gemma-3-27b-it GGUF Models

How to Use Gemma 3 Vision with llama.cpp

To utilize the experimental support for Gemma 3 Vision in llama.cpp, follow these steps:

  1. Clone the lastest llama.cpp Repository:

    git clone https://github.com/ggml-org/llama.cpp.git
    cd llama.cpp
    
  2. Build the Llama.cpp:

Build llama.cpp as usual : https://github.com/ggml-org/llama.cpp#building-the-project

Once llama.cpp is built Copy the ./llama.cpp/build/bin/llama-gemma3-cli to a chosen folder.

  1. Download the Gemma 3 gguf file:

https://huggingface.co/Mungert/gemma-3-4b-it-gguf/tree/main

Choose a gguf file without the mmproj in the name

Example gguf file : https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/google_gemma-3-4b-it-q4_k_l.gguf

Copy this file to your chosen folder.

  1. Download the Gemma 3 mmproj file

https://huggingface.co/Mungert/gemma-3-4b-it-gguf/tree/main

Choose a file with mmproj in the name

Example mmproj file : https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/google_gemma-3-4b-it-mmproj-bf16.gguf

Copy this file to your chosen folder.

  1. Copy images to the same folder as the gguf files or alter paths appropriately.

In the example below the gguf files, images and llama-gemma-cli are in the same folder.

Example image: image https://huggingface.co/Mungert/gemma-3-4b-it-gguf/resolve/main/car-1.jpg

Copy this file to your chosen folder.

  1. Run the CLI Tool:

From your chosen folder :

llama-gemma3-cli -m google_gemma-3-4b-it-q4_k_l.gguf --mmproj google_gemma-3-4b-it-mmproj-bf16.gguf
 Running in chat mode, available commands:
   /image <path>    load an image
   /clear           clear the chat history
   /quit or /exit   exit the program

> /image car-1.jpg
Encoding image car-1.jpg
Image encoded in 46305 ms
Image decoded in 19302 ms

> what is the image of
Here's a breakdown of what's in the image:

**Subject:** The primary subject is a black Porsche Panamera Turbo driving on a highway.

**Details:**

*   **Car:** It's a sleek, modern Porsche Panamera Turbo, identifiable by its distinctive rear design, the "PORSCHE" lettering, and the "Panamera Turbo" badge. The license plate reads "CVC-911".
*   **Setting:** The car is on a multi-lane highway, with a blurred background of trees, a distant building, and a cloudy sky. The lighting suggests it's either dusk or dawn.
*   **Motion:** The image captures the car in motion, with a slight motion blur to convey speed.

**Overall Impression:** The image conveys a sense of speed, luxury, and power. It's a well-composed shot that highlights the car's design and performance.

Do you want me to describe any specific aspect of the image in more detail, or perhaps analyze its composition?

Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)

Our latest quantization method introduces precision-adaptive quantization for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on Llama-3-8B. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.

Benchmark Context

All tests conducted on Llama-3-8B-Instruct using:

  • Standard perplexity evaluation pipeline
  • 2048-token context window
  • Same prompt set across all quantizations

Key Improvements

  • Dynamic Precision Allocation:
    • First/Last 25% of layers → IQ4_XS (selected layers)
    • Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
  • Critical Component Protection:
    • Embeddings/output layers use Q5_K
    • Reduces error propagation by 38% vs standard 1-2bit

Quantization Performance Comparison (Llama-3-8B)

Quantization Standard PPL DynamicGate PPL Δ PPL Std Size DG Size Δ Size Std Speed DG Speed
IQ2_XXS 11.30 9.84 -12.9% 2.5G 2.6G +0.1G 234s 246s
IQ2_XS 11.72 11.63 -0.8% 2.7G 2.8G +0.1G 242s 246s
IQ2_S 14.31 9.02 -36.9% 2.7G 2.9G +0.2G 238s 244s
IQ1_M 27.46 15.41 -43.9% 2.2G 2.5G +0.3G 206s 212s
IQ1_S 53.07 32.00 -39.7% 2.1G 2.4G +0.3G 184s 209s

