Dnn.Components.GemmaBlock Module Reference

Exported Modules
module	Dnn.Components.Gqa
module	Dnn.Components.RmsNorm
module	Serialization.ModelArchive
module	Compute.ExecutionContext
module	Compute.DeviceAllocation
module	Compute.DeviceTypeTraits
module	Compute.IExecutionContext
module	Dnn.Quantization.Weight.Policies
module	Dnn.Component
module	Compute.CpuMemoryResource
module	Compute.DeviceId
module	Dnn.Components.GemmaConfig
module	Compute.Device
module	Dnn.Components.Linear
module	Compute.Observation
module	Dnn.Tensor
module	Dnn.CompositeComponent
module	Dnn.Quantization.KvCache.Policy
module	Dnn.Components.GatedMLP
module	Dnn.Components.Router
module	Dnn.Components.Residual
module	Dnn.Components.Rope
module	Compute.ExecutionContextFactory
module	Serialization.SafeTensors
module	Dnn.ComponentType
module	Serialization.Mode
module	Dnn.Components.Swiglu
module	Dnn.ITensor
module	Compute.DeviceType
module	Serialization.Tensor
module	Dnn.ActivationType
module	Dnn.Components.ITransformerBlock
module	Dnn.Components.MixtureOfExperts
module	Dnn.TensorOps
module	Dnn.TensorDataType
module	Dnn.TensorDataTypeTraits
module	Compute.GqaState
module	Dnn.TensorTypes

Classes
struct	Mila::Dnn::GemmaBlock< TDeviceType, TPrecision, kGlobal, TWeightQuant, TKvPolicy, kDelegatedFeedForward, kMixtureOfExperts >::BlockBuildContexts
	The per-child build contexts and split-scratch geometry this block implies. More...
class	Mila::Dnn::GemmaBlock< TDeviceType, TPrecision, kGlobal, TWeightQuant, TKvPolicy, kDelegatedFeedForward, kMixtureOfExperts >
	One Gemma 4 decoder block; kGlobal selects the global (full-attention) geometry. More...
struct	Mila::Dnn::GemmaBlockWorkspace< TDeviceType, TPrecision >
	Transformer-owned shared activation workspace for GemmaBlock (pooling). More...
struct	Mila::Dnn::GemmaBlockWorkspaceWidths
	Max-geometry slot widths, shared by the allocation and the transformer's footprint. More...

Functions
GemmaBlockWorkspaceWidths	Mila::Dnn::gemmaBlockWorkspaceWidths (const GemmaConfig &config)
template<DeviceType TDeviceType, TensorDataType TPrecision> requires PrecisionSupportedOnDevice<TPrecision, TDeviceType>
GemmaBlockWorkspace< TDeviceType, TPrecision >	Mila::Dnn::makeGemmaBlockWorkspace (const GemmaConfig &config, DeviceId device, dim_t B, dim_t prefill_chunk, const std::string &name_prefix)
	Allocate the shared block workspace at the widths gemmaBlockWorkspaceWidths() reports.

Files
file	Mila/Src/Dnn/Components/Transformers/Gemma/Gemma.Block.ixx
	Gemma 4 decoder block (inference: prefill + decode), one type per attention kind.
file	Mila/Src/Dnn/Components/Transformers/Gemma/Gemma.Block.Workspace.ixx
	Transformer-owned shared activation slots for GemmaBlock (pooling), and their factory.