Reshape error with tflite model about tensorflowasr HOT 4 CLOSED

After I update the tf-nightly to the most recent version, I got following error:

RuntimeError: tensorflow/lite/kernels/reshape.cc:55 stretch_dim != -1 (0 != -1)Node number 187 (RESHAPE) failed to prepare.

from tensorflowasr.

It's because tflite model doesn't support tensor[index] (it returns a tensor of 0 elements instead of actual tensor of encoder, which has 144 elements), so I have to use tf.gather_nd instead. I fixed this issue in master :D
More specific:

...
def body(time, total, encoded, hypothesis):
    ytu, new_states = self.decoder_inference(
    # avoid using [index] in tflite
    encoded=tf.gather_nd(encoded, tf.expand_dims(time, axis=-1)),
    predicted=hypothesis.prediction.read(hypothesis.index),
    states=hypothesis.states
)
...

from tensorflowasr.

It is all good now, thx!
The inference speed is a little bit slow,
but it is an amazing project anyway!

from tensorflowasr.

It is all good now, thx!
The inference speed is a little bit slow,
but it is an amazing project anyway!

The beam search implementation is single thread so it's slow. I'm finding a way to improve it :D

from tensorflowasr.