Etc/Etc

ONNX, TVM

light_meal 2022. 8. 18. 23:51
728x90

ONNX, TVM

(ONNX + ONNX Runtime) & TVM

  ONNX + ONNX Runtime TVM
Develop & Support Microsoft & facebook & AWS Apache
System 1. ONNX Runtime Quantization
2. ONNX Runtime Compile
1. Integrated System( Lightweight Model + Compile )
Support ML Framework caffe24
Keras
Tensorflow
PyTorch
CoreML
mxnet
XGBoost
NCNN
...
Pytorch
CoreML
Tensorflow
Keras
ONNX
mxnet …
Deploy Target Device or Environment Arm
Arm
NN
CoreML
CUDA
Windows
AMD
Android
Intel
RKNPU
TensorRT
Vitis
FPGA
NPU
Arm
Android
TensorRT
Vitis
BNNS
DLPack
FPGA
NPU

 

ONNX + ONNX Runtime

https://blog.ml6.eu/bert-is-eating-your-cash-quantization-and-onnxruntime-to-save-ea6dc84dcd88

  • Resnet50 model → ONNX Runtime Quantization BenchmarkAvg: 23.95ms → Avg: 10.91ms

 

TVM

728x90