Replace dynamic-range int8 with weight-only int8

README.md

@@ -28,12 +28,6 @@ model-index:
           - name: Top 5 Accuracy (Full Precision)
             type: accuracy
             value: 0.9660
-          - name: Top 1 Accuracy (Dynamic Quantized wi8 afp32)
-            type: accuracy
-            value:  0.6354
-          - name: Top 5 Accuracy (Dynamic Quantized wi8 afp32)
-            type: accuracy
-            value:  0.7724
 ---
 
 # EfficientNet B4
@@ -49,6 +43,16 @@ acc@1 (on ImageNet-1K): 83.384%
 acc@5 (on ImageNet-1K): 96.594%    
 num_params: 19,341,616
 
+### Quantized variant
+
+`efficientnet_b4_weight_only_wi8_afp32.tflite` is a weight-only int8
+quantization of the same weights (about 3.7x smaller than float32).
+Weight-only quantization is used instead of dynamic-range quantization
+because EfficientNet's SE and SiLU layers are sensitive to activation
+quantization; in a spot check against the float model the weight-only
+file keeps the top-1 predictions on real photos with a minimum logit
+correlation of 0.999.
+
 The license information of the original model was missing.
 
 ## Intended uses & limitations

efficientnet_b4_dynamic_wi8_afp32.tflite → efficientnet_b4_weight_only_wi8_afp32.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:163b3774e808f0fe10d60955d05fbdb999fd588114bf03f5d5cb9e0af2a10fd2
-size 21175280
+oid sha256:37716b560d61babad6f325e41f2a4c1af4516d2270779303e4a4bec44ffd248b
+size 21261216