Skip to content
Audio

Whisper large-v3 Core ML (8-bit)

local speech-to-text on Apple devices at half the size

2 min readupdated

This is a Core ML conversion of the popular openai/whisper-large-v3 speech recognition model built for the WhisperKit framework. Its weights are quantized to 8-bit (compressed for faster execution and lower memory use), letting it run on the Apple Neural Engine at roughly 1.56 GB instead of the standard float16 version's 3.1 GB. It handles local transcription (converting audio to text) without sending recordings to the cloud while preserving the original model's accuracy. It is aimed at macOS, iOS, and iPadOS developers who need fast, high-quality on-device speech recognition.

Key features

  • Compact footprint: takes up roughly 1.56 GB in total (648 MB for the audio encoder and 910 MB for the text decoder), half the size of the float16 version.
  • Fast performance on Mac: averages a 0.21 RTF (processing time per second of audio), compared to 0.55 for the float16 build.
  • Virtually no accuracy loss: averages an 8.0% WER across 10 test sets, versus 7.9% for the original float16 version.
  • Production-proven: already powers the Sagt app on Mac, iPhone, and iPad.

What you can do

  • Build local voice recorders or voice memo transcription into iPhone, iPad, and Mac apps using the WhisperKit library.
  • Add offline transcription for calls, interviews, and podcasts in English, German, Dutch, Spanish, and French with zero cloud API costs.

Example

let folder = try await WhisperKit.download(
    variant: "openai_whisper-large-v3_8bit",
    from: "Selbyggen/whisper-large-v3-coreml"
)
let kit = try await WhisperKit(WhisperKitConfig(modelFolder: folder.path))
let result = try await kit.transcribe(audioPath: "recording.m4a")
Read moreDetails
Average WER across 10 datasets (FLEURS and VoxPopuli for en, de, nl, es, fr): 8.0% (vs. 7.9% for float16 and 8.7% for large-v3 turbo).
Speed: average RTF on Mac is 0.21 (lower is faster; float16 is 0.55).
Quantization: per-channel linear symmetric int8 via coremltools (8-bit weights, 16-bit activations, float16 mel spectrogram).
Compatibility: Apple devices supporting Core ML / Apple Neural Engine via WhisperKit; Apache 2.0 license.
Discuss Telegram 0 16

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.8 GB · ready files1.4 GB
  • FP8 / INT8recommendedweights 0.4 GB1.0 GB
  • 4-bitweights 0.2 GB0.8 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Selbyggen/whisper-large-v3-coreml
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Selbyggen/whisper-large-v3-coreml --local-dir ~/models/whisper-large-v3-coreml
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of whisper-large-v3 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 3All on Hugging Face

About the model

Selbyggen/whisper-large-v3-coreml
Author
Selbyggen
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

160

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments