Whisper Tiny (GGML Binary for RunSLM AI)
This repository provides the official quantized binary distribution of OpenAI's Whisper Tiny model converted into GGML format (ggml-tiny.bin) for on-device automatic speech recognition (ASR) inside RunSLM AI and whisper.cpp runtimes.
Model Summary
- Architecture: Whisper (Sequence-to-Sequence Encoder-Decoder)
- Parameters: ~39 Million
- File Format: GGML Binary (
.bin) - File Size: ~77.7 MB
- License: MIT License
- Target Deployment: Local mobile & tablet hardware (iOS Metal & Android Arm64 NEON)
Features
- 100% Offline Speech Recognition: Transcribe speech directly on device hardware with zero network connectivity.
- Multilingual Support: Supports multilingual speech transcription across major languages including English, Cantonese, Mandarin, Japanese, Spanish, French, and German.
- Low Memory Footprint: Requires only ~75 MB of device RAM, making it suitable for older devices and background execution.
Direct Download
The raw model binary can be downloaded directly from:
https://huggingface.co/RunSLM-AI/whisper-tiny-ggml/resolve/main/ggml-tiny.bin
Acknowledgements & Credits
- Original Model: OpenAI Whisper by OpenAI (MIT License).
- GGML Conversion & C++ Runtime: whisper.cpp by Georgi Gerganov and contributors (MIT License).
- Distribution Maintainer: PROJECTLAB LIMITED / RunSLM-AI.