| backward(const TensorType &input, const TensorType &output_grad) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| build(const BuildContext &context) final | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| build_context_ | Mila::Dnn::Component< TDeviceType, TComputePrecision > | protected |
| Component(const std::string &name) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineexplicit |
| ComponentBase typedef (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
| decode(const TensorType &q, const TensorType &k, const TensorType &v, dim_t position_offset) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| forward(const TensorType &input) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| getConfig() const noexcept (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| getDeviceId() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| getDeviceType() | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinestatic |
| getExecutionContext() const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| getGradients() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| getMemoryStats() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| getModelDim() const noexcept (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| getName() const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inline |
| getNumHeads() const noexcept (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| getNumKvHeads() const noexcept (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| getParameterNames() const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| getParameters() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| getPrecision() noexcept | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinestatic |
| getRequiredMemory(const BuildContext &context) const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| getTrainingMode() const noexcept | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inline |
| getType() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| GroupedQueryAttention(const std::string &name, const GqaConfig &config, std::optional< DeviceId > device_id=std::nullopt) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlineexplicit |
| hasExecutionContext() const noexcept | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| HostStagingMemoryResource typedef | Mila::Dnn::Component< TDeviceType, TComputePrecision > | protected |
| installSharedOutput(std::shared_ptr< TensorType > output) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| isBuilt() const final | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| kCacheDtype (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | static |
| kKvCompressed (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | static |
| KvCacheTensorType typedef (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
| load_(ModelArchive &archive, SerializationMode mode) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| loadParameter(const std::string &, const Serialization::ITensorBlob &) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| loadParameterFromBlob(const std::string ¶m_name, const Serialization::ITensorBlob &blob, Tensor< TParameterPrecision, TMemoryResource > &target, const shape_t &expected_shape) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| MR typedef (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
| onBuilding(const BuildContext &context) override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlineprotectedvirtual |
| onExecutionContextSet() override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlineprotectedvirtual |
| onTrainingModeChanging(TrainingMode training_mode) override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlineprotectedvirtual |
| OpType typedef (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
| parameterCount() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| prefill(const TensorType &q, const TensorType &k, const TensorType &v, dim_t position_offset) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| requireSerializableParameters() const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| resetKVCache() | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| rewindKvCache(dim_t position) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| save_(ModelArchive &, SerializationMode) const override (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| saveFlatTensors(Serialization::SafeTensorsWriter &writer, const std::string &prefix, Serialization::TensorSavePass pass) const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| saveParameterToArchive(ModelArchive &archive, const std::string ¶meter_name, const Tensor< TParameterPrecision, TMemoryResource > ¶meter) const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| saveParameterToWriter(Serialization::SafeTensorsWriter &writer, const std::string &flat_name, const Tensor< TParameterPrecision, TMemoryResource > ¶meter, Serialization::TensorSavePass pass) const | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| setExecutionContext(IExecutionContext *context) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlineprotected |
| setState(const GqaState &state) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| setTrainingMode(TrainingMode mode) | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inline |
| setUseFlashDecode(bool enabled) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| setUseFlashPrefill(bool enabled) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| supportsKVCache() const noexcept | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inline |
| synchronize() override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| TensorType typedef (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
| toString() const override | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | inlinevirtual |
| zeroGradients() | Mila::Dnn::Component< TDeviceType, TComputePrecision > | inlinevirtual |
| ~GroupedQueryAttention() override=default (defined in Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >) | Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy > | |
1.17.0