Modelpublic

Vana-Labs/stt-whisper

sourceHugging Facemitupdated 12h agoView on Hugging Face

0likes

Model Card

stt-whisper

Whisper models in whisper.cpp (GGML) format for Tetro, the local meeting transcriber by Vana Labs. They run fully on your computer.

This repository is a pinned mirror, so Tetro's downloads don't depend on third-party hosts. Each file is named ggml-<model id>.bin, where the model ID is the name Tetro uses everywhere.

Model ID	Notes
`tiny`, `base`, `small`, `medium`	OpenAI Whisper, full precision
`large-v3`, `large-v3-turbo`	OpenAI Whisper, full precision
`tiny-q5_1`, `base-q5_1`, `small-q5_1`	5-bit quantized
`medium-q5_0`, `large-v3-q5_0`, `large-v3-turbo-q5_0`	5-bit quantized
`large-v3-turbo-hy`	Armenian fine-tune; also keeps English and Russian words

Sources and licenses

—OpenAI Whisper files: unchanged copies from ggerganov/whisper.cpp (commit 5359861c739e955e79d9a303bcbc70fb988958b1). The Whisper weights are © OpenAI, MIT license.
—`large-v3-turbo-hy`: Chillarmo/whisper-large-v3-turbo-armenian (Apache-2.0), converted to GGML with whisper.cpp's conversion script.