quantize_tensors
Executes hardware-aware, piecewise asymmetric INT8 linear tensor quantization to compress high-density float32 weights arrays down to uniform integer sectors. Dynamically calculates block-wise scaling constants and zero-point alignment metrics to isolate distribution outliers. Clamps values strictly between the standard 0 and 255 byte limits to accelerate tensor processing array processing speeds.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| block_size | No | The discrete integer segment capacity (default: 128) over which independent scaling bounds and shifts are calculated to preserve weight density precision. | |
| tensor_data | Yes | A flattened array of float32 values representing the raw weight distributions or dense mathematical matrices to undergo byte quantization. |