MBZUAI · MobiLlama
MobiLlama 0.5B
A transparent 0.5B text model with shared feed-forward layers and published Snapdragon 685 measurements.
- Parameters
- 520M
- Input
- Text
- Output
- Text
- Access
- Open source
- Licence
- Apache 2.0
- Released
- Device evidence
- Measured on device
- Last verified
- Formats
- Safetensors, GGUF
- Runtimes
- llama.cpp, Transformers
- Platforms
- Android, Linux, macOS, Windows
The authors report 7.02 tokens per second and 770 MB of RAM for a 4-bit GGUF build on a Snapdragon 685 phone. The family also includes a 0.8B variant.