Migrate cadence

.lintrunner.toml

@@ -390,7 +390,6 @@ exclude_patterns = [
     "backends/arm/test/ops/**",
     "backends/vulkan/quantizer/**",
     "backends/vulkan/test/**",
-    "backends/cadence/aot/quantizer/**",
     "backends/qualcomm/quantizer/**",
     "examples/qualcomm/**",
     "backends/xnnpack/quantizer/**",

backends/cadence/aot/compiler.py

@@ -123,7 +123,7 @@ def prepare_and_convert_pt2(
     assert isinstance(model_gm, torch.fx.GraphModule)
 
     # Prepare
-    prepared_model = prepare_pt2e(model_gm, quantizer)  # pyre-ignore[6]
+    prepared_model = prepare_pt2e(model_gm, quantizer)
 
     # Calibrate
     # If no calibration data is provided, use the inputs

backends/cadence/aot/quantizer/TARGETS

@@ -9,6 +9,7 @@ python_library(
     ],
     deps = [
         "//caffe2:torch",
+        "//pytorch/ao:torchao",
     ],
 )
 
@@ -34,7 +35,6 @@ python_library(
         ":patterns",
         ":utils",
         "//caffe2:torch",
-        "//executorch/backends/xnnpack/quantizer:xnnpack_quantizer_utils",
     ],
 )
 

backends/cadence/aot/quantizer/patterns.py

@@ -15,7 +15,7 @@
 
 from torch import fx
 from torch._ops import OpOverload
-from torch.ao.quantization.quantizer import (
+from torchao.quantization.pt2e.quantizer import (
     DerivedQuantizationSpec,
     SharedQuantizationSpec,
 )

backends/cadence/aot/quantizer/quantizer.py

@@ -29,19 +29,20 @@
     is_annotated,
     no_outside_users,
 )
-from executorch.backends.xnnpack.quantizer.xnnpack_quantizer_utils import (
+
+from torch import fx
+
+from torchao.quantization.pt2e import HistogramObserver, MinMaxObserver
+from torchao.quantization.pt2e.quantizer import (
+    ComposableQuantizer,
+    DerivedQuantizationSpec,
     OperatorConfig,
     QuantizationAnnotation,
     QuantizationConfig,
     QuantizationSpec,
+    Quantizer,
 )
 
-from torch import fx
-
-from torch.ao.quantization.observer import HistogramObserver, MinMaxObserver
-from torch.ao.quantization.quantizer import DerivedQuantizationSpec, Quantizer
-from torch.ao.quantization.quantizer.composable_quantizer import ComposableQuantizer
-
 
 act_qspec_asym8s = QuantizationSpec(
     dtype=torch.int8,

backends/cadence/aot/quantizer/utils.py

@@ -14,13 +14,13 @@
 import torch
 from torch import fx
 from torch._ops import OpOverload
-from torch.ao.quantization import ObserverOrFakeQuantize
 
 from torch.fx import GraphModule
 from torch.fx.passes.utils.source_matcher_utils import (
     check_subgraphs_connected,
     SourcePartition,
 )
+from torchao.quantization.pt2e import ObserverOrFakeQuantize
 
 
 def quantize_tensor_multiplier(