README
1For C++ implementation, there are no plans to migrate quantization kernels. For sparsity, we plan to follow the structure below:
2
3./torch/aten/src/ATen
4├── (...)
5├── native
6│ ├── ao_sparse # New folder containing ops with sparse packed weights
7│ │ ├── cpu
8│ │ ├── cuda
9│ │ └── quantized # New folder with fused sparse and quantized packed weights
10│ │ ├── cpu
11│ │ └── cuda
12│ ├── quantized # Existing quantized kernels
13│ │ ├── cpu
14│ │ │ ├── kernels
15│ │ │ └── qnnpack # Existing mobile quantized and sparse kernels
16│ │ ├── cuda
17│ │ ├── QTensor.cpp
18│ │ ├── library.cpp
19│ │ └── (*.cpp, *.h)
20│ └── sparse # Existing folder containing torch.sparse methods
21└── quantized # Quantized Tensor definition and methods
22 ├── QTensorImpl.{cpp,h}
23 └── Quantizer.{cpp,h}
24