AI & Computing

Model File Size Calculator (Quantization)

Enter the number of parameters to compare model sizes at different precisions.

By TheCalcTools Updated Formula checked and tested

Your numbers

Quantised formats store small scales alongside weights, so 4-bit formats often average 4.5 to 5 bits

Your results

Size at your bits per weight (GB)
Size at your bits per weight (GB)
—
FP32 (GB)
—
FP16 / BF16 (GB)
—
INT8 (GB)
—
4-bit (GB)
—
Saving versus FP16
—

Model File Size Calculator (Quantization) by TheCalcTools

How the Model File Size Calculator (Quantization) works

Size in bytes = parameters × bits per weight ÷ 8. Quantisation stores weights with fewer bits, cutting memory and download size at a small cost in accuracy.

Sizes here are decimal gigabytes (10⁹ bytes). Files also contain the tokenizer and metadata, which add little.

The formula

Size (GB) = parameters (billions) × bits ÷ 8

Worked example

Using the default values shown in the calculator:

Parameters (billions)13
Your bits per weight4.85
  1. 13 billion × 4.85 bits ÷ 8 = 7.88 GB, against 26 GB at FP16.

Result:

Size at your bits per weight (GB)7.88
FP32 (GB)52.0
FP16 / BF16 (GB)26.0
INT8 (GB)13.0
4-bit (GB)6.50
Saving versus FP1669.7%

Frequently asked questions

Does 4-bit quantisation hurt quality?

Modern 4-bit methods usually lose only a little accuracy for large models; very small models and 2-3 bit formats lose more.

Sources and further reading

GPU Memory (VRAM) Calculator for LLMsEnter the model size, precision and context to estimate how much VRAM you need LLM Tokens per Second Calculator (Memory Bandwidth)Enter memory bandwidth and model size to estimate generation speed Data Storage ConverterWhy a 500 GB drive shows up smaller in your operating system, answered in numbers

More in AI & Computing · All calculators

This calculator gives estimates for information only. Check important results independently before you rely on them. See our disclaimer.