This PR adds support for vision inference using IREE, TensorRT and PyTorch.
For IREE, torch-mlir is used for emitting MLIR code.
For TensorRT, a builtin engine builder is used.
For PyTorch, torch.compile in full-graph mode is used for maximizing performance.
This PR adds support for vision inference using IREE, TensorRT and PyTorch.
For IREE,
torch-mlir
is used for emitting MLIR code. For TensorRT, a builtin engine builder is used. For PyTorch,torch.compile
in full-graph mode is used for maximizing performance.