Does a quantization under 16 GB fit in 16 GB VRAM?

Not necessarily. Download size, GPU weight size, resident memory and total runtime memory are different quantities. QFF records the size basis where available; a size filter is not a hardware-fit guarantee.

Allow for KV cache, temporary buffers, runtime overhead and other GPU users. Context length, batch size, cache precision and CPU offloading affect fit. Some models keep large components outside GPU memory, so compare like size bases.

For a practical recommendation, select a supported runtime and compatible measurement set, shortlist variants using the stated size basis, then measure peak GPU memory on your target hardware with the intended context and batch size. QFF's publisher claims can guide a shortlist, but do not establish throughput or successful loading.

QFF methodology and verification gates