mct-nightly 1.10.0.20231002.post426__py3-none-any.whl → 1.10.0.20231004.post404__py3-none-any.whl

{mct_nightly-1.10.0.20231002.post426.dist-info → mct_nightly-1.10.0.20231004.post404.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 1.10.0.20231002.post426
+Version: 1.10.0.20231004.post404
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -130,6 +130,14 @@ Main features:
 * <ins>Visualization:</ins> You can use TensorBoard to observe useful information for troubleshooting the quantized model's performance (for example, the model in different phases of the quantization, collected statistics, similarity between layers of the float and quantized model and bit-width configuration for mixed-precision quantization). For more details, please read the [visualization documentation](https://sony.github.io/model_optimization/docs/guidelines/visualization.html).   
 * <ins>Target Platform Capabilities:</ins> The Target Platform Capabilities (TPC) describes the target platform (an edge device with dedicated hardware). For more details, please read the [TPC README](model_compression_toolkit/target_platform_capabilities/README.md).   
 
+### Enhanced Post-Training Quantization (EPTQ)
+As part of the GPTQ we provide an advanced optimization algorithm called EPTQ.
+
+The specifications of the algorithm are detailed in the paper: _"**EPTQ: Enhanced Post-Training Quantization via Label-Free Hessian**"_ [4].
+
+More details on the how to use EPTQ via MCT can be found in the [EPTQ guidelines](model_compression_toolkit/gptq/README.md).
+
+
 
 #### Experimental features 
 
@@ -176,4 +184,5 @@ MCT aims at keeping a more up-to-date fork and welcomes contributions from anyon
 
 [3] [TORCHVISION.MODELS](https://pytorch.org/vision/stable/models.html) 
 
+[4] Gordon, O., Habi, H. V., & Netzer, A., 2023. [EPTQ: Enhanced Post-Training Quantization via Label-Free Hessian. arXiv preprint](https://arxiv.org/abs/2309.11531)
 

{mct_nightly-1.10.0.20231002.post426.dist-info → mct_nightly-1.10.0.20231004.post404.dist-info}/RECORD

@@ -53,7 +53,7 @@ model_compression_toolkit/core/common/matchers/function.py,sha256=kMwcinxn_PInve
 model_compression_toolkit/core/common/matchers/node_matcher.py,sha256=63cMwa5YbQ5LKZy8-KFmdchVc3N7mpDJ6fNDt_uAQsk,2745
 model_compression_toolkit/core/common/matchers/walk_matcher.py,sha256=xqfLKk6xZt72hSnND_HoX5ESOooNMypb5VOZkVsJ_nw,1111
 model_compression_toolkit/core/common/mixed_precision/__init__.py,sha256=sw7LOPN1bM82o3SkMaklyH0jw-TLGK0-fl2Wq73rffI,697
-model_compression_toolkit/core/common/mixed_precision/bit_width_setter.py,sha256=u8sbqlnTnMHZ4Oox1iLpqylAgqiaawHsew0Hw-9Upb8,6704
+model_compression_toolkit/core/common/mixed_precision/bit_width_setter.py,sha256=_qwE1RlvDx4eGUfxpFHfM1Jo1pA6gSUUrswdgfs6YU8,6774
 model_compression_toolkit/core/common/mixed_precision/configurable_quant_id.py,sha256=LLDguK7afsbN742ucLpmJr5TUfTyFpK1vbf2bpVr1v0,882
 model_compression_toolkit/core/common/mixed_precision/configurable_quantizer_utils.py,sha256=kmyBcqGh3qYqo42gIZzouQEljTNpF9apQt6cXEVkTQ0,3871
 model_compression_toolkit/core/common/mixed_precision/distance_weighting.py,sha256=x0cweemRG3_7FlvAbxFK5Zi77qpoKAGqtGndY8MtgwM,2222
@@ -429,8 +429,8 @@ model_compression_toolkit/trainable_infrastructure/keras/quantize_wrapper.py,sha
 model_compression_toolkit/trainable_infrastructure/keras/quantizer_utils.py,sha256=MVwXNymmFRB2NXIBx4e2mdJ1RfoHxRPYRgjb1MQP5kY,1797
 model_compression_toolkit/trainable_infrastructure/pytorch/__init__.py,sha256=huHoBUcKNB6BnY6YaUCcFvdyBtBI172ZoUD8ZYeNc6o,696
 model_compression_toolkit/trainable_infrastructure/pytorch/base_pytorch_quantizer.py,sha256=SbvRlIdE32PEBsINt1bhSqvrKL_zbM9V-aeSkOn-sw4,3083
-mct_nightly-1.10.0.20231002.post426.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-mct_nightly-1.10.0.20231002.post426.dist-info/METADATA,sha256=kR9pwPuw6SdQM3s7VF3e6XE8uoc4EVXAXCunQe3S_5g,15741
-mct_nightly-1.10.0.20231002.post426.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
-mct_nightly-1.10.0.20231002.post426.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-1.10.0.20231002.post426.dist-info/RECORD,,
+mct_nightly-1.10.0.20231004.post404.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+mct_nightly-1.10.0.20231004.post404.dist-info/METADATA,sha256=6imuKBIiVkvsgOisTy671wf6-OChPZOr7D8ai_J2sVo,16303
+mct_nightly-1.10.0.20231004.post404.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
+mct_nightly-1.10.0.20231004.post404.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-1.10.0.20231004.post404.dist-info/RECORD,,

