GPU Count Configuration with DeepSpeech Model Execution

I have deepspeech-gpu configured with following versions

DeepSpeech Version 0.9.3
Python 3.9.1
Tensorflow 2.3
libcuda 10.1
libcudart 10.1

I am using Pre-Trained Model “deepspeech-0.9.3-models.pbmm” and “deepspeech-0.9.3-models.scorer” for my voice-2-text usecase and this working fine.

However, although my machine has 2 GPUs, this model is only executing on 1GPU. I am primarily looking at achieving data parallelism and wondering how to go about doing the same. Can you please suggest how i can provide both my GPUs to the model. I have also tried setting up the environment variable CUDA_VISIBLE_DEVICES=“0,1” but that is not helping me.

Please suggest.