Do you recommend BF16 as the "save precision" for a 3090?
Do you recommend BF16 as the "save precision" for a 3090?
pip install tensorflow==2.1.0
pip install retinaface==0.0.6ERROR: Could not find a version that satisfies the requirement tensorflow==2.5.0sdxl_gen_img script from kohya? I see in diffusers people leverage both of the text encoders by actually sending in two prompts (and optionally two negative prompts). I'm guessing that using just a single prompt will only leverage one of the text encoders?learning_rate_te. What does that refer to vs learning_rate_te1?








