Learn · Tool

Model size calculator

How much storage do a model's weights take? Enter the parameter count from a model card and a precision. The answer is arithmetic, not a hardware requirement.

Parameters, in billions

The total parameter count from the model card, for example 8 for an 8B model.

Precision (bits per weight)

Or enter bits per weight

Weights alone, at 16 bits per weight for 8 billion parameters:

16.0 GB (14.9 GiB)

Weight storage for the same parameter count at common precisions
Precision	GB	GiB
32-bit	32.0	29.8
16-bit	16.0	14.9
8-bit	8.00	7.45
4-bit	4.00	3.73

How it works

Each parameter is stored as a number. At 16 bits per weight that is 2 bytes, at 8 bits 1 byte, and at 4 bits half a byte. So weight storage is the parameter count times the bits per weight, divided by 8. GB here means 1,000,000,000 bytes; GiB means 1,073,741,824 bytes, which some tools report instead.

What it leaves out

Running a model also needs working memory that grows with the length of the conversation and the number of requests handled at once. Real quantized files are often somewhat larger than the arithmetic suggests: the quantization explainer describes a 4-bit copy as about a quarter to a third the size of a 16-bit original. Use the result as a starting point and check the publisher's and the runtime's own documentation.

Support Us

Help keep USASI useful.

Leave a tip

Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).

About supporting this project