Mila 0.13.48
Deep Neural Network Library
Loading...
Searching...
No Matches
Mila::Dnn::GptBlock< TDeviceType, TPrecision > Member List

This is the complete list of members for Mila::Dnn::GptBlock< TDeviceType, TPrecision >, including all inherited members.

addComponent(ComponentPtr component)Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
AttentionType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
attn_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
backward(const TensorType &input, const TensorType &output_grad)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
build(const BuildContext &context) finalMila::Dnn::Component< TDeviceType, TPrecision >inlinevirtual
build_context_Mila::Dnn::Component< TDeviceType, TPrecision >protected
built_Mila::Dnn::Component< TDeviceType, TPrecision >private
cached_input_shape_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
child_component_map_Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >private
child_components_Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >private
childCount() const noexceptMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
clearComponents()Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
Component(const std::string &name)Mila::Dnn::Component< TDeviceType, TPrecision >inlineexplicit
ComponentBase typedefMila::Dnn::CompositeComponent< TDeviceType, TPrecision >
ComponentPtr typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
CompositeComponent(const std::string &name)Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlineexplicit
CompositeComponent(const CompositeComponent &)=deleteMila::Dnn::CompositeComponent< TDeviceType, TPrecision >
CompositeComponent(CompositeComponent &&) noexcept=defaultMila::Dnn::CompositeComponent< TDeviceType, TPrecision >
CompositeComponentBase typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
config_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
createGraph()Mila::Dnn::GptBlock< TDeviceType, TPrecision >inlineprivate
d_input_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
d_res1_accum_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
decode(const TensorType &input, int position)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
ensureBuilt(const char *method) constMila::Dnn::Component< TDeviceType, TPrecision >inlineprivate
exec_context_Mila::Dnn::Component< TDeviceType, TPrecision >private
ExecutionContextType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
ffn_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
findComponent(const std::string &path) constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
forward(const TensorType &input)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
forward_executed_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
getComponent(const std::string &name) constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
getComponentAs(const std::string &name) constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlineprotected
getComponents() constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
getDeviceId() const overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlinevirtual
getDeviceType()Mila::Dnn::Component< TDeviceType, TPrecision >inlinestatic
getExecutionContext() constMila::Dnn::Component< TDeviceType, TPrecision >inlineprotected
getGradients() const overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlinevirtual
getMemoryStats() const overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlinevirtual
getName() constMila::Dnn::Component< TDeviceType, TPrecision >inline
getParameterNames() constMila::Dnn::Component< TDeviceType, TPrecision >inlinevirtual
getParameters() const overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlinevirtual
getPrecision() noexceptMila::Dnn::Component< TDeviceType, TPrecision >inlinestatic
getRuntimeMode() const noexceptMila::Dnn::Component< TDeviceType, TPrecision >inline
getTrainingMode() const noexceptMila::Dnn::Component< TDeviceType, TPrecision >inline
getType() const overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlinevirtual
GptBlock(const std::string &name, const GptBlockConfig &config, std::optional< DeviceId > device_id=std::nullopt)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inlineexplicit
hasChildren() const noexceptMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
hasComponent(const std::string &name) constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
hasExecutionContext() const noexceptMila::Dnn::Component< TDeviceType, TPrecision >inlineprotected
initializeKVCache(int64_t max_seq_len)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
isBuilt() const finalMila::Dnn::Component< TDeviceType, TPrecision >inlinevirtual
isIdentifier(const std::string &s) noexceptMila::Dnn::Component< TDeviceType, TPrecision >inlineprivatestatic
isInferenceMode() const noexceptMila::Dnn::Component< TDeviceType, TPrecision >inline
isTrainingMode() const noexceptMila::Dnn::Component< TDeviceType, TPrecision >inline
last_attn_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_ffn_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_ln1_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_ln2_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_out_proj_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_qkv_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_res1_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
last_res2_out_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
LayerNormType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
LinearType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
ln1_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
ln2_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
load_(ModelArchive &archive, SerializationMode mode)Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
loadParameter(const std::string &name, const Serialization::ITensorBlob &blob)Mila::Dnn::Component< TDeviceType, TPrecision >inlinevirtual
loadParameterFromBlob(const std::string &param_name, const Serialization::ITensorBlob &blob, Tensor< TParameterPrecision, TMemoryResource > &target, const shape_t &expected_shape)Mila::Dnn::Component< TDeviceType, TPrecision >inlineprotected
MLPType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
MR typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
name_Mila::Dnn::Component< TDeviceType, TPrecision >private
onBuilding(const BuildContext &context) overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlineprotectedvirtual
onExecutionContextSet() overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlineprotectedvirtual
onTrainingModeChanging(TrainingMode training_mode) overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlineprotectedvirtual
operator=(const CompositeComponent &)=deleteMila::Dnn::CompositeComponent< TDeviceType, TPrecision >
operator=(CompositeComponent &&) noexcept=defaultMila::Dnn::CompositeComponent< TDeviceType, TPrecision >
optimize()Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlineprotectedvirtual
out_proj_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
owned_exec_context_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
parameterCount() const overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlinevirtual
qkv_proj_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
removeComponent(const std::string &name)Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
res1_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
res2_Mila::Dnn::GptBlock< TDeviceType, TPrecision >private
resetKVCache()Mila::Dnn::GptBlock< TDeviceType, TPrecision >inline
ResidualType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
save_(ModelArchive &archive, SerializationMode mode) const overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlinevirtual
setExecutionContext(IExecutionContext *context)Mila::Dnn::Component< TDeviceType, TPrecision >inlineprotected
setTrainingMode(TrainingMode mode)Mila::Dnn::Component< TDeviceType, TPrecision >inline
supportsKVCache() const noexceptMila::Dnn::GptBlock< TDeviceType, TPrecision >inline
synchronize() overrideMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inlinevirtual
TensorType typedefMila::Dnn::GptBlock< TDeviceType, TPrecision >
toString() const overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlinevirtual
training_mode_Mila::Dnn::Component< TDeviceType, TPrecision >private
training_mode_mutex_Mila::Dnn::Component< TDeviceType, TPrecision >private
tryFindComponent(const std::string &path) constMila::Dnn::CompositeComponent< TDeviceType, TPrecision >inline
validateBuildContext(const BuildContext &context) constMila::Dnn::GptBlock< TDeviceType, TPrecision >inlineprivate
validateInputShape(const shape_t &input_shape) constMila::Dnn::GptBlock< TDeviceType, TPrecision >inlineprivate
validateName(const std::string &name)Mila::Dnn::Component< TDeviceType, TPrecision >inlineprivatestatic
zeroGradients() overrideMila::Dnn::GptBlock< TDeviceType, TPrecision >inlinevirtual
~Component()=defaultMila::Dnn::Component< TDeviceType, TPrecision >virtual
~CompositeComponent()=defaultMila::Dnn::CompositeComponent< TDeviceType, TPrecision >virtual
~GptBlock() override=defaultMila::Dnn::GptBlock< TDeviceType, TPrecision >