Spaces:

lambdalabs

/

text-to-pokemon

like 73

Runtime error

Add ONNX CPUExecutionProvider since our Benchmark showed it can speed inference up by 40%~60% (depending on the types of CPUs).

CUDAExecutionProvider as a back-end also gets more promising. We can support it once a couple of things (1,2) are fully fixed in the upstreams.

Results from my local test (CPUExecutionProvider takes 40% less time compare to native PyTorch inference):