search

PABannier / bark.cpp

Suno AI's Bark model in C/C++ for fast text-to-speech
MIT License
630 stars 48 forks source link

model quantize bad encodec #167

Closed Gabri94x closed 1 month ago

Gabri94x commented 1 month ago

hi there, i'm not pretty sure if it a bug or i missed something in my quantization process but this is the result i have running the q4_0 model image this the code line i made for quantization image this is the running line image thank you for the project, it is really great and saves a lot of work!

PABannier commented 1 month ago

Hello @Gabri94x ! Actually encodec.cpp is not quantized due to the minor gains it brings to this model. I'll have a look at the quantization for bark to make sure it works.

bachittle commented 1 month ago

Yea I'm getting the same issue as well. It seems to occur with all quantization types, like q4_1, q8_0, etc.

PABannier commented 1 month ago

Hello all! This should be fixed in #176 . Could you try again and let me know?

bachittle commented 1 month ago

seems to be working now! now just new issues to address, like that the quantized models are slower than regular model. can probably close this issue now tho.

PABannier commented 1 month ago

Ok thanks for confirming :)