distil-whisper-large-v3-ct2-bfloat16

🗣️ Distil-Whisper Large-v3 — CTranslate2 (bfloat16)

This is HuggingFace's distil-large-v3 converted to CTranslate2 format with bfloat16 precision.

[!TIP] Also available in other precisions: float32 · float16

---

📋 Details

---

⚡ Quick Start

Install the inference library:

pip install whisper-s2t-reborn

Transcribe an audio file:

import whisper_s2t

model = whisper_s2t.load_model(
    model_identifier="distil-large-v3",
    compute_type="bfloat16",
    device="cuda",
)

result = model.transcribe_with_vad(
    ["audio.wav"],
    lang_codes=["en"],
    tasks=["transcribe"],
    initial_prompts=[None],
    batch_size=1,  # increase this to significantly improve throughput
)

for segment in result[0]:
    print(segment["text"])

[!NOTE] Models are auto-downloaded from this repo the first time you run inference. No manual download required.

See the whisper-s2t-reborn repository for the full list of available parameters.

---

📦 All Available CTranslate2 Whisper Models

Every model below is hosted at huggingface.co/ctranslate2-4you and works with whisper-s2t-reborn.

🌍 Standard Whisper (Multilingual)

🇺🇸 Whisper English-Only

⚡ Distilled Whisper

🚀 Whisper Large-v3 Turbo

---

🔗 Links

• 📦 Inference library — whisper-s2t-reborn
• 🏗️ CTranslate2 — github.com/OpenNMT/CTranslate2
• 🧠 Original model — distil-whisper/distil-large-v3