kpu / kenlm

KenLM: Faster and Smaller Language Model Queries
http://kheafield.com/code/kenlm/
Other
2.46k stars 514 forks source link

AttributeError: 'EncDecCTCModelBPE' object has no attribute '_wer'. Did you mean: 'wer'? #442

Open Dreahim opened 6 months ago

Dreahim commented 6 months ago

Hello, when I evaluate my ASR using this command, I get an error

python eval_beamsearch_ngram_ctc.py nemo_model_file=/home/user4/training/final_project/project6_nemo/models/python_model_conformer_ctc.nemo \               
             input_manifest=/home/user4/training/final_project/project6_nemo/data_manifest/dev_manifest4.json \
             kenlm_model_file=/home/user4/training/final_project/project6_nemo/LM/LM_models/old_LM_5gram \
             beam_width=[4,8,16,32,64] \
             beam_alpha=[-2.0,-1.0,0,1,2] \
             beam_beta=[-1.0,-0.5,0.0,0.5,1.0] \
             preds_output_folder=/home/user4/training/final_project/project6_nemo/LM/LM_models/pred \
             probs_cache_file=null \
             decoding_mode=beamsearch_ngram \
             decoding_strategy="flashlight"

The error is : _File "/home/user4/training/final_project/project6_nemo/LM/NeMo/eval_beamsearch_ngram_ctc.py", line 353, in main pred_text = asr_model._wer.decoding.ctc_decoder_predictions_tensor(preds_tensor)[0][0] File "/home/user4/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1695, in getattr raise AttributeError(f"'{type(self).name}' object has no attribute '{name}'") AttributeError: 'EncDecCTCModelBPE' object has no attribute 'wer'. Did you mean: 'wer'?

I tried to go to line 353 and change _wer to wer , ( I don't know if this was right), but after that It calculated WER and CER but I get this error: Beam search requires that consecutive ctc tokens are not folded

please tell me how to fix the problem, thank you