Fix imports to use torchao public api

Summary: Fixing imports to use torchao.utils API, instead of a nested call from torchao.quantization API.

Differential Revision: D65843742

Author: jainapurva

backends/vulkan/_passes/int4_weight_only_quantizer.py

@@ -35,7 +35,7 @@ def __init__(
         super().__init__()
         self.padding = not _check_linear_int4_k(in_features, groupsize, inner_k_tiles)
         if self.padding:
-            from torchao.quantization.utils import find_multiple
+            from torchao.utils import find_multiple
 
             self.origin_in_features = in_features
             in_features = find_multiple(in_features, (1024,))
@@ -203,7 +203,7 @@ def _create_quantized_state_dict(
                     if self.padding_allowed:
                         import torch.nn.functional as F
 
-                        from torchao.quantization.utils import find_multiple
+                        from torchao.utils import find_multiple
 
                         logging.warn(
                             f"warning: {fqn} is padded to satisfy in_features % 1024 == 0"

backends/xnnpack/test/ops/linear.py

@@ -30,8 +30,8 @@
     from torchao.quantization.quant_api import (
         int8_dynamic_activation_int4_weight,
         quantize_,
-        unwrap_tensor_subclass,
     )
+    from torchao.utils import unwrap_tensor_subclass
 
     torchao_installed = True
 except: