FunASR/funasr/runtime/python/benchmark_libtorch.md

Benchmark

Data set:

Aishell1 test set , the total audio duration is 36108.919 seconds.

Tools

• Install ModelScope and FunASR

  pip install "modelscope[audio_asr]" --upgrade -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
  git clone https://github.com/alibaba-damo-academy/FunASR.git && cd FunASR
  pip install --editable ./
  cd funasr/runtime/python/utils
  pip install -r requirements.txt
  

• recipe

  set the model, data path and output_dir

  nohup bash test_rtf.sh &> log.txt &
  

Paraformer-large

Intel(R) Xeon(R) Platinum 8269CY CPU @ 2.50GHz 16core-32processor with avx512_vnni

concurrent-tasks	processing time(s)	RTF	Speedup Rate
1 (torch fp32)	3522	0.0976	10.3
1 (torch int8)	1746	0.0484	20.7
32 (torch fp32)	236	0.0066	152.7
32 (torch int8)	114	0.0032	317.4
64 (torch fp32)	235	0.0065	153.7
64 (torch int8)	113	0.0031	319.2

Paraformer