about <onnx_decode_sentence.py>

[XhrLeokk]

hi, @frankyoujian , I have successfully ran the sherpa-onnx version, but

I got a batch of sentences want to be decode, I may need to use <onnx_decode_sentence.py> file to be more efficient.
when I try to run <onnx_decode_sentence.py> ,

1. firstly, I change "from model import Model" to "from model import Model_new as Model". to run the code.
2. secondly, as it goes to
   sp.load(args.bpe_model) got error
   RuntimeError: Internal: could not parse ModelProto from ./sherpa-onnx/download/sherpa-onnx-online-punct-en-2024-08-06/bpe.vocab

am I missing a "bpe file" that can be read by sp? or the code need to be updated?
I'm not an expert of ONNX, Thanks for the help.

[csukuangfj]

What is the output

ls -lh ./sherpa-onnx/download/sherpa-onnx-online-punct-en-2024-08-06/bpe.vocab

?

[XhrLeokk]

What is the output

ls -lh ./sherpa-onnx/download/sherpa-onnx-online-punct-en-2024-08-06/bpe.vocab

?

-rw-r--r-- 1 501 staff 146K Aug 5 03:19 *.bpe.vocab

[csukuangfj]

Does the file bpe.vocab exist?

The output only shows *.bpe.vocab.

Please make sure you have downloaded the model files correctly.

[XhrLeokk]

Does the file bpe.vocab exist?

The output only shows *.bpe.vocab.

Please make sure you have downloaded the model files correctly.

yes, the file exist.
I used exactly this file ran the decode demo in []https://k2-fsa.github.io/sherpa/onnx/punctuation/pretrained_models.html#sherpa-onnx-online-punct-en-2024-08-06-english-only
and it gives me the Puncted and Cased lines.
But I didn't find a way to efficiently decode a batch of sentences, that's why I turn back to use <onnx_decode_sentence.py>

the full output, I ignore the path of ".vocab"
-rw-r--r-- 1 501 staff 146K Aug 5 03:19 ./sherpa-onnx/download/sherpa-onnx-online-punct-en-2024-08-06/bpe.vocab

[csukuangfj]

Could you use an absolute path in your code?

[XhrLeokk]

your

yes, I can, the root path is /my_name.

[frankyoujian]

@XhrLeokk If you want to decode a batch of sentences, please concatenate them to one line sentence. It preprocesses the one line sentence by splitting the sentence into sequences of 200 tokens. This behavior is similar to that in onnx_decode_sentence.py.

[monkiravn]

I have same issue