Skip to content

SPEC.md

Latest commit

 

History

History
32 lines (19 loc) · 1.13 KB

SPEC.md

File metadata and controls

32 lines (19 loc) · 1.13 KB

Quantization Specification

Version: v1.2.3 | Status: Active | Last Updated: March 2026

Overview

Provides Int8 and FP4 (4-bit float) quantization for neural network weights. Supports symmetric and asymmetric Int8 quantization with nibble-packed FP4 storage, plus error metrics for quality assessment.

Functional Requirements

  1. Symmetric and asymmetric Int8 quantization with per-tensor scale and zero-point
  2. FP4 quantization with nibble-packed storage (8x compression ratio)
  3. Quantization error metrics: MSE, max absolute error, and SQNR

Interface

from codomyrmex.quantization import quantize_int8, dequantize_int8, quantize_fp4, dequantize_fp4, quantization_error

qt = quantize_int8(weights, scheme="asymmetric")
reconstructed = dequantize_int8(qt)
error = quantization_error(weights, reconstructed)

Exports

Int8Quantizer, QuantizedTensor, quantize_int8, dequantize_int8, FP4Quantizer, FP4Tensor, quantize_fp4, dequantize_fp4, compute_scale_zero_point, per_channel_scale, quantization_error

Navigation