| asInputA(const ITensor &t) | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | inlineprotectedstatic |
| asInputB(const ITensor &t) | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | inlineprotectedstatic |
| asOutputTensor(ITensor &t) | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | inlineprotectedstatic |
| backward(const ITensor &dQ_out, const ITensor &dK_out, ITensor &dQ_in, ITensor &dK_in) const override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| batch_size_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| build(const BuildContext &build_context) override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| cache_key_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| CacheKey typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| clearGradients() noexcept | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| ComputeType typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| config_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| ConfigType typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| context_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| cos_cache_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| CudaExecutionContext typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| CudaRopeOp(IExecutionContext *context, const RopeConfig &config) | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inline |
| CudaRopeOp(const CudaRopeOp &)=delete | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| CudaRopeOp(CudaRopeOp &&other) noexcept | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inline |
| data_type | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | static |
| DataTypeTraits typedef | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | |
| decode(const ITensor &Q_in, const ITensor &K_in, ITensor &Q_out, ITensor &K_out, int position) override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| device_type | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | static |
| dispatchForward(const ITensor &Q_in, const ITensor &K_in, ITensor &Q_out, ITensor &K_out, int B, int T, int position_offset) const | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlineprivate |
| ensureBuilt() const | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlineprivate |
| forward(const ITensor &Q_in, const ITensor &K_in, ITensor &Q_out, ITensor &K_out) const override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| getDataType() const | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| getDeviceType() const | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| getName() const override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| getOperationType() const override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| getStateMemorySize() const override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| is_built_ | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | protected |
| isBuilt() const | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| isEvalMode() const | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| makeCacheKey() const noexcept | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlineprivate |
| MR typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| operator=(const CudaRopeOp &)=delete | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| operator=(CudaRopeOp &&other) noexcept | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inline |
| owns_cache_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| prefill(const ITensor &Q_in, const ITensor &K_in, ITensor &Q_out, ITensor &K_out, int position_offset) override | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlinevirtual |
| releaseCache() noexcept | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlineprivate |
| seq_length_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| setGradients(ITensor *weight_grad, ITensor *bias_grad) | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| setParameters(ITensor *weight, ITensor *bias) | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| setTrainingMode(TrainingMode training_mode) | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | inlinevirtual |
| sin_cache_ | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | private |
| TensorInputAType typedef | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | |
| TensorInputBType typedef | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | |
| TensorOutputType typedef | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | |
| TensorType typedef | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | |
| training_mode_ | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | protected |
| validateRuntimeShape(int B, int T) const | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inlineprivate |
| ~CudaRopeOp() | Mila::Dnn::Compute::Cuda::Rope::CudaRopeOp< TComputePrecision > | inline |
| ~IPositionalPairedOp()=default | Mila::Dnn::Compute::IPositionalPairedOp | virtual |
| ~Operation()=default | Mila::Dnn::Compute::Operation< TDeviceType, TPrecision > | virtual |
| ~PairedOperation()=default | Mila::Dnn::Compute::PairedOperation< DeviceType::Cuda, TComputePrecision > | virtual |
1.15.0