Performance optimization

app.py

@@ -15,6 +15,7 @@ models_list = api.list_models(author="sd-concepts-library", sort="likes", direct
 models = []
 
 pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", use_auth_token=True, revision="fp16", torch_dtype=torch.float16).to("cuda")
+torch.backends.cudnn.benchmark = True
 
 def load_learned_embed_in_clip(learned_embeds_path, text_encoder, tokenizer, token=None):
   loaded_learned_embeds = torch.load(learned_embeds_path, map_location="cpu")
@@ -146,12 +147,11 @@ def checkbox_block():
   return checkbox
 
 def infer(text):
-  with autocast("cuda"):
-    images_list = pipe(
-              [text]*2,
-              num_inference_steps=50,
-              guidance_scale=7.5
-    )
+  images_list = pipe(
+      [text]*2,
+      num_inference_steps=50,
+      guidance_scale=7.5
+  )
   output_images = []
   for i, image in enumerate(images_list["sample"]):
     output_images.append(image)

requirements.txt

@@ -1,4 +1,4 @@
-diffusers
+-e git+https://github.com/Narsil/diffusers.git@6a4d2ef1e514a25ff5b511cafa1f06b039f0909b#egg=diffusers
 transformers
 nvidia-ml-py3
 ftfy