EVO Operator
TODO Operator
1 Hardware Level Optimize
1.1 Support CPU
Name |
ISA |
Company |
AVX |
amd64 |
Intel |
AMX |
amd64 |
Intel |
NEON |
aarch64 |
Arm |
RVV |
riscv |
UCB |
1.2 Support GPU
Name |
ISA |
Company |
CUDA |
|
|
Vulkan |
|
|
OpenCL |
|
|
Metal |
|
|
1.3 Support NPU
Name |
ISA |
Company |
CoreML |
|
|
HIAI |
|
|
NNAPI |
|
|
2 Hign Perfermance Operator Lib
Name |
ISA |
Company |
cuDNN |
|
|
MKLDNN |
|
|
3 onnx operator
total 162:
4 learnable parameters
- Conv:
- kernel: [1, 1, K_h, K_w]
- bias : []
- params: (K_h * K_w * C_in + 0/1) * C_out
- FLOPS : (K_h * K_w * C_in + 0/1) * C_out * (H_out * W_out)
- FLOPs : 2 *
- FC:
- weight: []
- bias : []
- params: (C_in + 0/1) * C_out
- FLOPS : (C_in + 0/1) * C_out
- FLOPs :
- BN:
- Activation:
5 hyper parameters
- learning rate
- batch size
- iterations
- epochs
data_size = 1200
batch_size = 100
epochs = 5
update_count = (1200 / 100) * 5 = 60