add

app.py

@@ -98,7 +98,6 @@ def load_models():
                 torch_dtype=torch.float16,
                 local_files_only=False,
                 resume_download=True,
-                token=True  # Use token-based auth
             )
             break
         except (ReadTimeout, ConnectionError) as e:
@@ -133,7 +132,6 @@ def load_models():
                 "YiftachEde/Sharp-It",
                 local_files_only=False,
                 resume_download=True,
-                token=True  # Use token-based auth
             ).to(torch.float16)
             break
         except (ReadTimeout, ConnectionError) as e:
@@ -157,7 +155,6 @@ def load_models():
                 repo_type="model",
                 local_files_only=False,
                 resume_download=True,
-                token=True,  # Use token-based auth
                 cache_dir="model_cache"  # Use a specific cache directory
             )
             break
@@ -293,7 +290,7 @@ class ShapERenderer:
             batch_size = 1
             guidance_scale = float(guidance_scale)
             
-            with torch.cuda.amp.autocast():  # Use automatic mixed precision
+            with torch.amp.autocast('cuda'):  # Use automatic mixed precision
                 latents = sample_latents(
                     batch_size=batch_size,
                     model=self.model,
@@ -320,7 +317,7 @@ class ShapERenderer:
             
             for i, (azimuth, elevation) in enumerate(zip(azimuths, elevations)):
                 cameras = create_custom_cameras(size, self.device, azimuths=[azimuth], elevations=[elevation], fov_degrees=30, distance=3.0)
-                with torch.cuda.amp.autocast():  # Use automatic mixed precision
+                with torch.amp.autocast('cuda'):  # Use automatic mixed precision
                     rendered_image = decode_latent_images(
                         self.xm,
                         latents[0],
@@ -381,7 +378,7 @@ class RefinerInterface:
                 input_image = Image.fromarray(new_layout)
             
             # Process with the pipeline (expects 960x640)
-            with torch.cuda.amp.autocast():  # Use automatic mixed precision
+            with torch.amp.autocast('cuda'):  # Use automatic mixed precision
                 refined_output_960x640 = self.pipeline.refine(
                     input_image,
                     prompt=prompt,
@@ -392,7 +389,7 @@ class RefinerInterface:
             torch.cuda.empty_cache()  # Clear GPU memory after refinement
             
             # Generate mesh using the 960x640 format
-            with torch.cuda.amp.autocast():  # Use automatic mixed precision
+            with torch.amp.autocast('cuda'):  # Use automatic mixed precision
                 vertices, faces, vertex_colors = create_mesh(
                     refined_output_960x640, 
                     self.model,