Update app.py

app.py

@@ -31,7 +31,17 @@ from safetensors.torch import load_file, save_file
 from diffusers import DiffusionPipeline, AnimateDiffPipeline, MotionAdapter, EulerDiscreteScheduler, DDIMScheduler, StableDiffusionXLPipeline, UNet2DConditionModel, AutoencoderKL, UNet3DConditionModel
 #import jax
 #import jax.numpy as jnp
-from numba import jit,njit
+from numba import njit as cpu1, jit as cpu2
+from numba.cuda import njit as cuda1, jit as cuda2
+
+# optimization:
+
+# @cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 
 # logging
 
@@ -117,7 +127,12 @@ function custom(){
 
 # functionality
 
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def run(cmd, assert_success=False, capture_output=False, env=None, dry_run=False):
     if dry_run:
         print(f"--> {cmd}")
@@ -132,7 +147,12 @@ def run(cmd, assert_success=False, capture_output=False, env=None, dry_run=False
 
     return result
     
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def translate(text,lang):
     if text == None or lang == None:
         return ""       
@@ -174,13 +194,23 @@ def translate(text,lang):
     print(ret)
     return ret
     
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def generate_random_string(length):
     characters = string.ascii_letters + string.digits
     return ''.join(random.choice(characters) for _ in range(length))
     
-#@spaces.GPU(duration=65)
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
+@spaces.GPU(duration=65)
 def Piper(image,positive,negative,motion):
     global last_motion
     global ip_loaded
@@ -217,7 +247,12 @@ def Piper(image,positive,negative,motion):
         num_frames=(fps*time)
     )
 
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def infer(pm):
         print("infer: started")
     
@@ -239,7 +274,12 @@ def infer(pm):
         export_to_gif(out.frames[0],name,fps=fps)
         return name
 
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def handle(i,m,p1,p2,*result):    
     p1_en = translate(p1,"english")
     p2_en = translate(p2,"english")
@@ -254,7 +294,12 @@ def handle(i,m,p1,p2,*result):
         ret.append(infer,pm)
     return ret
 
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def ui():
     with gr.Blocks(theme=gr.themes.Soft(),css=css,js=js) as demo:
         with gr.Column(elem_id="col-container"):
@@ -308,7 +353,12 @@ def ui():
         )
         demo.queue().launch()
 
-@njit(cache=True,nopython=True,parallel=True)
+@cuda1(cache=True)
+# @cuda2(cache=True)
+# @cpu1(cache=True,nopython=True,parallel=True)
+# @cpu2(cache=True,nopython=True,parallel=True)
+# @cpu1(cache=True)
+# @cpu2(cache=True)
 def pre():
     pipe = AnimateDiffPipeline.from_pretrained(base, vae=vae, motion_adapter=adapter, torch_dtype=dtype).to(device)
     pipe.scheduler = DDIMScheduler(