Product Engineering
Open-Source Contributor
Added int8 dynamic quantization support to Kyutai-Labs/pocket-tts, enabling efficient mixed-precision inference with:
- ~48 % lower runtime memory footprint
- ~27 % faster CPU inference (x86)
- CLI & API flags for opt-in quantization
- No measurable impact on audio quality 👉 PR: kyutai-labs/pocket-tts#147



