Merge pull request #6510 from brkirch/unet16-upcast-precision

Add upcast options, full precision sampling from float16 UNet and upcasting attention for inference using SD 2.1 models without --no-half

1 files changed, 2 insertions, 0 deletions

diff --git a/modules/shared.py b/modules/shared.py
index 5f713bee..6a0b96cb 100644
--- a/modules/shared.py
+++ b/modules/shared.py
@@ -45,6 +45,7 @@ parser.add_argument("--lowram", action='store_true', help="load stable diffusion
 parser.add_argument("--always-batch-cond-uncond", action='store_true', help="disables cond/uncond batching that is enabled to save memory with --medvram or --lowvram")

 parser.add_argument("--unload-gfpgan", action='store_true', help="does not do anything.")

 parser.add_argument("--precision", type=str, help="evaluate at this precision", choices=["full", "autocast"], default="autocast")

+parser.add_argument("--upcast-sampling", action='store_true', help="upcast sampling. No effect with --no-half. Usually produces similar results to --no-half with better performance while using less memory.")

 parser.add_argument("--share", action='store_true', help="use share=True for gradio and make the UI accessible through their site")

 parser.add_argument("--ngrok", type=str, help="ngrok authtoken, alternative to gradio --share", default=None)

 parser.add_argument("--ngrok-region", type=str, help="The region in which ngrok should start.", default="us")

@@ -409,6 +410,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), {
     "comma_padding_backtrack": OptionInfo(20, "Increase coherency by padding from the last comma within n tokens when using more than 75 tokens", gr.Slider, {"minimum": 0, "maximum": 74, "step": 1 }),

     "CLIP_stop_at_last_layers": OptionInfo(1, "Clip skip", gr.Slider, {"minimum": 1, "maximum": 12, "step": 1}),

     "extra_networks_default_multiplier": OptionInfo(1.0, "Multiplier for extra networks", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),

+    "upcast_attn": OptionInfo(False, "Upcast cross attention layer to float32"),

 }))

 

 options_templates.update(options_section(('compatibility', "Compatibility"), {