tensorflow-upstream: Error "Missing metadata for __global__ function"

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): Nope, entry level sample code. Code: https://pastebin.com/VyLQBsHi
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Mint 19.3
• TensorFlow installed from (source or binary): Binary
• TensorFlow version (use command below):

apt policy rocm-dkms
rocm-dkms:
  Installed: 3.3.0-19

pip3 list --format=columns | grep tens
tensorboard           2.1.1          
tensorflow-estimator  2.1.0          
tensorflow-rocm       2.1.1 

• Python version:

apt policy python3
python3:
  Installed: 3.6.7-1~18.04

• Bazel version (if compiling from source): NA
• GCC/Compiler version (if compiling from source): NA
• CUDA/cuDNN version: NA
• GPU model and memory: Red Devil Radeon RX 5700 XT 8GB

Scripts output (tried updating it to python3 and pip3): https://pastebin.com/cXKw9gBY

>>> print(tf.version.GIT_VERSION)
v2.1.0-22-g5f0b1a5
>>> print(tf.version.VERSION)
2.1.1

Describe the current behavior

train_loss = tf.keras.metrics.Mean(name='train_loss')

Returns:

terminate called after throwing an instance of 'std::runtime_error'
  what():  Missing metadata for __global__ function: _ZN5Eigen8internal15EigenMetaKernelINS_15TensorEvaluatorIKNS_14TensorAssignOpINS_9TensorMapINS_6TensorIbLi1ELi1EiEELi16ENS_11MakePointerEEEKNS_18TensorCwiseUnaryOpINS0_21scalar_boolean_not_opIbEEKNS4_INS5_IKbLi1ELi1EiEELi16ES7_EEEEEENS_9GpuDeviceEEEiEEvT_T0_
Aborted (core dumped)

Describe the expected behavior Not to crash. :) Sorry, new.

Standalone code to reproduce the issue Full source: https://pastebin.com/VyLQBsHi

Other info / logs Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached. Looks a lot like: https://github.com/ROCmSoftwarePlatform/tensorflow-upstream/issues/319