whisper-large-v3-turbo-mlx

whisper-large-v3-turbo, MLX format (float16)

OpenAI's whisper-large-v3-turbo, converted to MLX format. The weights are unchanged — this is a format conversion only. No fine-tuning, no quantization, no distillation.

Converted for Thematic Thing, an on-device qualitative-analysis application, so that the exact artefact it downloads is one with a documented, unbroken licence chain from the original MIT-licensed release.

What this is

Quantized Whisper builds degrade sharply on long-form audio while appearing fine on short clips. This build is deliberately unquantized.

Usage

import mlx_whisper
result = mlx_whisper.transcribe(audio, path_or_hf_repo="clementrahula/whisper-large-v3-turbo-mlx")

Licence and attribution

The model weights are the work of OpenAI, released under the MIT License:

Copyright (c) 2022 OpenAI

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR OTHER DEALINGS IN THE SOFTWARE.

The conversion tooling is MIT, Copyright (c) 2023 Apple Inc. Both licence texts are included in this repository as LICENSE-openai-whisper-MIT.txt and LICENSE-mlx-examples-MIT.txt.

Modification statement (MIT/attribution): the only change is tensor layout and naming to MLX's format. Numerical values are unmodified.