Reputation: 11
I'm doing Post Training Aware, and want to get type and weight of the activation of Conv1D, like 'relu'.
So I use tf.lite.experimental.Analyzer.analyze(model_content=tflite_model_quant)
to check, but I don't see anything about activation. How can I get it?
Analyzer Output
T#18(sequential_3/quant_conv1d_13/Conv1D) shape:[16, 1, 16, 4], type:INT8 RO 1024 bytes, buffer: 19, data:[., ., ., ., %, ...]
T#19(sequential_3/quant_conv1d_12/BiasAdd/ReadVariableOp) shape:[4], type:INT32 RO 16 bytes, buffer: 20, data:[-220, 862, 1177, -1029]
T#20(sequential_3/quant_conv1d_12/Conv1D) shape:[4, 1, 16, 1], type:INT8 RO 64 bytes, buffer: 21, data:[., :, ., ., ., ...]
T#21(sequential_3/quant_conv1d_12/Conv1D/ExpandDims) shape_signature:[-1, 1, 256, 1], type:INT8
T#22(sequential_3/quant_conv1d_12/Relu;sequential_3/quant_conv1d_12/BiasAdd;sequential_3/quant_conv1d_12/Conv1D/Squeeze;sequential_3/quant_conv1d_12/BiasAdd/ReadVariableOp;sequential_3/quant_conv1d_12/Conv1D) shape_signature:[-1, 1, 256, 4], type:INT8
T#23(sequential_3/quant_conv1d_12/Relu;sequential_3/quant_conv1d_12/BiasAdd;sequential_3/quant_conv1d_12/Conv1D/Squeeze;sequential_3/quant_conv1d_12/BiasAdd/ReadVariableOp) shape_signature:[-1, 256, 4], type:INT8
My Model
def build_model():
model = tf.keras.models.Sequential([
tf.keras.layers.Conv1D(4, 16, strides= 1,padding='same', activation= 'relu'),
tf.keras.layers.MaxPooling1D(pool_size=3, strides=2, padding='same'),
tf.keras.layers.Conv1D(16, 16, strides=1, padding='same', activation='relu'),
tf.keras.layers.MaxPooling1D(pool_size=3, strides=2, padding='same'),
tf.keras.layers.Conv1D(32, 16, strides=1, padding='same', activation='relu'),
tf.keras.layers.AveragePooling1D(pool_size=3, strides=2, padding='same'),
tf.keras.layers.Conv1D(64, 16, strides=1, padding= 'same', activation='relu'),
tf.keras.layers.MaxPooling1D(pool_size=3, strides=2, padding='same'),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
#tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(15, activation='softmax')
])
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['acc'])
return model
Upvotes: 0
Views: 18