Downloads · 30 days
121
33% of all-time downloads
gcoli/whisper-large-v3-swiss-german-mlx-q4
whisper-large-v3-swiss-german-mlx-q4 is a automatic speech recognition model from gcoli. Use it when you need speech turned into text. It is set up for mlx. The card lists the license as other.
Provenance: this checkpoint is openai/whisper-large-v3 with the LoRA adapter Flurin17/whisper-large-v3-peft-swiss-german merged into the weights (see mergeprovenance.json). The licence terms of that adapter and its tr…
Downloads · 30 days
121
33% of all-time downloads
All-time downloads
365
Public
Parameters
1.5B
878 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors878 MB · 100%
How the weights are stored.
U321.5B · 100%
From the Hugging Face model README
Provenance: this checkpoint is
openai/whisper-large-v3with the LoRA adapterFlurin17/whisper-large-v3-peft-swiss-germanmerged into the weights (seemerge_provenance.json). The licence terms of that adapter and its training data apply.
Q4 quantized MLX conversion of whisper-large-v3-swiss-german, derived from
openai/whisper-large-v3 plus a Swiss German PEFT adaptation trained on
SwissDial data.
This repository is intended for local speech-to-text inference with MLX-based Whisper runtimes, including oMLX through its mlx-audio STT integration.
model.safetensors with MLX tensor namesDownload the repository and point your MLX/oMLX Whisper STT engine at the local model directory or Hugging Face repo:
huggingface-cli download --local-dir whisper-large-v3-swiss-german-mlx-q4 gcoli/whisper-large-v3-swiss-german-mlx-q4
For oMLX, configure the model as a Whisper/STT model. Exact routing depends on the oMLX version and deployment config.
SwissDial CC BY-NC 4.0 with no-reidentification condition
The Swiss German adaptation is based on SwissDial data. Use of this model is therefore restricted to non-commercial purposes under Creative Commons Attribution-NonCommercial 4.0 International, plus the SwissDial condition that users must not attempt to determine the identity of speakers in the SwissDial dataset.
Additional practical terms:
The Whisper base model remains under its original Apache-2.0 license. The combined adapted checkpoint is distributed under the more restrictive terms above because of the SwissDial-trained adaptation.
This Q4 model was produced from the local FP16 MLX checkpoint by quantizing MLX Linear and Embedding weights to 4-bit affine groups. Conv, LayerNorm, positional embeddings and bias vectors remain dense.
See merge_provenance.json for the base and adapter source metadata inherited
from the FP16 conversion.
The repository contains SHA256SUMS for model.safetensors and config.json.
Structural validation checks:
config.jsonNo benchmark numbers are claimed for this Q4 variant until dialect-specific WER/CER evaluation has been run on a held-out Swiss German test set.