Falied to load gru layer weights to gru cell, Layer 'gru_cell' expected 3 variables, but received 0 variables during loading

[victorVoice]

Using tensorflow 2.16.1 with keras 3.5.0 falied to load pretrained gru layers weights to a gru cell.
the tow layer are defined as below

For gru layers:
t_rnn_1 = keras.layers.GRU(units=64, return_sequences=True)(t_in_1)
t_rnn_2 = keras.layers.GRU(units=64, return_sequences=True)(t_rnn_1)
t_dense_c = keras.layers.Dense(80)(t_rnn_2)
t_dense_c = tf.keras.layers.ReLU(max_value=6.)(t_dense_c)

For gru cells:
t_rnn_1, cell_out1 = keras.layers.GRUCell(units=64)(t_in_1, states=cell_in1)
t_rnn_2, cell_out2 = keras.layers.GRUCell(units=64)(t_rnn_1, states=cell_in2)
t_dense_2= keras.layers.Dense(80)(t_rnn_2)
t_dense_2 = tf.keras.layers.ReLU(max_value=6.)(t_dense_2)

when loading got flowing error message

Traceback (most recent call last):
File "/home/victoryu/project/se_tf/subband_model_streaming.py", line 319, in
tf_model.create_tf_lite_model(weights_file=args.ckpt, target_name='./crn_cplx')
File "/home/victoryu/project/se_tf/subband_model_streaming.py", line 128, in create_tf_lite_model
self.model.load_weights(weights_file)
File "/home/victoryu/miniconda3/envs/tf2.16/lib/python3.11/site-packages/keras/src/utils/traceback_utils.py", line 122, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/home/victoryu/miniconda3/envs/tf2.16/lib/python3.11/site-packages/keras/src/saving/saving_lib.py", line 593, in _raise_loading_failure
raise ValueError(msg)
ValueError: A total of 2 objects could not be loaded. Example error message for object :

Layer 'gru_cell' expected 3 variables, but received 0 variables during loading. Expected: ['kernel', 'recurrent_kernel', 'bias']

It works fine when using tensorflow 2.13.0 + keras 2.13.1.

When visulize the weight.h5 file the differnce between to layers are as below

wonder is the cause the problem, and how to fix it.

[mehtamansi29]

Hi @victorVoice -

Thanks for reporting the issue. Can you help me what you defined here t_in_1 or any full sample code for both gru or gru_cell layer ?

[victorVoice]

@mehtamansi29 Sure,
t_in_1 for gru layer is a tensor with shape [batchsize, time_steps, feature_dims] in the acture model is like a tensor with shape [32, 63, 80]
for gru_cell layer, since it process 1 time step at each time the t_in_1 here is [batch_siz，feature_dims], like [32, 80]

here is some sample code for gru cell layer

inp = keras.Input(batch_shape=(1, 5, 16))
cell_in1 = keras.Input(batch_shape=(1, 64))
cell_in2 = keras.Input(batch_shape=(1, 64))

t_in_1 = keras.layers.Reshape([5 * 16])(inp)
t_rnn_1, cell_out1 = keras.layers.GRUCell(units=64)(t_in_1, states=cell_in1)
t_rnn_2, cell_out2 = keras.layers.GRUCell(units=64)(t_rnn_1, states=cell_in2)
t_dense_2= keras.layers.Dense(80)(t_rnn_2)
t_dense_2 = tf.keras.layers.ReLU(max_value=6.)(t_dense_2)
s2_out = keras.layers.Reshape([1, 5, 16])(t_dense_2)

here is the code for gru layers

inp = keras.Input(batch_shape=(32, 63, 5, 16))

t_in_1 = keras.layers.Reshape([64, 5 * 16])(inp)
t_rnn_1 =  keras.layers.GRU(units=64, return_sequences=True)(t_in_1)
t_rnn_2 =  keras.layers.GRU(units=64, return_sequences=True)(t_rnn_1)
t_dense_2= keras.layers.Dense(80)(t_rnn_2)
t_dense_2 = tf.keras.layers.ReLU(max_value=6.)(t_dense_2)
s2_out = keras.layers.Reshape([1, 63, 5, 16])(t_dense_2)

[mehtamansi29]

Hi @victorVoice -

Thanks for the sample code. I replicate the sample code with GRU layer or GRU_cell in latest keras(3.6.0) and it is working fine for me.
Attached gist for the reference.

[victorVoice]

@mehtamansi29 Thanks i will try keras(3.6.0) first, thx for the help.