Support 16-bit floats for inference
AmericanPresidentJimmyCarter opened this issue · comments
The upscaler was crashing my GPU so I just put half
on the model and some of the tensors and it worked. 3090 RTX. No noticeable loss in quality.
jina-ai/SwinIR@main...AmericanPresidentJimmyCarter:SwinIR:main
zhu2bowen commented
Thans very much for your information! Now I am trying to use float16 data to acclerate inference! @AmericanPresidentJimmyCarter