Remove workaround for old pytorch. (Comfy-Org#12480)

Author: comfyanonymous

comfy/ldm/modules/diffusionmodules/model.py

@@ -102,19 +102,7 @@ def forward(self, x):
         return self.conv(x)
 
 def interpolate_up(x, scale_factor):
-    try:
-        return torch.nn.functional.interpolate(x, scale_factor=scale_factor, mode="nearest")
-    except: #operation not implemented for bf16
-        orig_shape = list(x.shape)
-        out_shape = orig_shape[:2]
-        for i in range(len(orig_shape) - 2):
-            out_shape.append(round(orig_shape[i + 2] * scale_factor[i]))
-        out = torch.empty(out_shape, dtype=x.dtype, layout=x.layout, device=x.device)
-        split = 8
-        l = out.shape[1] // split
-        for i in range(0, out.shape[1], l):
-            out[:,i:i+l] = torch.nn.functional.interpolate(x[:,i:i+l].to(torch.float32), scale_factor=scale_factor, mode="nearest").to(x.dtype)
-        return out
+    return torch.nn.functional.interpolate(x, scale_factor=scale_factor, mode="nearest")
 
 class Upsample(nn.Module):
     def __init__(self, in_channels, with_conv, conv_op=ops.Conv2d, scale_factor=2.0):