PyPI - mct-nightly - Versions diffs - 2.1.0.20240623.439__tar.gz → 2.1.0.20240625.423__tar.gz - Mend

mct-nightly 2.1.0.20240623.439tar.gz → 2.1.0.20240625.423tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (530) hide show

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/PKG-INFO RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.1.0.20240623.439
+Version: 2.1.0.20240625.423
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -56,11 +56,13 @@ Description: # Model Compression Toolkit (MCT)
         Currently, MCT is being tested on various Python, Pytorch and TensorFlow versions:
-        |             |  PyTorch 1.13                                                                                                                                                                                                               | PyTorch 2.0                                                                                                                                                                                                              | PyTorch 2.1                                                                                                                                                                                                              |
+        |             |  PyTorch 2.1                                                                                                                                                                                                               | PyTorch 2.2                                                                                                                                                                                                              | PyTorch 2.3                                                                                                                                                                                                              |
         |-------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-        | Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   |
-        | Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) |
-        | Python 3.11 | | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) |
+        | Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml)   |
+        | Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml) |
+        | Python 3.11 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml) |
         |             | TensorFlow 2.12                                                                                                                                                                                                                        | TensorFlow 2.13                                                                                                                                                                                                                        | TensorFlow 2.14                                                                                                                                                                                                                        | TensorFlow 2.15                                                                                                                                                                                                        |

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/README.md RENAMED Viewed

@@ -50,11 +50,13 @@ for hands-on learning. For example:
 Currently, MCT is being tested on various Python, Pytorch and TensorFlow versions:
-|             |  PyTorch 1.13                                                                                                                                                                                                               | PyTorch 2.0                                                                                                                                                                                                              | PyTorch 2.1                                                                                                                                                                                                              |
+|             |  PyTorch 2.1                                                                                                                                                                                                               | PyTorch 2.2                                                                                                                                                                                                              | PyTorch 2.3                                                                                                                                                                                                              |
 |-------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   |
-| Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) |
-| Python 3.11 | | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) |
+| Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml)   |
+| Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml) |
+| Python 3.11 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml) |
 |             | TensorFlow 2.12                                                                                                                                                                                                                        | TensorFlow 2.13                                                                                                                                                                                                                        | TensorFlow 2.14                                                                                                                                                                                                                        | TensorFlow 2.15                                                                                                                                                                                                        |

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/mct_nightly.egg-info/PKG-INFO RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.1.0.20240623.439
+Version: 2.1.0.20240625.423
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -56,11 +56,13 @@ Description: # Model Compression Toolkit (MCT)
         Currently, MCT is being tested on various Python, Pytorch and TensorFlow versions:
-        |             |  PyTorch 1.13                                                                                                                                                                                                               | PyTorch 2.0                                                                                                                                                                                                              | PyTorch 2.1                                                                                                                                                                                                              |
+        |             |  PyTorch 2.1                                                                                                                                                                                                               | PyTorch 2.2                                                                                                                                                                                                              | PyTorch 2.3                                                                                                                                                                                                              |
         |-------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-        | Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch113.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch20.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   |
-        | Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch112.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch113.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) |
-        | Python 3.11 | | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch20.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) |
+        | Python 3.9  | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch21.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch22.yml)   | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_pytorch23.yml)   |
+        | Python 3.10 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_pytorch23.yml) |
+        | Python 3.11 | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch21.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch22.yml) | [![Run Tests](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml/badge.svg)](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_pytorch23.yml) |
         |             | TensorFlow 2.12                                                                                                                                                                                                                        | TensorFlow 2.13                                                                                                                                                                                                                        | TensorFlow 2.14                                                                                                                                                                                                                        | TensorFlow 2.15                                                                                                                                                                                                        |

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/__init__.py RENAMED Viewed

@@ -27,4 +27,4 @@ from model_compression_toolkit import data_generation
 from model_compression_toolkit import pruning
 from model_compression_toolkit.trainable_infrastructure.keras.load_model import keras_load_quantized_model
-__version__ = "2.1.0.20240623.000439"
+__version__ = "2.1.0.20240625.000423"

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/core/common/graph/functional_node.py RENAMED Viewed

