DartQuant
A learning-based orthogonal rotation quantization method that improves quantization performance by iteratively optimizing the rotation matrix during calibration. However, it has high memory usage and long runtime, imposing a considerable burden in practical deployment.
Timeline (last 90 days)
Existing orthogonal transform methods fall roughly into two categories: data-independent fixed transforms (such as the Hadamard rotation) and rotation matrices learned during the calibration stage (such as SpinQuant and DartQuant).
Learned rotation methods (such as SpinQuant and DartQuant) require repeated iterative optimization of the rotation matrix before quantization, resulting in high memory usage and long runtime.
All Facts (2)
Existing orthogonal transform methods fall roughly into two categories: data-independent fixed transforms (such as the Hadamard rotation) and rotation matrices learned during the calibration stage (such as SpinQuant and DartQuant).
50%UnverifiedLearned rotation methods (such as SpinQuant and DartQuant) require repeated iterative optimization of the rotation matrix before quantization, resulting in high memory usage and long runtime.
50%