whisper-large-v3-ggml

Quantized ggml build of OpenAI's Whisper Large v3 model. This repository is intended to host GitHub releases so the model can be fetched globally without relying on Hugging Face mirrors.

Files expected in releases

Filename	Description
ggml-large-v3-q8_0.bin	Q8 quantized version produced from ggml-large-v3.bin using whisper.cpp's quantize q8_0.
checksums.txt	SHA-256 digest for every release artifact.
CONVERSION.md	Conversion + quantization log.

Download

RELEASE_URL="https://github.com/sergheinenov/whisper-large-v3-ggml/releases/download/v1.0.0"
curl -L -o ggml-large-v3-q8_0.bin "$RELEASE_URL/ggml-large-v3-q8_0.bin"
curl -L -o checksums.txt "$RELEASE_URL/checksums.txt"
shasum -a 256 -c checksums.txt

A helper shell script (download.sh) automates the same steps with checksum verification.

Reproducing the quantization

See CONVERSION.md for the precise commands used to fetch, quantize, and checksum the model.

License

The original Whisper weights are released by OpenAI under the MIT License. This derived artifact inherits those terms. See LICENSE for the license text.

Verification

Published checksums are generated via:

shasum -a 256 ggml-large-v3-q8_0.bin > checksums.txt

Always verify the checksum before loading the model locally.

Support

Issues and pull requests are welcome once the repository is published under github.com/sergheinenov.