Key:

  • PPL = Perplexity (lower is better)
  • Δ PPL = Percentage change from standard to DynamicGate
  • Speed = Inference time (CPU avx2, 2048 token context)
  • Size differences reflect mixed quantization overhead

Key Improvements:

  • 🔥 IQ1_M shows massive 43.9% perplexity reduction (27.46 → 15.41)
  • 🚀 IQ2_S cuts perplexity by 36.9% while adding only 0.2GB
  • IQ1_S maintains 39.7% better accuracy despite 1-bit quantization

Tradeoffs:

  • All variants have modest size increases (0.1-0.3GB)
  • Inference speeds remain comparable (<5% difference)

When to Use These Models

📌 Fitting models into GPU VRAMMemory-constrained deploymentsCpu and Edge Devices where 1-2bit errors can be tolerated
Research into ultra-low-bit quantization

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

  • A 16-bit floating-point format designed for faster computation while retaining good precision.
  • Provides similar dynamic range as FP32 but with lower memory usage.
  • Recommended if your hardware supports BF16 acceleration (check your device’s specs).
  • Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.


F16 (Float 16) – More widely supported than BF16

  • A 16-bit floating-point high precision but with less of range of values than BF16.
  • Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
  • Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.


Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

  • Lower-bit models (Q4_K)Best for minimal memory usage, may have lower precision.
  • Higher-bit models (Q6_K, Q8_0)Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).


Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for extreme memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

  • IQ3_XS: Ultra-low-bit quantization (3-bit) with extreme memory efficiency.

    • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
    • Trade-off: Lower accuracy compared to higher-bit quantizations.
  • IQ3_S: Small block size for maximum memory efficiency.

    • Use case: Best for low-memory devices where IQ3_XS is too aggressive.
  • IQ3_M: Medium block size for better accuracy than IQ3_S.

    • Use case: Suitable for low-memory devices where IQ3_S is too limiting.
  • Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

    • Use case: Best for low-memory devices where Q6_K is too large.
  • Q4_0: Pure 4-bit quantization, optimized for ARM devices.

    • Use case: Best for ARM-based devices or low-memory environments.

Summary Table: Model Format Selection

Model Format Precision Memory Usage Device Requirements Best Use Case
BF16 Highest High BF16-supported GPU/CPUs High-speed inference with reduced memory
F16 High High FP16-supported devices GPU inference when BF16 isn’t available
Q4_K Medium Low Low CPU or Low-VRAM devices Best for memory-constrained environments
Q6_K Medium Moderate CPU with more memory Better accuracy while still being quantized
Q8_0 High Moderate CPU or GPU with enough VRAM Best accuracy among quantized models
IQ3_XS Very Low Very Low Ultra-low-memory devices Extreme memory efficiency and low accuracy
Q4_0 Low Low ARM or low-memory devices llama.cpp can optimize for ARM devices

Included Files & Details

gemma-3-27b-it-bf16.gguf

  • Model weights preserved in BF16.
  • Use this if you want to requantize the model into a different format.
  • Best if your device supports BF16 acceleration.

gemma-3-27b-it-f16.gguf

  • Model weights stored in F16.
  • Use if your device supports FP16, especially if BF16 is not available.

gemma-3-27b-it-bf16-q8_0.gguf

  • Output & embeddings remain in BF16.
  • All other layers quantized to Q8_0.
  • Use if your device supports BF16 and you want a quantized version.

gemma-3-27b-it-f16-q8_0.gguf

  • Output & embeddings remain in F16.
  • All other layers quantized to Q8_0.

gemma-3-27b-it-q4_k.gguf

  • Output & embeddings quantized to Q8_0.
  • All other layers quantized to Q4_K.
  • Good for CPU inference with limited memory.

