Nondeterministic behavior with tf.concat on zipped tensors: Eager sometimes works, sometimes fails with InvalidArgumentError

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

binary

TensorFlow version

2.20.0

Custom code

Yes

OS platform and distribution

linux ubuntu 24.04

Mobile device

No response

Python version

3.9

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

CUDA Version 13.0

GPU model and memory

No response

Current behavior?

I encountered a nondeterministic behavior in TensorFlow 2.20.0
The same script produces two completely different results on two consecutive runs:

Run 1:

Eager mode: runs successfully
XLA compile mode (jit_compile=True): fails with OperatorNotAllowedInGraphError
Run 2:

Eager mode: fails with InvalidArgumentError (ConcatV2 kernel type mismatch)
Error does not appear in Run 1
This indicates nondeterministic execution paths inside AutoGraph/XLA when Python
builtins (filter/map/zip) are used with Tensors.

The model uses zip() to combine two Tensor lists, and tf.concat() on the resulting list.
In the first run, TensorFlow internally coerces the zipped tuples into tensors.
In the second run, TensorFlow attempts to run ConcatV2 with an incorrect dtype (DT_QINT32),
leading to:

Expected behavior:

Eager mode should be deterministic and should either always succeed or always raise a type error.
Kernel dtype should never unexpectedly switch to DT_QINT32.
Graph/XLA mode should fail with a stable error message, not vary between runs.
This inconsistent behavior may indicate nondeterministic paths in AutoGraph or graph construction.

Standalone code to reproduce the issue

import tensorflow as tf

class TestModel(tf.keras.Model):

    def __init__(self):
        super().__init__()
        self.d1 = tf.keras.layers.Dense(32, activation='relu')
        self.d2 = tf.keras.layers.Dense(16, activation='tanh')
        self.d3 = tf.keras.layers.Dense(8)

    def call(self, x):
        filtered_features = list(filter(lambda z: tf.reduce_sum(z) > 0.5, [x, x * 2, x * 3]))
        mapped_features = list(map(lambda z: tf.nn.sigmoid(z), filtered_features))
        zipped_data = list(zip(mapped_features, [tf.ones_like(x) for _ in range(len(mapped_features))]))
        combined = tf.concat(zipped_data, axis=-1)
        return self.d3(combined)

def get_default_model():
    return TestModel()

def get_sample_inputs():
    x = tf.random.normal([4, 16])
    return (x,)

def main():
    model = get_default_model()
    inputs = get_sample_inputs()
    eager_out = model(*inputs)
    print('Eager Input shape:', inputs[0].shape)
    print('Eager Output shape:', eager_out.shape)
    @tf.function(jit_compile=True)
    def compiled_forward(*args):
        return model(*args)
    compiled_out = compiled_forward(*inputs)
    print('XLA Output shape:', compiled_out.shape)

if __name__ == '__main__':
    main()
‵‵‵

Relevant log output

- Run 1:
    - Eager mode: runs successfully
    - XLA compile mode (jit_compile=True): fails with OperatorNotAllowedInGraphError

    OperatorNotAllowedInGraphError: Exception encountered when calling TestModel.call().
    
    Using a symbolic `tf.Tensor` as a Python `bool` is not allowed. You can attempt the following resolutions to the problem: If you are running in Graph mode, use Eager execution mode or decorate this function with @tf.function. If you are using AutoGraph, you can try decorating this function with @tf.function. If that does not work, then you may be using an unsupported feature or your source code may not be visible to AutoGraph. See https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/autograph/g3doc/reference/limitations.md#access-to-source-code for more information.
    
    Arguments received by TestModel.call():
      • x=tf.Tensor(shape=(4, 16), dtype=float32)

- Run 2:
    - Eager mode: fails with InvalidArgumentError (ConcatV2 kernel type mismatch)
    - Error does not appear in Run 1

tensorflow.python.framework.errors_impl.InvalidArgumentError: Exception encountered when calling TestModel.call().

OpKernel 'ConcatV2' has constraint on attr 'T' not in NodeDef '[N=0, Tidx=DT_INT32]', KernelDef: 'op: "ConcatV2" device_type: "CPU" constraint { name: "T" allowed_values { list { type: DT_QINT32 } } } host_memory_arg: "axis"' [Op:ConcatV2] name: concat

Arguments received by TestModel.call():
  • x=tf.Tensor(shape=(4, 16), dtype=float32)