HQQ
A post-training quantization method designed for extremely low bit-widths. It minimizes quantization error through a half-quadratic optimization framework, significantly reducing performance degradation at INT4 precision compared to standard uniform quantization, making it suitable for edge device deployment.
Timeline (last 90 days)
When using INT4 HQQ quantization, catastrophic transcription loops occurring when the model processes West African accents were amplified by 5 to 7 times.
HQQ (Half-Quadratic Quantization) is a post-training quantization method designed for extremely low bit-widths that minimizes quantization error through a half-quadratic optimization framework.
All Facts (2)
When using INT4 HQQ quantization, catastrophic transcription loops occurring when the model processes West African accents were amplified by 5 to 7 times.
50%UnverifiedHQQ (Half-Quadratic Quantization) is a post-training quantization method designed for extremely low bit-widths that minimizes quantization error through a half-quadratic optimization framework.
50%