Evaluating with NGPU-LM failed while training.

[hoangtch-namitech]

Describe the bug

GreedyBatchedCTCInfer, BeamBatchedCTCInfer, BeamBatchedRNNTInfer, BeamBatchedTDTInfer are not subclasses of WithOptionalCudaGraphs. Hence, validating using greedy_batch/beam_batch/malsd_batch decoding with NGPU-LM in training ASR model will cause RuntimeError: CUDA error: an illegal memory access was encountered

Steps/Code to reproduce bug

Train and eval ASR model using greedy_batch/beam_batch/malsd_batch decoding with NGPU-LM.

[hoangtch-namitech]

@lilithgrigoryan I have opened a PR (#13995) for this issue, would you mind checking this pls? Thanks

[github-actions]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

[artbataev]

Fixed in #13995