LLMs · Google

Gemma 3 270M

Name: Gemma 3 270M
Author: Google

Parameters: 0.3B
Architecture: Dense
Max context: 32K tokens
License: gemma
Released: 2025-08

How to run it

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

ollama run gemma3:270m

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

huggingface.co/unsloth/gemma-3-270m-it-GGUF (opens in a new tab)

ollama run hf.co/unsloth/gemma-3-270m-it-GGUF

Original weights · Hugging Face

Full-precision files from Google, for fine-tuning, vLLM, Transformers.

huggingface.co/google/gemma-3-270m-it (opens in a new tab)

Free Hugging Face account needed: request access on the model page, and Google approves it (usually within minutes to a few hours).

Official repo · GitHub

Reference code, inference scripts and docs from Google. The weights themselves download from Hugging Face.

github.com/google-deepmind/gemma (opens in a new tab)

git clone https://github.com/google-deepmind/gemma

Google DeepMind’s official Gemma library.

Links checked 5 days ago against Hugging Face, Ollama and GitHub.

Memory needed

Weights plus KV cache and runtime overhead. Leave a little extra free for the operating system and other apps.

KV cache: sliding-window hybrid: the KV estimate counts the global-attention layers plus a small allowance for the sliding-window layers.

Quantization	Quality	Weights	Total at 8K context	Total at 32K context	Download
`Q2_K`	Low	0.1 GB	0.7 GB	0.8 GB	GGUF (opens in a new tab)
`Q3_K_M`	Low	0.1 GB	0.8 GB	0.9 GB	GGUF (opens in a new tab)
`Q4_K_M`	Good	0.2 GB	0.8 GB	0.9 GB	GGUF (opens in a new tab)
`Q5_K_M`	High	0.2 GB	0.8 GB	0.9 GB	GGUF (opens in a new tab)
`Q6_K`	High	0.2 GB	0.9 GB	1.0 GB	GGUF (opens in a new tab)
`Q8_0`	Very high	0.3 GB	0.9 GB	1.0 GB	GGUF (opens in a new tab)
`F16`	Full	0.5 GB	1.2 GB	1.3 GB	GGUF (opens in a new tab) Original weights (opens in a new tab)

Best hardware for Gemma 3 270M

The least expensive GPU in our catalog that runs it without offloading is the Intel Arc A380 (launch price $139) at F16.

Grade · points	Hardware	Quant	Memory	Speed	Fit
47 ptsUsable (C) · 47 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 4.5/15	NVIDIA GeForce RTX 4060	`F16`	1.6 GB / 8.0 GB	290 tok/s	Runs well
47 ptsUsable (C) · 47 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 4.4/15	Apple M1 (16 GB)	`F16`	2.0 GB / 10.7 GB	62.7 tok/s	Runs well
47 ptsUsable (C) · 47 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 4.4/15	Apple M4 (16 GB)	`F16`	2.0 GB / 10.7 GB	110 tok/s	Runs well
46 ptsUsable (C) · 46 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 3.5/15	Apple M4 (24 GB)	`F16`	2.5 GB / 16.1 GB	110 tok/s	Runs well
46 ptsUsable (C) · 46 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 3.5/15	Apple M4 Pro (24 GB)	`F16`	2.5 GB / 16.1 GB	251 tok/s	Runs well
46 ptsUsable (C) · 46 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 3.4/15	Intel Arc B580	`F16`	1.8 GB / 12.0 GB	372 tok/s	Runs well
46 ptsUsable (C) · 46 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 3.4/15	NVIDIA GeForce RTX 3060 12GB	`F16`	1.8 GB / 12.0 GB	384 tok/s	Runs well
46 ptsUsable (C) · 46 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 3.4/15	NVIDIA GeForce RTX 4070 SUPER	`F16`	1.8 GB / 12.0 GB	538 tok/s	Runs well
45 ptsUsable (C) · 45 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 2.8/15	AMD Radeon RX 9070 XT	`F16`	2.0 GB / 16.0 GB	588 tok/s	Runs well
45 ptsUsable (C) · 45 of 100 points: Model quality 2.7/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 2.8/15	NVIDIA GeForce RTX 4060 Ti 16GB	`F16`	2.0 GB / 16.0 GB	307 tok/s	Runs well

Check it on your own hardware

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.