MODEL / GOOGLE

Gemma 3 12B

Google reports an INT4 path on an 8 GB laptop GPU, but context and runtime overhead can make 8 GB tight.

一次情報

総パラメータ12B

推定メモリ~9.4 GBINT4

コンテキスト128KText + Image

LicenseGemma140+ languages

01 / 用途

vision
knowledge work
multilingual

02 / 推定適合

Mac Studio M5 Ultra · 96 GB余裕あり Mac mini M6 · 24 GB余裕あり MacBook Pro M5 Max · 64 GB余裕あり Framework Desktop · Ryzen AI Max+ 395 · 128 GB余裕あり Radeon AI PRO R9700 · 32 GB余裕あり

03 / 実行環境

These are runtime examples, not validated installation recipes. Fit estimates assume quantized files; official repositories may hold full-precision weights. MLX, GGUF and TensorRT require matching conversions or engines.

OllamaEasyollama run gemma3:12b

llama.cppMediumllama-server -m /path/to/verified-model.gguf --port 9931

LM StudioEasySearch “google/gemma-3-12b-it” → Load a Q4/INT4 build → Start Server

MLX LMMediummlx_lm.server --model google/gemma-3-12b-it

注意

Google reports an INT4 path on an 8 GB laptop GPU, but context and runtime overhead can make 8 GB tight.