migrate convert/prepare to torchao

Differential Revision: D75095744

Pull Request resolved: #11015

Author: metascroy

.lintrunner.toml

@@ -378,3 +378,31 @@ command = [
     '--',
     '@{{PATHSFILE}}',
 ]
+
+[[linter]]
+code = "TORCH_AO_IMPORT"
+include_patterns = ["**/*.py"]
+exclude_patterns = [
+    "third-party/**",
+]
+
+command = [
+  "python3",
+  "-m",
+  "lintrunner_adapters",
+  "run",
+  "grep_linter",
+  "--pattern=\\bfrom torch\\.ao\\.quantization\\.(?:quantize_pt2e)(?:\\.[A-Za-z0-9_]+)*\\b",
+  "--linter-name=TorchAOImport",
+  "--error-name=Prohibited torch.ao.quantization import",
+  """--error-description=\
+  Imports from torch.ao.quantization are not allowed. \
+  Please import from torchao.quantization.pt2e instead.\n \
+  * torchao.quantization.pt2e (includes all the utils, including observers, fake quants etc.) \n \
+  * torchao.quantization.pt2e.quantizer (quantizer related objects and utils) \n \
+  * torchao.quantization.pt2e.quantize_pt2e (prepare_pt2e, prepare_qat_pt2e, convert_pt2e) \n\n \
+  If you need something from torch.ao.quantization, you can add your file to an exclude_patterns for TORCH_AO_IMPORT in .lintrunner.toml. \
+  """,
+  "--",
+  "@{{PATHSFILE}}",
+]

.mypy.ini

@@ -97,3 +97,6 @@ ignore_missing_imports = True
 
 [mypy-zstd]
 ignore_missing_imports = True
+
+[mypy-torchao.*]
+follow_untyped_imports = True

backends/apple/coreml/test/test_coreml_quantizer.py

@@ -15,12 +15,12 @@
 )
 
 from executorch.backends.apple.coreml.quantizer import CoreMLQuantizer
-from torch.ao.quantization.quantize_pt2e import (
+from torch.export import export_for_training
+from torchao.quantization.pt2e.quantize_pt2e import (
     convert_pt2e,
     prepare_pt2e,
     prepare_qat_pt2e,
 )
-from torch.export import export_for_training
 
 
 class TestCoreMLQuantizer:

backends/cadence/aot/compiler.py

@@ -37,9 +37,9 @@
 from executorch.exir.passes.sym_shape_eval_pass import HintBasedSymShapeEvalPass
 from executorch.exir.program._program import to_edge_with_preserved_ops
 from torch._inductor.decomposition import remove_decompositions
-from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 from torch.export.exported_program import ExportedProgram
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 from .passes import get_cadence_passes
 
@@ -123,7 +123,7 @@ def prepare_and_convert_pt2(
     assert isinstance(model_gm, torch.fx.GraphModule)
 
     # Prepare
-    prepared_model = prepare_pt2e(model_gm, quantizer)
+    prepared_model = prepare_pt2e(model_gm, quantizer)  # pyre-ignore[6]
 
     # Calibrate
     # If no calibration data is provided, use the inputs

backends/cortex_m/test/test_replace_quant_nodes.py

@@ -17,14 +17,14 @@
 )
 from executorch.exir.dialects._ops import ops as exir_ops
 from torch.ao.quantization.observer import HistogramObserver
-from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
 from torch.ao.quantization.quantizer.quantizer import (
     QuantizationAnnotation,
     QuantizationSpec,
     Quantizer,
 )
 from torch.export import export, export_for_training
 from torch.fx import GraphModule
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 
 @dataclass(eq=True, frozen=True)

backends/example/example_partitioner.py

@@ -19,9 +19,9 @@
 )
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.graph_module import get_control_flow_submodules
-from torch.ao.quantization.pt2e.graph_utils import find_sequential_partitions
 from torch.export import ExportedProgram
 from torch.fx.passes.operator_support import OperatorSupportBase
+from torchao.quantization.pt2e.graph_utils import find_sequential_partitions
 
 
 @final

backends/example/example_quantizer.py

@@ -9,11 +9,14 @@
 
 import torch
 from executorch.backends.example.example_operators.ops import module_to_annotator
-from executorch.backends.xnnpack.quantizer.xnnpack_quantizer_utils import OperatorConfig
 from torch import fx
-from torch.ao.quantization.observer import HistogramObserver, MinMaxObserver
-from torch.ao.quantization.pt2e.graph_utils import find_sequential_partitions
-from torch.ao.quantization.quantizer import QuantizationSpec, Quantizer
+from torchao.quantization.pt2e.graph_utils import find_sequential_partitions
+from torchao.quantization.pt2e.observer import HistogramObserver, MinMaxObserver
+from torchao.quantization.pt2e.quantizer import (
+    OperatorConfig,
+    QuantizationSpec,
+    Quantizer,
+)
 
 
 def get_uint8_tensor_spec(observer_or_fake_quant_ctr):

backends/example/test_example_delegate.py

@@ -17,10 +17,10 @@
     DuplicateDequantNodePass,
 )
 from executorch.exir.delegate import executorch_call_delegate
-
-from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
 from torch.export import export
 
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
+
 from torchvision.models.quantization import mobilenet_v2
 
 

backends/nxp/tests/executorch_pipeline.py

@@ -20,7 +20,7 @@
     to_edge_transform_and_lower,
 )
 from torch import nn
-from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 
 def _quantize_model(model, calibration_inputs: list[tuple[torch.Tensor]]):

backends/nxp/tests/test_quantizer.py

@@ -8,7 +8,7 @@
 import executorch.backends.nxp.tests.models as models
 import torch
 from executorch.backends.nxp.quantizer.neutron_quantizer import NeutronQuantizer
-from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
+from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
 
 def _get_target_name(node):