@@ -1,4 +1,4 @@
-from typing import Dict, Any, Tuple, Type
+from typing import Dict, Any, Tuple, Type, List, Union
 from model_compression_toolkit.constants import FOUND_TF
 from model_compression_toolkit.core.common.graph.base_node import BaseNode
@@ -25,7 +25,7 @@ class FunctionalNode(BaseNode):
                  functional_op: Any = None,
                  inputs_as_list: bool = False,
                  has_activation: bool = True,
-                 tensor_input_allocs = None):
+                 tensor_input_allocs: List[Union[int, str]] = None):
         """
         Init a FunctionalNode object.
@@ -44,8 +44,7 @@ class FunctionalNode(BaseNode):
             functional_op: The op the node implements.
             inputs_as_list: Whether to pass the node its input tensors as a list or not when calling the layer.
             has_activation: Whether the node has activations that we might want to quantize.
-            tensor_input_allocs: A list of indices for activation tensors in the node's input tensor list
+            tensor_input_allocs: A list of indices and strings for allocations input tensors in the node's args and kwargs.
         """
         super().__init__(name,

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/core/pytorch/back2framework/pytorch_model_builder.py RENAMED Viewed

@@ -106,7 +106,7 @@ def _run_operation(n: BaseNode,
                    input_tensors: List,
                    op_func: Any,
                    quantize_node_activation_fn,
-                   use_activation_quantization: bool) -> Tuple[Union[List, torch.Tensor], Union[List, torch.Tensor]]:
+                   use_activation_quantization: bool) -> Tuple[List[torch.Tensor], List[torch.Tensor]]:
     """
     Applying the layer (op_func) to the input tensors (input_tensors).
     If quantized is set to True, and the layer's corresponding node (n) has quantization
@@ -126,17 +126,17 @@ def _run_operation(n: BaseNode,
     op_call_args = n.op_call_args if isinstance(n, FunctionalNode) else []
     functional_kwargs = n.op_call_kwargs if isinstance(n, FunctionalNode) else {}
-    if not (isinstance(n, FunctionalNode) and isinstance(op_func, PytorchQuantizationWrapper)):
-        # Insert positional weights only when not a quantized functional node, because quantized functional nodes
-        # insert the quantized weights in the wrapper.
+    # Insert positional weights only when not a quantized functional node, because quantized functional nodes
+    # insert the quantized weights in the wrapper.
+    if isinstance(n, FunctionalNode) and isinstance(op_func, PytorchQuantizationWrapper):
+        _tensor_input_allocs = [i for i in n.tensor_input_allocs if i not in n.weights]
+    else:
         input_tensors = n.insert_positional_weights_to_input_list(input_tensors)
         # convert inputs from positional weights (numpy arrays) to tensors. Must handle each element in the
         # list separately, because in FX the tensors are FX objects and fail to_torch_tensor
         input_tensors = [to_torch_tensor(t, numpy_type=t.dtype) if isinstance(t, np.ndarray) else t
                          for t in input_tensors]
         _tensor_input_allocs = None
-    else:
-        _tensor_input_allocs = [i for i in n.tensor_input_allocs if i not in n.weights]
     if isinstance(n, FunctionalNode) and n.inputs_as_list:
         out_tensors_of_n_float = op_func(input_tensors, *op_call_args, **functional_kwargs)
@@ -152,6 +152,8 @@ def _run_operation(n: BaseNode,
             out_tensors_of_n_float = torch.cat(out_tensors_of_n_float, dim=0)
         out_tensors_of_n = quantize_node_activation_fn(out_tensors_of_n_float)
+    if not isinstance(out_tensors_of_n, list):
+        out_tensors_of_n, out_tensors_of_n_float = [out_tensors_of_n], [out_tensors_of_n_float]
     return out_tensors_of_n, out_tensors_of_n_float
@@ -318,12 +320,8 @@ class PytorchModel(torch.nn.Module):
                                                                       quantize_node_activation_fn=activation_quantization_fn,
                                                                       use_activation_quantization=use_activation_quantization)
-            if isinstance(out_tensors_of_n, list):
-                node_to_output_tensors_dict.update({node: out_tensors_of_n})
-                node_to_output_tensors_dict_float.update({node: out_tensors_of_n_float})
-            else:
-                node_to_output_tensors_dict.update({node: [out_tensors_of_n]})
-                node_to_output_tensors_dict_float.update({node: [out_tensors_of_n_float]})
+            node_to_output_tensors_dict.update({node: out_tensors_of_n})
+            node_to_output_tensors_dict_float.update({node: out_tensors_of_n_float})
         if self.append2output:
             outputs = _generate_outputs(self.append2output,

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/const_holder_conv.py RENAMED Viewed

