Will It Fit? — VRAM calculator for local AI models

Your model

Model size billions of parameters B Quantization how heavily the model is compressed Context length how much the model can "keep in mind" tokens Context memory GB needed per 1,000 tokens GB / 1K Depends on the model architecture — typical chat models use 0.1–0.35. Model cards often list this as "KV cache".

7.4 GB

of memory needed — weights + context + runtime overhead

Does it fit?

Expert details

weights = params × bpw ÷ 8 · kv = context ÷ 1000 × kv_per_1k · overhead = 1.5 GB · a device fits when total ≤ 90% of its memory