gemma-3-27b-it-q4_k_s.gguf

  • Smallest Q4_K variant, using less memory at the cost of accuracy.
  • Best for very low-memory setups.

gemma-3-27b-it-q6_k.gguf

  • Output & embeddings quantized to Q8_0.
  • All other layers quantized to Q6_K .

gemma-3-27b-it-q8_0.gguf

  • Fully Q8 quantized model for better accuracy.
  • Requires more memory but offers higher precision.

gemma-3-27b-it-iq3_xs.gguf

  • IQ3_XS quantization, optimized for extreme memory efficiency.
  • Best for ultra-low-memory devices.

gemma-3-27b-it-iq3_m.gguf

  • IQ3_M quantization, offering a medium block size for better accuracy.
  • Suitable for low-memory devices.

gemma-3-27b-it-q4_0.gguf

  • Pure Q4_0 quantization, optimized for ARM devices.
  • Best for low-memory environments.
  • Prefer IQ4_NL for better accuracy.

🚀 If you find these models useful

Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 Network Monitor Assitant.

💬 Click the chat icon (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.

What I'm Testing

I'm experimenting with function calling against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".

🟡 TestLLM – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .

The other Available AI Assistants

🟢 TurboLLM – Uses gpt-4o-mini Fast! . Note: tokens are limited since OpenAI models are pricey, but you can Login or Download the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .

🔵 HugLLM – Runs open-source Hugging Face models Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)

gemma-3-27b-it GGUF Models

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

  • A 16-bit floating-point format designed for faster computation while retaining good precision.
  • Provides similar dynamic range as FP32 but with lower memory usage.
  • Recommended if your hardware supports BF16 acceleration (check your device’s specs).
  • Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.


F16 (Float 16) – More widely supported than BF16

  • A 16-bit floating-point high precision but with less of range of values than BF16.
  • Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
  • Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.


Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

  • Lower-bit models (Q4_K)Best for minimal memory usage, may have lower precision.
  • Higher-bit models (Q6_K, Q8_0)Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).


Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for extreme memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

  • IQ3_XS: Ultra-low-bit quantization (3-bit) with extreme memory efficiency.

    • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
    • Trade-off: Lower accuracy compared to higher-bit quantizations.
  • IQ3_S: Small block size for maximum memory efficiency.

    • Use case: Best for low-memory devices where IQ3_XS is too aggressive.
  • IQ3_M: Medium block size for better accuracy than IQ3_S.

    • Use case: Suitable for low-memory devices where IQ3_S is too limiting.
  • Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

    • Use case: Best for low-memory devices where Q6_K is too large.
  • Q4_0: Pure 4-bit quantization, optimized for ARM devices.

    • Use case: Best for ARM-based devices or low-memory environments.

Summary Table: Model Format Selection

Model Format Precision Memory Usage Device Requirements Best Use Case
BF16 Highest High BF16-supported GPU/CPUs High-speed inference with reduced memory
F16 High High FP16-supported devices GPU inference when BF16 isn’t available
Q4_K Medium Low Low CPU or Low-VRAM devices Best for memory-constrained environments
Q6_K Medium Moderate CPU with more memory Better accuracy while still being quantized
Q8_0 High Moderate CPU or GPU with enough VRAM Best accuracy among quantized models
IQ3_XS Very Low Very Low Ultra-low-memory devices Extreme memory efficiency and low accuracy
Q4_0 Low Low ARM or low-memory devices llama.cpp can optimize for ARM devices

Included Files & Details

gemma-3-27b-it-bf16.gguf

  • Model weights preserved in BF16.
  • Use this if you want to requantize the model into a different format.
  • Best if your device supports BF16 acceleration.

gemma-3-27b-it-f16.gguf

  • Model weights stored in F16.
  • Use if your device supports FP16, especially if BF16 is not available.

