alpindale

/

Llama-2-7b-ONNX

like 8

I expected that inferencing with onnxruntime should be faster than normal pytorch inferencing.
but it is much slower than pytorch inf.
I wonder how it could be.