model_compression_toolkit/core/common/mixed_precision/bit_width_setter.py

@@ -50,20 +50,21 @@ def set_bit_widths(mixed_precision_enable: bool,
                 _set_node_final_qc(bit_widths_config,
                                    node,
                                    node_index_in_graph)
-            elif node.is_activation_quantization_enabled():
-                # If we are here, this means that we are in weights-only mixed-precision
-                # (i.e., activations are quantized with fixed bitwidth or not quantized)
-                # and that this node doesn't have weights to quantize
-                assert len(node.candidates_quantization_cfg) > 0, \
-                    "Node need to have at least one quantization configuration in order to quantize its activation"
-                node.final_activation_quantization_cfg = copy.deepcopy(node.candidates_quantization_cfg[0].activation_quantization_cfg)
-            elif node.is_weights_quantization_enabled():
-                # If we are here, this means that we are in activation-only mixed-precision
-                # (i.e., weights are quantized with fixed bitwidth or not quantized)
-                # and that this node doesn't have activations to quantize
-                assert len(node.candidates_quantization_cfg) > 0, \
-                    "Node need to have at least one quantization configuration in order to quantize its activation"
-                node.final_weights_quantization_cfg = copy.deepcopy(node.candidates_quantization_cfg[0].weights_quantization_cfg)
+            else:
+                if node.is_activation_quantization_enabled():
+                    # If we are here, this means that we are in weights-only mixed-precision
+                    # (i.e., activations are quantized with fixed bitwidth or not quantized)
+                    # and that this node doesn't have weights to quantize
+                    assert len(node.candidates_quantization_cfg) > 0, \
+                        "Node need to have at least one quantization configuration in order to quantize its activation"
+                    node.final_activation_quantization_cfg = copy.deepcopy(node.candidates_quantization_cfg[0].activation_quantization_cfg)
+                if node.is_weights_quantization_enabled():
+                    # If we are here, this means that we are in activation-only mixed-precision
+                    # (i.e., weights are quantized with fixed bitwidth or not quantized)
+                    # and that this node doesn't have activations to quantize
+                    assert len(node.candidates_quantization_cfg) > 0, \
+                        "Node need to have at least one quantization configuration in order to quantize its activation"
+                    node.final_weights_quantization_cfg = copy.deepcopy(node.candidates_quantization_cfg[0].weights_quantization_cfg)
 
     # When working in non-mixed-precision mode, there's only one bitwidth, and we simply set the
     # only candidate of the node as its final weight and activation quantization configuration.