Arm backend: Make quantize_with_submodules private

Change-Id: If9d3aade49e92f45af4266d14e0bcc149c371a21
Signed-off-by: Sebastian Larsson <sebastian.larsson@arm.com>

Author: Sebastian-Larsson

backends/arm/quantizer/arm_quantizer.py

@@ -754,7 +754,7 @@ def validate(self, model: GraphModule) -> None:
                         f"Quantizer detected operator {node.name} with different device inputs: {devices}."
                     )
 
-    def quantize_with_submodules(
+    def _quantize_with_submodules(
         self,
         model: GraphModule,
         calibration_samples: list[tuple],

backends/arm/test/tester/quantize.py

@@ -50,11 +50,11 @@ def run(
             raise ValueError("ArmQuantizer can only run with TOSAQuantizer.")
 
         if self.calibration_samples is not None:
-            converted = self.quantizer.quantize_with_submodules(
+            converted = self.quantizer._quantize_with_submodules(
                 captured_graph, self.calibration_samples, bool(self.is_qat), self.fold_quantize  # type: ignore
             )
         else:
-            converted = self.quantizer.quantize_with_submodules(
+            converted = self.quantizer._quantize_with_submodules(
                 captured_graph, [inputs], bool(self.is_qat), self.fold_quantize
             )
 

docs/source/backends/arm-ethos-u/arm-ethos-u-quantization.md

@@ -34,29 +34,6 @@ def EthosUQuantizer.add_quantizer(self, quantizer: 'Quantizer') -> 'TOSAQuantize
 ```
 Insert a quantizer with highest precedence.
 
-```python
-def EthosUQuantizer.quantize_with_submodules(self, model: 'GraphModule', calibration_samples: 'list[tuple]', is_qat: 'bool' = False, fold_quantize: 'bool' = True):
-```
-Quantizes a GraphModule in a way such that conditional submodules are
-handled properly.
-
-Note: torchao's prepare_pt2e and convert_pt2e natively handle
-while_loop body_fn submodules, so we only manually process cond
-branches and while_loop cond_fn here.
-
-Args:
-- **model (GraphModule)**: The model to quantize.
-- **calibration_samples (list[tuple])**: A list of inputs to used to
-        calibrate the model during quantization. To properly calibrate a
-        model with submodules, at least one sample per code path is
-        needed.
-- **is_qat (bool)**: Whether to do quantization aware training or not.
-- **fold_quantize (bool)**: Enables or disables constant folding when quantization
-        is completed.
-
-Returns:
-- **GraphModule**: The quantized model.
-
 ```python
 def EthosUQuantizer.set_global(self, quantization_config: 'Optional[QuantizationConfig]') -> 'TOSAQuantizer':
 ```

docs/source/backends/arm-vgf/arm-vgf-quantization.md

@@ -53,29 +53,6 @@ def VgfQuantizer.add_quantizer(self, quantizer: 'Quantizer') -> 'TOSAQuantizer':
 ```
 Insert a quantizer with highest precedence.
 
-```python
-def VgfQuantizer.quantize_with_submodules(self, model: 'GraphModule', calibration_samples: 'list[tuple]', is_qat: 'bool' = False, fold_quantize: 'bool' = True):
-```
-Quantizes a GraphModule in a way such that conditional submodules are
-handled properly.
-
-Note: torchao's prepare_pt2e and convert_pt2e natively handle
-while_loop body_fn submodules, so we only manually process cond
-branches and while_loop cond_fn here.
-
-Args:
-- **model (GraphModule)**: The model to quantize.
-- **calibration_samples (list[tuple])**: A list of inputs to used to
-        calibrate the model during quantization. To properly calibrate a
-        model with submodules, at least one sample per code path is
-        needed.
-- **is_qat (bool)**: Whether to do quantization aware training or not.
-- **fold_quantize (bool)**: Enables or disables constant folding when quantization
-        is completed.
-
-Returns:
-- **GraphModule**: The quantized model.
-
 ```python
 def VgfQuantizer.set_global(self, quantization_config: 'Optional[QuantizationConfig]') -> 'TOSAQuantizer':
 ```