gemma-3-27b-it-bf16-q8_0.gguf

  • Output & embeddings remain in BF16.
  • All other layers quantized to Q8_0.
  • Use if your device supports BF16 and you want a quantized version.

gemma-3-27b-it-f16-q8_0.gguf

  • Output & embeddings remain in F16.
  • All other layers quantized to Q8_0.

gemma-3-27b-it-q4_k.gguf

  • Output & embeddings quantized to Q8_0.
  • All other layers quantized to Q4_K.
  • Good for CPU inference with limited memory.

gemma-3-27b-it-q4_k_s.gguf

  • Smallest Q4_K variant, using less memory at the cost of accuracy.
  • Best for very low-memory setups.

gemma-3-27b-it-q6_k.gguf

  • Output & embeddings quantized to Q8_0.
  • All other layers quantized to Q6_K .

gemma-3-27b-it-q8_0.gguf

  • Fully Q8 quantized model for better accuracy.
  • Requires more memory but offers higher precision.

gemma-3-27b-it-iq3_xs.gguf

  • IQ3_XS quantization, optimized for extreme memory efficiency.
  • Best for ultra-low-memory devices.

gemma-3-27b-it-iq3_m.gguf

  • IQ3_M quantization, offering a medium block size for better accuracy.
  • Suitable for low-memory devices.

gemma-3-27b-it-q4_0.gguf

  • Pure Q4_0 quantization, optimized for ARM devices.
  • Best for low-memory environments.
  • Prefer IQ4_NL for better accuracy.

🚀 If you find these models useful

Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 Network Monitor Assitant.

💬 Click the chat icon (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.

What I'm Testing

I'm experimenting with function calling against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".

🟡 TestLLM – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .

The other Available AI Assistants

🟢 TurboLLM – Uses gpt-4o-mini Fast! . Note: tokens are limited since OpenAI models are pricey, but you can Login or Download the Free Network Monitor agent to get more tokens, Alternatively use the TestLLM .

🔵 HugLLM – Runs open-source Hugging Face models Fast, Runs small models (≈8B) hence lower quality, Get 2x more tokens (subject to Hugging Face API availability)

Gemma 3 model card

Model Page: Gemma

Resources and Technical Documentation:

Terms of Use: Terms

Authors: Google DeepMind

Model Information

Summary description and brief definition of inputs and outputs.

Description

Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 models are multimodal, handling text and image input and generating text output, with open weights for both pre-trained variants and instruction-tuned variants. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions. Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as laptops, desktops or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone.

Inputs and outputs

  • Input:

    • Text string, such as a question, a prompt, or a document to be summarized
    • Images, normalized to 896 x 896 resolution and encoded to 256 tokens each
    • Total input context of 128K tokens for the 4B, 12B, and 27B sizes, and 32K tokens for the 1B size
  • Output:

    • Generated text in response to the input, such as an answer to a question, analysis of image content, or a summary of a document
    • Total output context of 8192 tokens

Usage

Below there are some code snippets on how to get quickly started with running the model. First, install the Transformers library. Gemma 3 is supported starting from transformers 4.50.0.

$ pip install -U transformers

Then, copy the snippet from the section that is relevant for your use case.

Running with the pipeline API

You can initialize the model and processor for inference with pipeline as follows.

from transformers import pipeline
import torch

pipe = pipeline(
    "image-text-to-text",
    model="google/gemma-3-27b-it",
    device="cuda",
    torch_dtype=torch.bfloat16
)

With instruction-tuned models, you need to use chat templates to process our inputs first. Then, you can pass it to the pipeline.

messages = [
    {
        "role": "system",
        "content": [{"type": "text", "text": "You are a helpful assistant."}]
    },
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    }
]

output = pipe(text=messages, max_new_tokens=200)
print(output[0]["generated_text"][-1]["content"])
# Okay, let's take a look! 
# Based on the image, the animal on the candy is a **turtle**. 
# You can see the shell shape and the head and legs.

