examples/asr/asr_hybrid_transducer_ctc/speech_to_text_hybrid_rnnt_ctc_bpe.py not working with numba-cuda>0.15.1

[ican24]

Dear Team,

examples/asr/asr_hybrid_transducer_ctc/speech_to_text_hybrid_rnnt_ctc_bpe.py is not working with numba-cuda>0.15.1. Only after installation of numba-cuda with less version it works.

pip install numba-cuda==0.15.1

Please list minimal steps or code snippet for us to be able to reproduce the bug.

python -u examples/asr/asr_hybrid_transducer_ctc/speech_to_text_hybrid_rnnt_ctc_bpe.py \
    --config-path="../conf/fastconformer/hybrid_cache_aware_streaming" \
    --config-name="fastconformer_hybrid_transducer_ctc_bpe_streaming" \
    model.train_ds.manifest_filepath=datanew/train0 \
    model.validation_ds.manifest_filepath=datanew/test0 \
    model.tokenizer.dir=datanew/tokenize/tokenizer_spe_bpe_v1024 \
    model.tokenizer.type="bpe" \
    trainer.max_epochs=100 \
    model.optim.name="adamw" \
    model.optim.weight_decay=0.0001 \
    model.optim.sched.warmup_steps=2000 \
    model.aux_ctc.ctc_loss_weight=0.3 \
    model.optim.lr=0.005 \
    model.optim.betas=[0.9,0.999]

The error

  File "cuda/bindings/nvjitlink.pyx", line 124, in cuda.bindings.nvjitlink.add_data
  File "cuda/bindings/nvjitlink.pyx", line 143, in cuda.bindings.nvjitlink.add_data
  File "cuda/bindings/nvjitlink.pyx", line 75, in cuda.bindings.nvjitlink.check_status
cuda.bindings.nvjitlink.nvJitLinkError: ERROR_INTERNAL (6)
Linker error log: ERROR 4 in nvvmAddNVVMContainerToProgram, may need newer version of nvJitLink library

Environment details

If NVIDIA docker image is used you don't need to specify these. Otherwise, please provide:

• OS version Ubuntu 24.04

• PyTorch version 2.9.0

• Python version 3.12 Additional context

Add any other context about the problem here. Example: GPU model NVIDIA RTX 5090