Whisper models in ggml format, mirrored for Dictum
This repository holds the Whisper files that Dictum offers to download. They are byte-for-byte copies of
ggerganov/whisper.cpp at commit 5359861.
Nothing here was converted, quantized or changed.
| File | Model | Size | SHA-256 |
|---|---|---|---|
ggml-large-v3-turbo-q8_0.bin |
Whisper large-v3-turbo, 8-bit | 874,188,075 bytes | 317eb69c11673c9de1e1f0d459b253999804ec71ac4c23c17ecf5fbe24e259a1 |
ggml-medium-q8_0.bin |
Whisper medium, 8-bit | 823,369,779 bytes | 42a1ffcbe4167d224232443396968db4d02d4e8e87e213d3ee2e03095dea6502 |
ggml-small-q8_0.bin |
Whisper small, 8-bit | 264,464,607 bytes | 49c8fb02b65e6049d5fa6c04f81f53b867b5ec9540406812c643f177317f779f |
ggml-tiny.en.bin |
Whisper tiny, English only | 77,704,715 bytes | 921e4cf8686fdd993dcd081a5da5b6c365bfde1162e72b08d75ac75289920b1f |
The files run with whisper.cpp and the libraries built on it, such as Whisper.net.
Licence
The Whisper weights are © 2022 OpenAI. The conversion to ggml is © the whisper.cpp authors. Both are released under the MIT licence; the two texts are in LICENSE.
Why a mirror
Dictum pins each download to a commit of this repository and checks the file's SHA-256 before it uses the file. The mirror keeps those links working whatever happens upstream. Its history is never rewritten, so a link pinned to a commit stays valid.