Running the model on a single/multi GPU

# pip install accelerate

from transformers import AutoProcessor, Gemma3ForConditionalGeneration
from PIL import Image
import requests
import torch

model_id = "google/gemma-3-27b-it"

model = Gemma3ForConditionalGeneration.from_pretrained(
    model_id, device_map="auto"
).eval()

processor = AutoProcessor.from_pretrained(model_id)

messages = [
    {
        "role": "system",
        "content": [{"type": "text", "text": "You are a helpful assistant."}]
    },
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image in detail."}
        ]
    }
]

inputs = processor.apply_chat_template(
    messages, add_generation_prompt=True, tokenize=True,
    return_dict=True, return_tensors="pt"
).to(model.device, dtype=torch.bfloat16)

input_len = inputs["input_ids"].shape[-1]

with torch.inference_mode():
    generation = model.generate(**inputs, max_new_tokens=100, do_sample=False)
    generation = generation[0][input_len:]

decoded = processor.decode(generation, skip_special_tokens=True)
print(decoded)

# **Overall Impression:** The image is a close-up shot of a vibrant garden scene, 
# focusing on a cluster of pink cosmos flowers and a busy bumblebee. 
# It has a slightly soft, natural feel, likely captured in daylight.

Citation

@article{gemma_2025,
    title={Gemma 3},
    url={https://goo.gle/Gemma3Report},
    publisher={Kaggle},
    author={Gemma Team},
    year={2025}
}

Model Data

Data used for model training and how the data was processed.

Training Dataset

These models were trained on a dataset of text data that includes a wide variety of sources. The 27B model was trained with 14 trillion tokens, the 12B model was trained with 12 trillion tokens, 4B model was trained with 4 trillion tokens and 1B with 2 trillion tokens. Here are the key components:

  • Web Documents: A diverse collection of web text ensures the model is exposed to a broad range of linguistic styles, topics, and vocabulary. The training dataset includes content in over 140 languages.
  • Code: Exposing the model to code helps it to learn the syntax and patterns of programming languages, which improves its ability to generate code and understand code-related questions.
  • Mathematics: Training on mathematical text helps the model learn logical reasoning, symbolic representation, and to address mathematical queries.
  • Images: A wide range of images enables the model to perform image analysis and visual data extraction tasks.

The combination of these diverse data sources is crucial for training a powerful multimodal model that can handle a wide variety of different tasks and data formats.

Data Preprocessing

Here are the key data cleaning and filtering methods applied to the training data:

  • CSAM Filtering: Rigorous CSAM (Child Sexual Abuse Material) filtering was applied at multiple stages in the data preparation process to ensure the exclusion of harmful and illegal content.
  • Sensitive Data Filtering: As part of making Gemma pre-trained models safe and reliable, automated techniques were used to filter out certain personal information and other sensitive data from training sets.
  • Additional methods: Filtering based on content quality and safety in line with our policies.

Implementation Information

Details about the model internals.

Hardware

Gemma was trained using Tensor Processing Unit (TPU) hardware (TPUv4p, TPUv5p and TPUv5e). Training vision-language models (VLMS) requires significant computational power. TPUs, designed specifically for matrix operations common in machine learning, offer several advantages in this domain:

  • Performance: TPUs are specifically designed to handle the massive computations involved in training VLMs. They can speed up training considerably compared to CPUs.
  • Memory: TPUs often come with large amounts of high-bandwidth memory, allowing for the handling of large models and batch sizes during training. This can lead to better model quality.
  • Scalability: TPU Pods (large clusters of TPUs) provide a scalable solution for handling the growing complexity of large foundation models. You can distribute training across multiple TPU devices for faster and more efficient processing.
  • Cost-effectiveness: In many scenarios, TPUs can provide a more cost-effective solution for training large models compared to CPU-based infrastructure, especially when considering the time and resources saved due to faster training.
  • These advantages are aligned with Google's commitments to operate sustainably.

