Will this model run on your machine?

A check before you download 40 GB for nothing. The arithmetic is written out in full, and the link keeps your settings.

Architecture read from unsloth/gemma-3-4b-it, since the official repository requires accepting a licence. The file there is identical.

Installed memory

The processor and the GPU share this memory.

03 Quantization

About 4.9 bits per parameter in a GGUF file.

04 Context

8,192tokens

Prompt + answer. The model accepts 131,072 tokens at most.

05 Parallel requests

One conversation at a time in almost every case.

What fits on this machine

The largest models in the catalog that fit at Q4_K_M with a 8,192-token context.

34layers

2,560width

8heads

4KV heads

262Kvocabulary

131KContext

Sliding-window attention: the real cache can be smaller than this estimate.