language model quantization

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

https://nvidia.github.io/OpenSeq2Seq

Apache License 2.0

1.54k stars 369 forks source link

language model quantization #364

Open SibtainRazaJamali opened 5 years ago

SibtainRazaJamali commented 5 years ago

Why is the language model being quantized in script/build_lm.py? Is it a trade of between performance and error rate. How can we use trie and lm.binary file without quantizing?