Install from WinGet (Windows)
winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf dollarland/dolyzer-gemma-models:Q4_K_M# Run inference directly in the terminal:
llama cli -hf dollarland/dolyzer-gemma-models:Q4_K_MBuild from source code
git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf dollarland/dolyzer-gemma-models:Q4_K_M# Run inference directly in the terminal:
./build/bin/llama cli -hf dollarland/dolyzer-gemma-models:Q4_K_MUse Docker
docker model run hf.co/dollarland/dolyzer-gemma-models:Q4_K_MDolyzer Gemma Models
Welcome to the official Gemma model repository for Dolyzer, an offline-first mobile AI application.
This repository hosts selected models from Google's Gemma family, including supported quantized and converted variants intended for on-device AI inference.
Our goal is to make compatible Gemma models accessible for local AI applications while preserving their original licensing requirements, attribution, and responsible-use restrictions.
About This Repository
Repository: dollarland/dolyzer-gemma-models
Maintained by: Dolyzer
Original model developer: Google
Model family: Gemma
Supported file formats: GGUF and MediaPipe/LiteRT .task, where available and compatible.
Primary application: Dolyzer Mobile AI
Key Features
- On-device AI inference
- Offline operation after downloading
- Support for multiple Gemma model variants
- Mobile-oriented quantized models
- User-controlled model downloads and storage
- Versioned models and future updates
Model performance, hardware support, and compatibility depend on the selected model and inference runtime.
Available Models
The repository is designed to accommodate additional Gemma models over time.
| Model | Format | Quantization | Status |
|---|---|---|---|
| Gemma 2 2B Instruct | GGUF | Q4_K_M | Add when uploaded and verified |
Additional models will be documented here when they become available.
Future additions may include different Gemma model sizes, quantizations, and supported inference formats, subject to compatibility testing and licensing requirements.
Model Sources and Attribution
Gemma models are developed by Google.
Official resources:
Dolyzer does not claim ownership of the original Gemma architectures or model weights.
For each redistributed model, its upstream source, conversion method, modifications, quantization, and applicable terms should be recorded.
Using Models in Dolyzer
Dolyzer is designed to allow users to download supported AI models and execute inference locally on compatible devices.
General workflow:
- Open Dolyzer's AI model library.
- Select an available Gemma model.
- Review its storage, compatibility, and licensing information.
- Download and verify the selected model.
- Activate it using the compatible on-device inference engine.
- Use the model offline after installation.
GGUF files require a compatible GGUF inference engine. MediaPipe .task files require a compatible MediaPipe/LiteRT runtime. These formats are not interchangeable.
Availability inside Dolyzer depends on successful integration and device compatibility testing.
License and Terms of Use
License: Google Gemma Terms of Use
The Gemma models in this repository are governed by the applicable Google Gemma Terms of Use, not the Apache 2.0 License.
Official terms:
https://ai.google.dev/gemma/terms
Prohibited Use Policy:
https://ai.google.dev/gemma/prohibited_use_policy
The Gemma Terms contain requirements governing use, modification, redistribution, notices, and prohibited uses.
Where applicable, distributions must include the complete agreement, the required notice, and enforceable use restrictions.
See the accompanying LICENSE.txt and NOTICE files when provided.
Different model generations may have different license terms. Each additional model must be reviewed against its applicable license before distribution.
Model Modifications and Quantization
Some models may be converted or quantized to support local inference on mobile devices.
Quantization can reduce storage and memory requirements but may affect model quality, performance, and compatibility.
Modified model files must have appropriate modification notices and provenance documentation.
Conversion or quantization does not remove the applicable obligations under Google's model license.
Limitations
- Model responses may be inaccurate or incomplete.
- Models may generate misleading or inappropriate content.
- Runtime memory usage can exceed model file size.
- Performance varies by device, available memory, processor, and context length.
- Not every model is compatible with every phone or inference runtime.
- Users remain responsible for appropriate use of generated content.
Future Model Additions
This repository may expand to include additional Gemma models and optimized variants.
Each new model entry should document:
- Model name and generation
- Original source and developer
- File format and quantization
- File size and checksum
- Supported inference runtime
- Tested hardware requirements
- Model version and release date
- Conversion details and modifications
- Applicable license and required notices
Disclaimer
Dolyzer is an independent application and is not affiliated with, sponsored by, or endorsed by Google.
Gemma models and related intellectual property remain subject to the applicable rights and terms of their respective owners.
Maintained by Dolyzer
Building accessible, offline-first AI experiences for mobile devices.
- Downloads last month
- 18
4-bit
Install (macOS, Linux)
# Start a local OpenAI-compatible server with a web UI: llama serve -hf dollarland/dolyzer-gemma-models:Q4_K_M# Run inference directly in the terminal: llama cli -hf dollarland/dolyzer-gemma-models:Q4_K_M