@@ -19,6 +19,7 @@ from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common.graph.base_graph import Graph
 from model_compression_toolkit.core.common.graph.base_node import BaseNode
+from model_compression_toolkit.core.common.graph.functional_node import FunctionalNode
 from model_compression_toolkit.core.pytorch.constants import IN_CHANNELS, OUT_CHANNELS, KERNEL_SIZE, KERNEL, BIAS
 from model_compression_toolkit.core.common import FrameworkInfo
@@ -37,7 +38,7 @@ class FunctionalConvSubstitution(common.BaseSubstitution):
     def substitute(self,
                    graph: Graph,
-                   func_node: BaseNode) -> Graph:
+                   func_node: FunctionalNode) -> Graph:
         """
         Substitute functional and conv/linear layer with torch layer
         Args:
@@ -60,9 +61,15 @@ class FunctionalConvSubstitution(common.BaseSubstitution):
         # Create new node of layer convolution
         if 1 not in func_node.weights:
             Logger.critical(f'Weight input missing for node {func_node.name}.')  # pragma: no cover
-        weight = func_node.weights[1]
-        bias = func_node.weights.get(2)
-        framework_attr = func_node.framework_attr
+        # Extract index of kernel and bias according to tensor_input_allocs if they were input as kwargs. If
+        # they were input as args, use their fixed positions.
+        weight_index = func_node.tensor_input_allocs.index(KERNEL) if KERNEL in func_node.tensor_input_allocs else 1
+        bias_index = func_node.tensor_input_allocs.index(BIAS) if BIAS in func_node.tensor_input_allocs else 2
+        if weight_index not in func_node.weights:
+            Logger.critical(f'Mismatch between tensor_input_allocs and weight index in node {func_node.name}.')  # pragma: no cover
+        weight = func_node.weights[weight_index]
+        bias = func_node.weights.get(bias_index)
+        framework_attr = func_node.op_call_kwargs
         framework_attr.update({OUT_CHANNELS: weight.shape[out_channel_index]})
         framework_attr.update({IN_CHANNELS: weight.shape[in_channel_index]})
         framework_attr.update({KERNEL_SIZE: weight.shape[2:]})

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/functional_batch_norm.py RENAMED Viewed

@@ -20,6 +20,7 @@ import torch.nn.functional as F
 from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
 from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common import BaseNode, Graph
+from model_compression_toolkit.core.common.graph.functional_node import FunctionalNode
 from model_compression_toolkit.core.pytorch.constants import *
 from model_compression_toolkit.logger import Logger
@@ -37,9 +38,12 @@ class FunctionalBatchNorm(common.BaseSubstitution):
         super().__init__(matcher_instance=bn_node)
     @staticmethod
-    def get_attributes_from_weights(node: BaseNode) -> Dict:
+    def get_attributes_from_weights(node: FunctionalNode) -> Dict:
         """
-        convert functional batch_norm positional weights to BatchNorm2d weights
+        Convert functional batch_norm positional weights to BatchNorm2d weights. Extract indices of gamma
+        and beta according to tensor_input_allocs if they were input as kwargs. If they were input as args,
+        use their fixed positions.
         Args:
             node: functional batch_norm node.
@@ -53,23 +57,22 @@ class FunctionalBatchNorm(common.BaseSubstitution):
                         GAMMA: np.ones(node.weights[1].shape),
                         BETA: np.zeros(node.weights[1].shape)}
-        has_weight = WEIGHT not in node.framework_attr
-        has_bias = BIAS not in node.framework_attr
+        # Check if weight and/or bias were not given.
+        if KERNEL in node.tensor_input_allocs:
+            weights_dict[GAMMA] = node.weights[node.tensor_input_allocs.index(KERNEL)]
+        elif KERNEL not in node.op_call_kwargs:
+            weights_dict[GAMMA] = node.weights[3]
-        if 3 in node.weights:
-            if has_weight:
-                weights_dict[GAMMA] = node.weights[3]
-            else:
-                weights_dict[BETA] = node.weights[3]
-        if 4 in node.weights:
-            assert has_bias
+        if BIAS in node.tensor_input_allocs:
+            weights_dict[BETA] = node.weights[node.tensor_input_allocs.index(BIAS)]
+        elif BIAS not in node.op_call_kwargs:
             weights_dict[BETA] = node.weights[4]
         return weights_dict
     def substitute(self,
                    graph: Graph,
-                   node: BaseNode) -> Graph:
+                   node: FunctionalNode) -> Graph:
         """
         Substitute functional.batch_norm and its inputs with BatchNorm2d.
         Args:
@@ -87,10 +90,13 @@ class FunctionalBatchNorm(common.BaseSubstitution):
         bn_node_weights = self.get_attributes_from_weights(node)
         if not bn_node_weights:
             return graph
+        framework_attr = {NUM_FEATURES: out_channels}
+        if EPSILON in node.op_call_kwargs:
+            framework_attr.update({EPSILON: node.op_call_kwargs[EPSILON]})
+        if MOMENTUM in node.op_call_kwargs:
+            framework_attr.update({MOMENTUM: node.op_call_kwargs[MOMENTUM]})
         new_batchnorm2d = BaseNode(name=node.name + '_into_BatchNorm2d',
-                                   framework_attr={NUM_FEATURES: out_channels,
-                                                   EPSILON: EPSILON_VAL,
-                                                   MOMENTUM: MOMENTUM_VAL},
+                                   framework_attr=framework_attr,
                                    input_shape=node.output_shape,
                                    output_shape=node.output_shape,
                                    weights=bn_node_weights,