Software

Training was done using JAX and ML Pathways.

JAX allows researchers to take advantage of the latest generation of hardware, including TPUs, for faster and more efficient training of large models. ML Pathways is Google's latest effort to build artificially intelligent systems capable of generalizing across multiple tasks. This is specially suitable for foundation models, including large language models like these ones.

Together, JAX and ML Pathways are used as described in the [paper about the Gemini family of models][gemini-2-paper]; "the 'single controller' programming model of Jax and Pathways allows a single Python process to orchestrate the entire training run, dramatically simplifying the development workflow."

Evaluation

Model evaluation metrics and results.

Benchmark Results

These models were evaluated against a large collection of different datasets and metrics to cover different aspects of text generation:

Reasoning and factuality

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 few-shot 28.4 50.9 72.6 77.7
DROP 1-shot 42.4 60.1 72.2 77.2

STEM and code

Benchmark Metric Gemma 3 PT 4B Gemma 3 PT 12B Gemma 3 PT 27B
MMLU 5-shot 59.6 74.5 78.6
MMLU (Pro COT) 5-shot 29.2 45.3 52.2
AGIEval 3-5-shot 42.1 57.4 66.2
MATH 4-shot 24.2 43.3 50.0
GSM8K 8-shot 38.4 71.0 82.6
GPQA 5-shot 15.0 25.4 24.3
MBPP 3-shot 46.0 60.4 65.6
HumanEval 0-shot 36.0 45.7 48.8

Multilingual

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
WMT24++ (ChrF) 36.7 48.4 53.9 55.7
FloRes 29.5 39.2 46.0 48.8
XQuAD (all) 43.9 68.0 74.5 76.8
ECLeKTic 4.69 11.0 17.2 24.4
IndicGenBench 41.4 57.2 61.7 63.4

Multimodal

Benchmark Gemma 3 PT 4B Gemma 3 PT 12B Gemma 3 PT 27B
COCOcap 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
ReMI 27.3 38.5 44.8
AI2D 63.2 75.2 79.0
ChartQA 63.6 74.7 76.3
VQAv2 63.9 71.2 72.9
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

Ethics and Safety

Ethics and safety evaluation approach and results.

Evaluation Approach

Our evaluation methods include structured evaluations and internal red-teaming testing of relevant content policies. Red-teaming was conducted by a number of different teams, each with different goals and human evaluation metrics. These models were evaluated against a number of different categories relevant to ethics and safety, including:

  • Child Safety: Evaluation of text-to-text and image to text prompts covering child safety policies, including child sexual abuse and exploitation.
  • Content Safety: Evaluation of text-to-text and image to text prompts covering safety policies including, harassment, violence and gore, and hate speech.
  • Representational Harms: Evaluation of text-to-text and image to text prompts covering safety policies including bias, stereotyping, and harmful associations or inaccuracies.

In addition to development level evaluations, we conduct "assurance evaluations" which are our 'arms-length' internal evaluations for responsibility governance decision making. They are conducted separately from the model development team, to inform decision making about release. High level findings are fed back to the model team, but prompt sets are held-out to prevent overfitting and preserve the results' ability to inform decision making. Assurance evaluation results are reported to our Responsibility & Safety Council as part of release review.

Evaluation Results

For all areas of safety testing, we saw major improvements in the categories of child safety, content safety, and representational harms relative to previous Gemma models. All testing was conducted without safety filters to evaluate the model capabilities and behaviors. For both text-to-text and image-to-text, and across all model sizes, the model produced minimal policy violations, and showed significant improvements over previous Gemma models' performance with respect to ungrounded inferences. A limitation of our evaluations was they included only English language prompts.

Usage and Limitations

These models have certain limitations that users should be aware of.

Intended Usage

Open vision-language models (VLMs) models have a wide range of applications across various industries and domains. The following list of potential uses is not comprehensive. The purpose of this list is to provide contextual information about the possible use-cases that the model creators considered as part of model training and development.

