOfExperts Module Reference

Exported Modules

module  Compute.DeviceAllocation
module  Compute.Observation
module  Dnn.ActivationType
module  Dnn.TensorTypes
module  Dnn.Component
module  Dnn.TensorDataType
module  Dnn.TensorDataTypeTraits
module  Serialization.ModelArchive
module  Dnn.Quantization.Weight.Policies
module  Dnn.ComponentType
module  Compute.ExecutionContextFactory
module  Serialization.Tensor
module  Dnn.TensorOps
module  Compute.OperationTraits
module  Dnn.Tensor
module  Compute.DeviceId
module  Serialization.SafeTensors
module  Dnn.Components.MixtureOfExpertsConfig
module  Compute.IExecutionContext
module  Compute.DeviceTypeTraits
module  Dnn.Components.Activation
module  Serialization.Mode
module  Compute.ExecutionContext
module  Compute.DeviceType
module  Dnn.ITensor

Classes

class  Mila::Dnn::MixtureOfExperts< TDeviceType, TPrecision, TGate, TWeightQuantization >
 Stacked experts: gate_up_proj [E, 2I, H] and down_proj [E, H, I], checkpoint-named. More...

Files

file  Mila/Src/Dnn/Components/MixtureOfExperts/MixtureOfExperts.ixx
 Mixture-of-experts bank: the stacked experts and the weighted combine, routed from outside.