whisper-large-v3-french-distil-dec16-ggml
Quantizations of the authors' own whisper.cpp f16 conversion of bofenghuang/whisper-large-v3-french-distil-dec16, in q4_0 / q5_0 / q8_0 (phones use q4/q5, Macs q8).
Source checkpoint: bofenghuang/whisper-large-v3-french-distil-dec16 by bofenghuang · License: mit (unchanged; this repo only re-packages the weights)
Engine: Load with whisper.cpp (whisper-cli -m <file>), or any app that embeds it.
Files
| File | Quantization | Size | Note |
|---|---|---|---|
ggml-whisper-large-v3-french-distil-dec16-q8_0.bin |
q8_0 | 1209 MB | |
ggml-whisper-large-v3-french-distil-dec16-q5_0.bin |
q5_0 | 791 MB | |
ggml-whisper-large-v3-french-distil-dec16-q4_0.bin |
q4_0 | 652 MB |
f16 is the lossless conversion; q8_0 is nearly identical in accuracy at ~55 % of the size; q5_0/q5_k are the phone-friendly choice; q4_* is smallest with a small accuracy cost.
How these were made
Converted from the upstream checkpoint with the engine's own converter, then quantized with the engine's quantizer. Each variant was checked by transcribing short Portuguese and English samples before upload.
Attribution
Weights are derivative works of the upstream model and keep its license. Please cite the original authors (bofenghuang). The conversion and hosting here are maintained by JoaoZaokk so that the download links used by the Odysseus / Open WebUI native apps stay stable. No warranty.