wav2vec2-hidet

Warning

This test has moved to https://github.com/egorsmkv/wav2vec2-hidet-0.4.0

NOTE: this is a work-in-progress thing, do not use it anywhere except dev machine

Installation:

git clone https://github.com/egorsmkv/wav2vec2-hidet
cd wav2vec2-hidet

conda create -n w2v2-hidet python=3.10
conda activate w2v2-hidet

pip install 'torch>2.0.0' torchaudio
pip install evaluate soundfile transformers jiwer
pip install https://github.com/kpu/kenlm/archive/master.zip pyctcdecode

conda install -c conda-forge nvtx
pip install hidet

conda install -c conda-forge cudatoolkit-dev

Running:

wget https://huggingface.co/Yehor/wav2vec2-xls-r-300m-uk-with-small-lm/resolve/main/language_model/5gram_correct.arpa
wget https://huggingface.co/Yehor/wav2vec2-xls-r-300m-uk-with-small-lm/raw/main/language_model/unigrams.txt

python export_gpu_version.py

python recognize_gpu_with_lm_and_hidet.py

Links:

• https://arxiv.org/abs/2006.11477
• https://huggingface.co/docs/transformers/model_doc/wav2vec2
• https://pytorch.org/blog/introducing-hidet/
• https://docs.hidet.org/stable/gallery/tutorials/optimize-pytorch-model.html
• https://discuss.huggingface.co/t/integration-transformers-and-hidet-for-wav2vec2/44094