Converting Model to TFLite using Tensorflow 2.16.1 fails - works w/ 2.15.0 #63987
Comments
|
It seems like you might have a mix of |
|
@gsexton, |
tilakrayal
added
the
stat:awaiting response
|
@tilakrayal The model is built via: model = tf.keras.Sequential([
tf.keras.layers.Dense(1024, activation='relu'),
tf.keras.layers.Dense(2048, activation='relu'),
tf.keras.layers.Dense(2048, activation='relu'),
tf.keras.layers.Dense(1, activation='sigmoid')
])
model.compile(
loss=tf.keras.losses.binary_crossentropy,
optimizer=tf.keras.optimizers.Adam(learning_rate=0.003),
metrics=[
tf.keras.metrics.BinaryAccuracy(name='accuracy'),
tf.keras.metrics.Precision(name='precision'),
tf.keras.metrics.Recall(name='recall')
]
)
history = model.fit(X_train, y_train, epochs=NUM_TRAINING_EPOCHS)the convert is: converter = tf.lite.TFLiteConverter.from_keras_model(best_model)so it seems like this should be OK. My first effort for conversion was: tf.saved_model.save(best_model,save_path)
converter = tf.lite.TFLiteConverter.from_saved_model(save_path)
tflite_model = converter.convert()That too gives: @Aloqeely I'm not sure I follow what to do with this: Also, please try to execute the model code with the tf-keras, as the tensorflow v2.16 by default uses keras 3.0 version. Thank you! I think it is already doing what you suggest. Using tf.keras. |
google-ml-butler
bot
removed
the
stat:awaiting response
|
Here is a complete simple program and data set that replicates the problem. You can alternate paths by |
|
Hi @gsexton, I have reproduced the code with TF 2.15 and TF 2.16. With 2.15 it's working fine as expected. As you mentioned, it's getting aborted with TF2.16.1 at TFLite conversion . Right now, TF2.16 has an issue with Keras 3.0. As a workaround install Keras 2 as follows. Thank You |
LakshmiKalaKadali
added
the
stat:awaiting response
|
I have tested with the tf_keras installed, and the problem is resolved. |
|
I also experienced this bug, and it did not disappear with #!/usr/bin/env python3
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Input, DepthwiseConv2D, Conv2D, Flatten, Dense
# Define a simple model with depthwise convolutions
model = Sequential()
model.add(Input(shape=(64, 64, 3)))
model.add(DepthwiseConv2D(kernel_size=(3, 3), activation="relu"))
model.add(Conv2D(64, (1, 1), activation="relu"))
model.add(Flatten())
model.add(Dense(100, activation="relu"))
model.add(Dense(10, activation="softmax"))
# Compile the model
model.compile(optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"])
# Define the representative dataset generator
def representative_dataset_gen():
for _ in range(100):
# Here you should provide a sample from your actual dataset
# For illustration, we'll use random data
# Yield a batch of input data (in this case, a single sample)
yield [tf.random.normal((1, 64, 64, 3))]
# Set the TensorFlow Lite converter
converter = tf.lite.TFLiteConverter.from_keras_model(model)
# Set the optimization to default for int8 quantization
converter.optimizations = [tf.lite.Optimize.DEFAULT]
# Define the representative dataset for quantization
converter.representative_dataset = representative_dataset_gen
# Restrict the target spec to int8 for full integer quantization
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
# Instruct the converter to make the input and output layer as integer
converter.inference_input_type = tf.int8
converter.inference_output_type = tf.int8
# Convert the model
tflite_model = converter.convert()
# Save the model to a file
with open("model.tflite", "wb") as f:
f.write(tflite_model)
print("Model has been successfully converted to TFLite and saved as 'model.tflite'.")This crashes with the error: |
|
Is there any chance we can fix this issue at the root cause? We found the same problem when saving a keras model with |
|
I have the same problem; In TF2.16 converting a .keras model to .tflite gives |
|
Same issue. |
|
Same issue here. Try: pip install -U tf_keras # Keras 2 but didn't work. Saving to tflite... |
Have you made sure to set the env variable before importing the tflow module? this fixed it for me |
Do you have an example how to do it? @filipang |
1. System information
2. Code
3. Failure after conversion
5. (optional) Any other info / logs
This issue was reported on #62610 originally.
I've tested this on Linux AMD 64, and on Mac OS 14.4 ARM and the same error occurs.
Per the comment from @adamantivm, I tried using Tensorflow 2.15.0 on Linux AMD64, and it worked as expected.
I tried setting the env variable for MLIR_CRASH... and running, but no output was found.
The text was updated successfully, but these errors were encountered: