Update app.py

app.py

@@ -199,7 +199,6 @@ def compose_language_descriptions(prompt):
   return out_img
 
 # create model for CLEVR Objects
-timestep_respacing =  100
 clevr_options = model_and_diffusion_defaults_for_clevr()
 
 flags = {
@@ -215,7 +214,7 @@ flags = {
     "num_classes": '2',
     "dataset": "clevr_pos",
     "use_fp16": has_cuda,
-    "timestep_respacing": str(timestep_respacing)
+    "timestep_respacing": '100'
 }
 
 for key, val in flags.items():
@@ -228,6 +227,7 @@ if has_cuda:
     
 clevr_model.to(device)
 clevr_model.load_state_dict(th.load(download_model('clevr_pos'), device))
+print('total clevr_pos parameters', sum(x.numel() for x in clevr_model.parameters()))
 
 def compose_clevr_objects(prompt):
     print(prompt)
@@ -248,14 +248,13 @@ def compose_clevr_objects(prompt):
         half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)
         eps = th.cat([half_eps] * x_t.size(0), dim=0)
         return th.cat([eps, rest], dim=1)
-
-    masks = [True] * (len(coordinates) - 1) + [False]
-    model_kwargs = dict(
-        y=th.tensor(coordinates, dtype=th.float, device=device),
-        masks=th.tensor(masks, dtype=th.bool, device=device)
-    )
     
     def sample(coordinates):
+        masks = [True] * (len(coordinates) - 1) + [False]
+        model_kwargs = dict(
+            y=th.tensor(coordinates, dtype=th.float, device=device),
+            masks=th.tensor(masks, dtype=th.bool, device=device)
+        )
         samples = clevr_diffusion.p_sample_loop(
             model_fn,
             (len(coordinates), 3, options["image_size"], options["image_size"]),