Hardware-Aware Quantisation Explorer

Visualise how neural network weights map to fixed-point and low-bit formats

Select Number Format

Bit Layout

Decoded Value

0.0

Representable Number Line

Distribution Settings

Error Metrics

--
MSE
--
SNR (dB)
--
Max Error
--
Clipped %

Weight Histogram (FP32 vs Quantised)

FP32 Original
Quantised

Scheme Settings

Quantisation Parameters

Quantisation Mapping

GPTQ-Style Column-wise Effect

Quantising one column introduces error that propagates to remaining columns. Watch how compensating updates reduce total error.

Toy Network: 2D Classification

A 3-layer MLP trained on a spiral dataset. Quantise and see how the decision boundary degrades.

FP32 Decision Boundary

--
Accuracy

Quantised Decision Boundary

--
Accuracy
--
Degradation

Layer Sensitivity Analysis

Hardware Cost Comparison

Model Size Calculator

Area-Accuracy Tradeoff

Relative multiplier area vs estimated accuracy retention. The Pareto-optimal frontier defines the best achievable tradeoffs.

Layer Precision Assignment

Assign precision per layer. Sensitive layers benefit from higher precision.

Summary

--
Model Size
--
Est. Accuracy
--
vs FP16 Size