Hugging Face's logo

Kim2091
/
UltraSharpV2

Image-to-Image
ONNX
PyTorch
super-resolution
Model card Files Community
3

The ONNX format model of UltraSharp V2 FP16 has an inference error.

#3
by nukui - opened
Discussion
nukui

I am using ONNX Runtime version 1.22.0, which performs well for FP32 model inference and successfully runs the UltraSharp Lite FP16 model. However, for the UltraSharp FP16 model, it returns a tensor filled with NaN values.
Can you explain your approach to running inference on the FP16 ONNX model? I'd really appreciate it.

Kim2091
 Owner

@nukui Apologies. I actually should not have uploaded an FP16 version as DAT2 is not stable in that format. The conversion is not usable.

nukui

Thanks for your reply, the other models work great!

nukui changed discussion status to closed

Sign up or log in to comment