mct-nightly 1.10.0.20231128.post5436__py3-none-any.whl → 1.10.0.20231129.post414__py3-none-any.whl

{mct_nightly-1.10.0.20231128.post5436.dist-info → mct_nightly-1.10.0.20231129.post414.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 1.10.0.20231128.post5436
+Version: 1.10.0.20231129.post414
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN

{mct_nightly-1.10.0.20231128.post5436.dist-info → mct_nightly-1.10.0.20231129.post414.dist-info}/RECORD

@@ -168,7 +168,7 @@ model_compression_toolkit/core/keras/graph_substitutions/substitutions/softmax_s
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/virtual_activation_weights_composition.py,sha256=wH9ocMLL725-uUPU-zCxdd8NwT5nyd0ZShmI7iuTwF8,1462
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/weights_activation_split.py,sha256=rjIheZW7LbSPv9bzMSmC8wl6UUxaTkd4J2IHinObT-Y,1814
 model_compression_toolkit/core/keras/hessian/__init__.py,sha256=lNJ29DYxaLUPDstRDA1PGI5r9Fulq_hvrZMlhst1Z5g,697
-model_compression_toolkit/core/keras/hessian/activation_trace_hessian_calculator_keras.py,sha256=tj3GtzMY75Wumj2j-teFqlN4EzgUgupCmjtoEVKvpv4,9665
+model_compression_toolkit/core/keras/hessian/activation_trace_hessian_calculator_keras.py,sha256=RL2c7JkvdK37Mgzq2CdSS_1Cg0R1cJ0Gv0Q3cRgV4H4,9642
 model_compression_toolkit/core/keras/hessian/trace_hessian_calculator_keras.py,sha256=uwDcC6Fr0H-NKaX0NERRkF6rrnPRbxuo9PsapnuPsRo,3952
 model_compression_toolkit/core/keras/hessian/weights_trace_hessian_calculator_keras.py,sha256=SIi7_wChifkulp6aoc-v7v4I1NrQAwvU4ymG2HI66Yk,10478
 model_compression_toolkit/core/keras/mixed_precision/__init__.py,sha256=sw7LOPN1bM82o3SkMaklyH0jw-TLGK0-fl2Wq73rffI,697
@@ -226,7 +226,7 @@ model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/softmax
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/virtual_activation_weights_composition.py,sha256=WmEa8Xjji-_tIbthDxlLAGSr69nWk-YKcHNaVqLa7sg,1375
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/weights_activation_split.py,sha256=tp78axmUQc0Zpj3KwVmV0PGYHvCf7sAW_sRmXXw7gsY,1616
 model_compression_toolkit/core/pytorch/hessian/__init__.py,sha256=lNJ29DYxaLUPDstRDA1PGI5r9Fulq_hvrZMlhst1Z5g,697
-model_compression_toolkit/core/pytorch/hessian/activation_trace_hessian_calculator_pytorch.py,sha256=ZrBQTpBclUpsOKnmSkgGG1y4Rs1OjBaf93Fcsn6trUg,8340
+model_compression_toolkit/core/pytorch/hessian/activation_trace_hessian_calculator_pytorch.py,sha256=6EHwxuOdb57DEBHiYfFODCQHWlUQDArEHLaym19Ml5A,8297
 model_compression_toolkit/core/pytorch/hessian/trace_hessian_calculator_pytorch.py,sha256=c80LiFbY06mosfZI4yjigiqDsF1ZFdRZ14Xox0kD_94,3480
 model_compression_toolkit/core/pytorch/hessian/weights_trace_hessian_calculator_pytorch.py,sha256=4TsL-27I672758_B742Zbe-_Ro4OLp7ISxWtWD4oKD8,6899
 model_compression_toolkit/core/pytorch/mixed_precision/__init__.py,sha256=Rf1RcYmelmdZmBV5qOKvKWF575ofc06JFQSq83Jz99A,696
@@ -447,8 +447,8 @@ 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-1.10.0.20231128.post5436.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-mct_nightly-1.10.0.20231128.post5436.dist-info/METADATA,sha256=xgYuHWMlREV_QtNvSqQmQD82x8tSx3Ht5SlNk8MoY6A,16233
-mct_nightly-1.10.0.20231128.post5436.dist-info/WHEEL,sha256=oiQVh_5PnQM0E3gPdiz09WCNmwiHDMaGer_elqB3coM,92
-mct_nightly-1.10.0.20231128.post5436.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-1.10.0.20231128.post5436.dist-info/RECORD,,
+mct_nightly-1.10.0.20231129.post414.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+mct_nightly-1.10.0.20231129.post414.dist-info/METADATA,sha256=BlcFLeMfxtrdsyKQxgLa5xkorVHJlwwK7qGId7PKegQ,16232
+mct_nightly-1.10.0.20231129.post414.dist-info/WHEEL,sha256=oiQVh_5PnQM0E3gPdiz09WCNmwiHDMaGer_elqB3coM,92
+mct_nightly-1.10.0.20231129.post414.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-1.10.0.20231129.post414.dist-info/RECORD,,

model_compression_toolkit/core/keras/hessian/activation_trace_hessian_calculator_keras.py

@@ -152,7 +152,7 @@ class ActivationTraceHessianCalculatorKeras(TraceHessianCalculatorKeras):
                     # Compute the final approximation for each output index
                     num_node_outputs = len(interest_point_scores[0])
                     for output_idx in range(num_node_outputs):
-                        final_approx_per_output.append(2 * tf.reduce_mean([x[output_idx] for x in interest_point_scores]) / output.shape[-1])
+                        final_approx_per_output.append(tf.reduce_mean([x[output_idx] for x in interest_point_scores]))
 
                     # final_approx_per_output is a list of all approximations (one per output), thus we average them to
                     # get the final score of a node.

model_compression_toolkit/core/pytorch/hessian/activation_trace_hessian_calculator_pytorch.py

@@ -131,8 +131,8 @@ class ActivationTraceHessianCalculatorPytorch(TraceHessianCalculatorPytorch):
                             break
 
                     trace_hv.append(hessian_trace_approx)
-                ipts_hessian_trace_approx.append(2 * torch.mean(torch.stack(trace_hv)) / output.shape[
-                    -1])  # Get averaged Hessian trace approximation
+
+                ipts_hessian_trace_approx.append(torch.mean(torch.stack(trace_hv)))  # Get averaged Hessian trace approximation
 
             # If a node has multiple outputs, it means that multiple approximations were computed
             # (one per output since granularity is per-tensor). In this case we average the approximations.