Replace the dynamic int8 file with a weight-only int8 file

README.md

@@ -51,10 +51,20 @@ The license information of the original model was missing.
 | File | Description |
 | --- | --- |
 | `mobilenet_v3_small.tflite` | Full precision LiteRT/TFLite model. |
-| `mobilenet_v3_small_dynamic_wi8_afp32.tflite` | Dynamic weight-only INT8 model with FP32 activations. |
+| `mobilenet_v3_small_weight_only_wi8_afp32.tflite` | Weight-only INT8 model with FP32 activations. |
 | `mobilenet_v3_small_Google_Tensor_G5_apply_plugin.tflite` | AOT-compiled artifact for the Google Tensor G5 target. |
 | `mobilenet_v3_small_int8_channelwise.tflite` | Static INT8 model with channelwise INT8 weights and asymmetric INT8 activations. |
 
+### Quantized variant
+
+`mobilenet_v3_small_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 MobileNetV3's SE and hard-swish 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 0.991.
+
 ## Quantization
 
 `mobilenet_v3_small_int8_channelwise.tflite` was produced with the `STATIC_WI8_AI8` quantization recipe. Weights are signed INT8 and use symmetric channelwise quantization for weight tensors. Activations are signed INT8 with asymmetric quantization parameters.

mobilenet_v3_small_dynamic_wi8_afp32.tflite → mobilenet_v3_small_weight_only_wi8_afp32.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a93f73c57e09b1f0224ee9965ae92af6c7da16373b2b02f9db232721f5b0ced6
-size 2741424
+oid sha256:04f7ee3ab9a50dc7c828f82d040712ab0b85f0d54c13258b885482038005b5a8
+size 2746704