1. cache examples. 2. concurrency=2

app.py

@@ -29,6 +29,7 @@ model = RiffusionPipeline.from_pretrained(
 
 if torch.cuda.is_available():
   model.to("cuda")
+  model.enable_xformers_memory_efficient_attention()
 
 pipe_inpaint = StableDiffusionInpaintPipeline.from_pretrained(repo_id, torch_dtype=torch.float16, safety_checker=lambda images, **kwargs: (images, False),)
 pipe_inpaint.scheduler = DPMSolverMultistepScheduler.from_config(pipe_inpaint.scheduler.config)
@@ -37,6 +38,7 @@ pipe_inpaint.scheduler = DPMSolverMultistepScheduler.from_config(pipe_inpaint.sc
 
 if torch.cuda.is_available():
     pipe_inpaint = pipe_inpaint.to("cuda")
+    pipe.enable_xformers_memory_efficient_attention()
 
 
 def get_init_image(image, overlap, feel):
@@ -309,7 +311,7 @@ with gr.Blocks(css=css) as app:
             ["typing", "dance beat", "og_beat", 10],
             ["synthwave", "jazz", "agile", 10],
             ["rap battle freestyle", "", "og_beat", 10],
-            ["techno club banger", "", "og_beat", 10],
+            # ["techno club banger", "", "og_beat", 10],
             ["reggae dub beat", "sunset chill", "og_beat", 10],
             ["acoustic folk ballad", "", "agile", 10],
             ["blues guitar riff", "", "agile", 5],
@@ -318,7 +320,8 @@ with gr.Blocks(css=css) as app:
             ["rock and roll power chord", "", "motorway", 5],
             ["soulful R&B love song", "", "marim", 10],
             ["country western twangy guitar", "", "agile", 10]],
-        inputs=[prompt_1, prompt_2, feel, num_iterations])
+        inputs=[prompt_1, prompt_2, feel, num_iterations].
+        cache_examples=True)
     
     gr.HTML("""
     <div style="border-top: 1px solid #303030;">
@@ -331,4 +334,4 @@ with gr.Blocks(css=css) as app:
     </div>
     """)
 
-app.launch()
+app.queue(max_size=250, concurrency_count=2).launch()