lizhaohu

The TCNN network has both float16 and float32 components. Is it possible to perform loss calculation without enabling mixed precision?

1

The TCNN network has both float16 and float32 components. Is it possible to perform loss calculation without enabling mixed precision?

code

2

Very great work. When will the code be released?

about LORA

1

How to determine which layer should be fine tuned using LORA

Do you have plans to release a training code

The function implementation for _get_background_color was not found.