{mct-nightly-2.1.0.20240623.439 → mct-nightly-2.1.0.20240625.423}/model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/functional_layer_norm.py RENAMED Viewed

@@ -21,6 +21,7 @@ from typing import Dict, Tuple, List
 from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
 from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common import BaseNode, Graph
+from model_compression_toolkit.core.common.graph.functional_node import FunctionalNode
 from model_compression_toolkit.core.pytorch.constants import *
 from model_compression_toolkit.logger import Logger
@@ -38,9 +39,11 @@ class FunctionalLayerNorm(common.BaseSubstitution):
         super().__init__(matcher_instance=ln_node)
     @staticmethod
-    def get_attributes_from_weights(node: BaseNode, normalized_shape: [Tuple, List, int]) -> Dict:
+    def get_attributes_from_weights(node: FunctionalNode, normalized_shape: [Tuple, List, int]) -> Dict:
         """
-        Parse layer_norm(input, normalized_shape, weight=None, bias=None)
+        Convert functional layer_norm positional weights to LayerNorm weights. Extract indices of gamma
+        and beta according to tensor_input_allocs if they were input as kwargs. If they were input as args,
+        use their fixed positions.
         Args:
             node: Node that match the pattern in the substitution init.
             normalized_shape: nn.LayerNorm "normalized_shape" argument
@@ -50,28 +53,26 @@ class FunctionalLayerNorm(common.BaseSubstitution):
         """
         # Define default weight and bias
-        weights_dict = {GAMMA: np.ones(normalized_shape), # Default value in case weight is not given
-                        BETA: np.zeros(normalized_shape) # Default value in case bias is not given
+        weights_dict = {GAMMA: np.ones(normalized_shape),  # Default value in case weight is not given
+                        BETA: np.zeros(normalized_shape)  # Default value in case bias is not given
                         }
         # Check if weight and/or bias were not given.
-        has_weight = WEIGHT not in node.framework_attr
-        has_bias = BIAS not in node.framework_attr
+        if KERNEL in node.tensor_input_allocs:
+            weights_dict[GAMMA] = node.weights[node.tensor_input_allocs.index(KERNEL)]
+        elif KERNEL not in node.op_call_kwargs:
+            weights_dict[GAMMA] = node.weights[1]
-        if 1 in node.weights:
-            if has_weight:
-                weights_dict[GAMMA] = node.weights[1]
-            else:
-                weights_dict[BETA] = node.weights[1]
-        if 2 in node.weights:
-            assert has_bias
+        if BIAS in node.tensor_input_allocs:
+            weights_dict[BETA] = node.weights[node.tensor_input_allocs.index(BIAS)]
+        elif BIAS not in node.op_call_kwargs:
             weights_dict[BETA] = node.weights[2]
         return weights_dict
     def substitute(self,
                    graph: Graph,
-                   node: BaseNode) -> Graph:
+                   node: FunctionalNode) -> Graph:
         """
         Substitute functional.layer_norm and its inputs with LayerNorm.
         Args:
@@ -85,10 +86,11 @@ class FunctionalLayerNorm(common.BaseSubstitution):
         ln_node_weights = self.get_attributes_from_weights(node, normalized_shape)
+        framework_attr = {NORMALIZED_SHAPE: normalized_shape}
+        if EPSILON in node.op_call_kwargs:
+            framework_attr.update({EPSILON: node.op_call_kwargs[EPSILON]})
         new_layernorm = BaseNode(name=node.name + '_into_LayerNorm',
-                                 framework_attr={NORMALIZED_SHAPE: normalized_shape,
-                                                 EPSILON: node.framework_attr.get('eps'),
-                                                 },
+                                 framework_attr=framework_attr,
                                  input_shape=node.output_shape,
                                  output_shape=node.output_shape,
                                  weights=ln_node_weights,

mct-nightly 2.1.0.20240623.439__tar.gz → 2.1.0.20240625.423__tar.gz

mct-nightly 2.1.0.20240623.439tar.gz → 2.1.0.20240625.423tar.gz