The code is a zig replica of what explained by Marios Fournarakis and Markus Nagel ine the following paper: A White Paper on Neural Network Quantization
| Task | Status |
|---|---|
| matrix normalization [0, 1] | ✅ |
| vector normalization [0, 1] | ✅ |
| matrix quantization to u8 | ✅ |
| vector quantization to i32 | ✅ |
| vector quantization to u8 | 🚧 |
| qunatization error computation | 🚧 |
| parallelized dot product | 🚧 |
| fix vector = udefined, not good practice | |
| check comptime | |
| tests |