|
Mila 0.13.48
Deep Neural Network Library
|
CUDA implementation of AdamW optimizer. More...
#include <vector>#include <memory>#include <stdexcept>#include <string>#include <sstream>#include <cstdint>#include <cuda_runtime.h>#include "Kernels/CudaOptimizers.h"import Compute.ExecutionContext;import Compute.CudaDeviceMemoryResource;import Compute.OptimizerBase;import Compute.DeviceType;import Dnn.TensorInitializers;import Dnn.TensorDataTypeTraits;import Compute.CudaTensorDataType;import Dnn.TensorDataType;import Dnn.TensorTypes;import Dnn.Tensor;import Dnn.ITensor;import Dnn.Optimizers.AdamWConfig;Classes | |
| class | Mila::Dnn::Compute::CudaAdamWOptimizer< TPrecision > |
| CUDA-specific AdamW optimizer implementation. More... | |
Namespaces | |
| namespace | Mila |
| Mila main API namespace. | |
| namespace | Mila::Dnn |
| namespace | Mila::Dnn::Compute |
Typedefs | |
| using | Mila::Dnn::Compute::OptimizerConfig = Mila::Dnn::Optimizers::AdamWConfig |
CUDA implementation of AdamW optimizer.
Bridges the Mila optimizer interface to the AdamW CUDA kernel (adamw.cuh). Manages per-parameter state tensors (momentum, variance) and performs parameter updates using the optimized CUDA kernel implementation.