Gemma 3 27B
Parameters
27B
Context Length
128K
Modality
Multimodal
Architecture
Dense
License
Gemma Terms of Use
Release Date
12 Mar 2025
Knowledge Cutoff
Aug 2024
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
3x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
128,000 tokens
Consumer
5x RTX 4090
24GB VRAM
Datacenter
2x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
Rank
#132
| Benchmark | Score | Rank |
|---|---|---|
QA Assistant | 0.913 | 13 |
Summarization | 0.802 | 13 |
StackUnseen | 0.372 | 28 |
Coding | 0.05 | 36 |
General Text | 1366 | 107 |
Rankings
Overall Rank
#132
Coding Rank
#99
About Gemma 3 27B
Gemma 3 is a family of lightweight, state-of-the-art models developed by Google DeepMind, designed with research and technology derived from the Gemini models. The Gemma 3 27B variant is a multimodal model engineered to process both textual and image inputs, generating text-based outputs. This model variant is intended for broad application across various generation tasks, including question answering, summarization, and complex reasoning, and supports over 140 languages. Its design focuses on enabling deployment on a range of hardware, from consumer-grade devices like laptops and workstations to specialized cloud infrastructure.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
64
Key-Value Heads
16
Attention Head Dimension
-
Position Embedding
ROPE
RoPE Theta
-
Sliding Window Attention
-
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
-
Dimensions
Hidden Dimension Size
4,096
Number of Layers
46
FFN Intermediate Size (Dense)
-
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
-
About Gemma 3
Gemma 3 is a family of open, lightweight models from Google. It introduces multimodal image and text processing, supports over 140 languages, and features extended context windows up to 128K tokens. Models are available in multiple parameter sizes for diverse applications.