LLM Quantization Calculator
Compare model weight size across FP32, FP16, INT8 and the low-bit formats.
Runs entirely in your browser. Nothing you type is uploaded.
About this tool
Enter a parameter count to see the theoretical weight size at each precision, the size with an overhead allowance, and the saving against FP16. Use it to judge whether a model fits before downloading a quantized checkpoint.
Questions & answers
- Why is a real GGUF file bigger than the theoretical size?
- Quantized formats store a scale and often a zero-point for each small group of weights, and usually keep embeddings and normalisation layers at higher precision. The overhead field is there to approximate that.
- How is the size calculated?
- Parameters multiplied by bits per weight, divided by eight. A 7B model at 4-bit is 7,000,000,000 × 4 / 8 = 3.5 GB before overhead.
- Does lower precision hurt quality?
- Generally yes, and increasingly below about 4 bits, but how much depends on the model and the quantization method. This tool only reports size; it makes no claim about output quality.