Skip to content
Audio

norwegian-parakeet

норвежский speech-to-text на открытых данных

1 min read

Модель от ahdrage для распознавания норвежской речи (букмол, nb/no) на базе moondream/parakeet-ultra — посттренированного NVIDIA Parakeet TDT 0.6B v3. Дообучена только на речи с открытыми лицензиями: около 1 770 часов уникального аудио. Пишет текст с регистром и пунктуацией, декодирует жадно, без языковой модели.

Ключевые особенности

  • 0,63B параметров, архитектура Parakeet TDT; в обучении участвовали все параметры.
  • Около 1 770 ч уникального аудио; источники и лицензии перечислены в docs/DATA.md.
  • Четыре прохода обучения с понижением learning rate; итоговый WER улучшался от прохода к проходу.
  • Обучение шло на одной RTX PRO 4000 Blackwell (24 ГБ), суммарно около 35 часов.
  • Клипы до 30 с; длинные записи нужно сначала нарезать — это не тестировалось.

Пример

import torch
from transformers import AutoProcessor, ParakeetForTDT

processor = AutoProcessor.from_pretrained("ahdrage/norwegian-parakeet")
model = ParakeetForTDT.from_pretrained("ahdrage/norwegian-parakeet").eval()
inputs = processor(audio, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    out = model.generate(**inputs)
print(processor.batch_decode(out.sequences, skip_special_tokens=True)[0])
ПодробнееWER: Tuva 3,1%; NST 4,2%; NPSC 6,1%; FLEURS 7,4%; NB Samtale 17,7%.
Оценка на клипах до 30 с; на Stortinget и NorGovPCC транскрипты отредактированы, поэтому WER завышен.
Веса — CC BY 4.0, код — Apache-2.0; для коммерческого использования есть нерешённые вопросы по лицензиям NST и Stortinget.
Discuss Telegram 0 21

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 2.3 GB · ready files3.1 GB
  • BF16weights 1.2 GB1.8 GB
  • FP8 / INT8recommendedweights 0.6 GB1.1 GB
  • 4-bitweights 0.3 GB0.9 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 · F32 — 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.

    # ahdrage/norwegian-parakeet
    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 ahdrage/norwegian-parakeet --local-dir ~/models/norwegian-parakeet
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to norwegian-parakeet
  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 parakeet-ultra (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

ahdrage/norwegian-parakeet
Author
ahdrage
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

210

Ask AI about this model

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

Comments

Audio

Swanada

Swanada