  • Content Creation and Communication
    • Text Generation: These models can be used to generate creative text formats such as poems, scripts, code, marketing copy, and email drafts.
    • Chatbots and Conversational AI: Power conversational interfaces for customer service, virtual assistants, or interactive applications.
    • Text Summarization: Generate concise summaries of a text corpus, research papers, or reports.
    • Image Data Extraction: These models can be used to extract, interpret, and summarize visual data for text communications.
  • Research and Education
    • Natural Language Processing (NLP) and VLM Research: These models can serve as a foundation for researchers to experiment with VLM and NLP techniques, develop algorithms, and contribute to the advancement of the field.
    • Language Learning Tools: Support interactive language learning experiences, aiding in grammar correction or providing writing practice.
    • Knowledge Exploration: Assist researchers in exploring large bodies of text by generating summaries or answering questions about specific topics.

Limitations

  • Training Data
    • The quality and diversity of the training data significantly influence the model's capabilities. Biases or gaps in the training data can lead to limitations in the model's responses.
    • The scope of the training dataset determines the subject areas the model can handle effectively.
  • Context and Task Complexity
    • Models are better at tasks that can be framed with clear prompts and instructions. Open-ended or highly complex tasks might be challenging.
    • A model's performance can be influenced by the amount of context provided (longer context generally leads to better outputs, up to a certain point).
  • Language Ambiguity and Nuance
    • Natural language is inherently complex. Models might struggle to grasp subtle nuances, sarcasm, or figurative language.
  • Factual Accuracy
    • Models generate responses based on information they learned from their training datasets, but they are not knowledge bases. They may generate incorrect or outdated factual statements.
  • Common Sense
    • Models rely on statistical patterns in language. They might lack the ability to apply common sense reasoning in certain situations.

Ethical Considerations and Risks

The development of vision-language models (VLMs) raises several ethical concerns. In creating an open model, we have carefully considered the following:

  • Bias and Fairness
    • VLMs trained on large-scale, real-world text and image data can reflect socio-cultural biases embedded in the training material. These models underwent careful scrutiny, input data pre-processing described and posterior evaluations reported in this card.
  • Misinformation and Misuse
    • VLMs can be misused to generate text that is false, misleading, or harmful.
    • Guidelines are provided for responsible use with the model, see the Responsible Generative AI Toolkit.
  • Transparency and Accountability:
    • This model card summarizes details on the models' architecture, capabilities, limitations, and evaluation processes.
    • A responsibly developed open model offers the opportunity to share innovation by making VLM technology accessible to developers and researchers across the AI ecosystem.

Risks identified and mitigations:

  • Perpetuation of biases: It's encouraged to perform continuous monitoring (using evaluation metrics, human review) and the exploration of de-biasing techniques during model training, fine-tuning, and other use cases.
  • Generation of harmful content: Mechanisms and guidelines for content safety are essential. Developers are encouraged to exercise caution and implement appropriate content safety safeguards based on their specific product policies and application use cases.
  • Misuse for malicious purposes: Technical limitations and developer and end-user education can help mitigate against malicious applications of VLMs. Educational resources and reporting mechanisms for users to flag misuse are provided. Prohibited uses of Gemma models are outlined in the Gemma Prohibited Use Policy.
  • Privacy violations: Models were trained on data filtered for removal of certain personal information and other sensitive data. Developers are encouraged to adhere to privacy regulations with privacy-preserving techniques.

Benefits

At the time of release, this family of models provides high-performance open vision-language model implementations designed from the ground up for responsible AI development compared to similarly sized models.

Using the benchmark evaluation metrics described in this document, these models have shown to provide superior performance to other, comparably-sized open model alternatives.

Downloads last month
1,732
GGUF
Model size
27B params
Architecture
gemma3

1-bit

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Mungert/gemma-3-27b-it-GGUF

Quantized
(14)
this model