Mixed Quantization

Per-Tensor Quantization

Method applying a single set of quantization parameters to an entire tensor, ensuring scale consistency for all values. This approach simplifies hardware implementation but may reduce precision for extended distributions.

← Geri