Add both 32-bit (normal) and 8-bit (quantized) versions of the model weights, in ONNX format.
cc @sgugger
LGTM
· Sign up or log in to comment