put sam2 cuda utils in its tab

app.py

@@ -20,12 +20,6 @@ def float32_high_matmul_precision():
     finally:
         torch.set_float32_matmul_precision("highest")
 
-# use bfloat16 for the entire notebook
-torch.autocast("cuda", dtype=torch.bfloat16).__enter__()
-# turn on tfloat32 for Ampere GPUs (https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices)
-if torch.cuda.get_device_properties(0).major >= 8:
-    torch.backends.cuda.matmul.allow_tf32 = True
-    torch.backends.cudnn.allow_tf32 = True
 
 pipe = FluxFillPipeline.from_pretrained(
     "black-forest-labs/FLUX.1-Fill-dev", torch_dtype=torch.bfloat16
@@ -140,6 +134,12 @@ def rmbg(image=None, url=None):
 
 
 def mask_generation(image=None, d=None):
+    # use bfloat16 for the entire notebook
+    # torch.autocast("cuda", dtype=torch.bfloat16).__enter__()
+    # # turn on tfloat32 for Ampere GPUs (https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices)
+    # if torch.cuda.get_device_properties(0).major >= 8:
+    #     torch.backends.cuda.matmul.allow_tf32 = True
+    #     torch.backends.cudnn.allow_tf32 = True
     d = eval(d)  # convert this to dictionary
     with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
         predictor = SAM2ImagePredictor.from_pretrained("facebook/sam2.1-hiera-large")