↑

Gemini / gemma3

multimodal

gemma3GemmaGoogle

0 0 0 更新于 2026-05-06 22:07

概述

Gemma 3

Gemma 是 Google 基于 Gemini 技术构建的轻量级模型系列。Gemma 3 模型是多模态模型（可处理文本和图像），并具有 128K 上下文窗口，支持超过 140 种语言。Gemma 3 提供 270M、1B、4B、12B 和 27B 的参数大小，在问答、摘要和推理等任务中表现出色，其紧凑的设计使其能够在资源有限的设备上部署。目前，Gemma3模型是单个 GPU 上运行功能最强大的模型。

模型

名称	尺寸	上下文	输入	Ollama 下载命令
gemma3:latest	3.3GB	128K	Text, Image	`ollama run gemma3:latest`
gemma3:270m	292MB	32K	Text	`ollama run gemma3:270m`
gemma3:1b	815MB	32K	Text	`ollama run gemma3:1b`
gemma3:4b	3.3GB	128K	Text, Image	`ollama run gemma3:4b`
gemma3:12b	8.1GB	128K	Text, Image	`ollama run gemma3:12b`
gemma3:27b	17GB	128K	Text, Image	`ollama run gemma3:27b`
gemma3:270m-it-qat	241MB	32K	Text	`ollama run gemma3:270m-it-qat`
gemma3:270m-it-q8_0	292MB	32K	Text	`ollama run gemma3:270m-it-q8_0`
gemma3:270m-it-fp16	543MB	32K	Text	`ollama run gemma3:270m-it-fp16`
gemma3:270m-it-bf16	543MB	32K	Text	`ollama run gemma3:270m-it-bf16`
gemma3:1b-it-qat	1.0GB	32K	Text	`ollama run gemma3:1b-it-qat`
gemma3:1b-it-q4_K_M	815MB	32K	Text	`ollama run gemma3:1b-it-q4_K_M`
gemma3:1b-it-q8_0	1.1GB	32K	Text	`ollama run gemma3:1b-it-q8_0`
gemma3:1b-it-fp16	2.0GB	32K	Text	`ollama run gemma3:1b-it-fp16`
gemma3:4b-it-qat	4.0GB	128K	Text, Image	`ollama run gemma3:4b-it-qat`
gemma3:4b-it-q4_K_M	3.3GB	128K	Text, Image	`ollama run gemma3:4b-it-q4_K_M`
gemma3:4b-it-q8_0	5.0GB	128K	Text, Image	`ollama run gemma3:4b-it-q8_0`
gemma3:4b-it-fp16	8.6GB	128K	Text, Image	`ollama run gemma3:4b-it-fp16`
gemma3:12b-it-qat	8.9GB	128K	Text, Image	`ollama run gemma3:12b-it-qat`
gemma3:12b-it-q4_K_M	8.1GB	128K	Text, Image	`ollama run gemma3:12b-it-q4_K_M`
gemma3:12b-it-q8_0	13GB	128K	Text, Image	`ollama run gemma3:12b-it-q8_0`
gemma3:12b-it-fp16	24GB	128K	Text, Image	`ollama run gemma3:12b-it-fp16`
gemma3:27b-it-qat	18GB	128K	Text, Image	`ollama run gemma3:27b-it-qat`
gemma3:27b-it-q4_K_M	17GB	128K	Text, Image	`ollama run gemma3:27b-it-q4_K_M`
gemma3:27b-it-q8_0	30GB	128K	Text, Image	`ollama run gemma3:27b-it-q8_0`
gemma3:27b-it-fp16	55GB	128K	Text, Image	`ollama run gemma3:27b-it-fp16`

评估

Gemma 3 270M

Benchmark	n-shot	Gemma 3 270m instruction tuned
HellaSwag	0-shot	37.7
PIQA	0-shot	66.2
ARC-C	0-shot	28.2
WinoGrande	0-shot	52.3
BIG-Bench Hard	few-shot	26.7
IF Eval	0-shot	51.2

这些模型针对大量不同的数据集和指标进行了评估，以涵盖文本生成的不同方面：

推理、逻辑和代码能力

Benchmark	Metric	Gemma 3 PT 1B	Gemma 3 PT 4B	Gemma 3 PT 12B	Gemma 3 PT 27B
HellaSwag	10-shot	62.3	77.2	84.2	85.6
BoolQ	0-shot	63.2	72.3	78.8	82.4
PIQA	0-shot	73.8	79.6	81.8	83.3
SocialIQa	0-shot	48.9	51.9	53.4	54.9
TriviaQA	5-shot	39.8	65.8	78.2	85.5
Natural Questions	5-shot	9.48	20.0	31.4	36.1
ARC-C	25-shot	38.4	56.2	68.9	70.6
ARC-e	0-shot	73.0	82.4	88.3	89.0
WinoGrande	5-shot	58.2	64.7	74.3	78.8
BIG-Bench Hard	—	28.4	50.9	72.6	77.7
DROP	3-shot, F1	42.4	60.1	72.2	77.2
AGIEval	3–5-shot	22.2	42.1	57.4	66.2
MMLU	5-shot, top-1	26.5	59.6	74.5	78.6
MATH	4-shot	—	24.2	43.3	50.0
GSM8K	5-shot, maj@1	1.36	38.4	71.0	82.6
GPQA	—	9.38	15.0	25.4	24.3
MMLU (Pro)	5-shot	11.2	23.7	40.8	43.9
MBPP	3-shot	9.80	46.0	60.4	65.6
HumanEval	pass@1	6.10	36.0	45.7	48.8
MMLU (Pro COT)	5-shot	9.7	NaN	NaN	NaN

多语言能力

文件 1：多语言 / 文本任务

Benchmark	Gemma 3 PT 1B	Gemma 3 PT 4B	Gemma 3 PT 12B	Gemma 3 PT 27B
MGSM	2.04	34.7	64.3	74.3
Global-MMLU-Lite	24.9	57.0	69.4	75.7
Belebele	26.6	59.4	78.0	—
WMT24± (ChrF)	36.7	48.4	53.9	55.7
FloRes	29.5	39.2	46.0	48.8
XL-Sum	4.82	8.55	12.2	14.9
XQuAD (all)	43.9	68.0	74.5	76.8

多式联运能力

Benchmark	Gemma 3 PT 4B	Gemma 3 PT 12B	Gemma 3 PT 27B
COCO Captions (CIDEr)	102	111	116
DocVQA (val)	72.8	82.3	85.6
InfoVQA (val)	44.1	54.8	59.4
MMMU (pt)	39.2	50.3	56.1
TextVQA (val)	58.9	66.5	68.6
RealWorldQA	45.5	52.2	53.9
ReMl	27.3	38.5	44.8
AI2D	63.2	75.2	79.0
ChartQA	45.4	60.9	63.8
ChartQA (augmented)	81.8	88.5	88.7
VQAv2	—	—	—
BLINK	38.0	35.9	39.6
OKVQA	51.0	58.7	60.2
TallyQA	42.5	51.8	54.3
SpatialSense VQA	50.9	60.0	59.4
CountBenchQA	26.1	17.8	68.0