Skip to content
Audio

Moonshine Base Welsh Compact

валлийский ASR на 53 млн параметров

1 min read

Компактная версия валлийского распознавателя rebreda/moonshine-base-welsh, дообученного из Moonshine Base. 53.2M параметров, веса в fp16, скачивание 106 MB. Меньше и точнее предыдущей версии за счёт трёх правок весов без дополнительного обучения.

Ключевые особенности

  • 53.2M параметров вместо 61.5M; 106 MB вместо 246 MB.
  • Три правки без дообучения: усреднение чекпоинтов LoRA, отключение пяти голов decoder self-attention и урезанная токен-таблица до 12,735 токенов.
  • Отключение голов снизило WER на Siarad с 87.0 до 76.4.
  • Вход — 16 kHz mono, короткие фразы; выход — текст с регистром и пунктуацией.
  • Для диктовки, голосовых команд, чтения и подкастов на валлийском; английские слова внутри валлийской речи.

Пример

from transformers import pipeline

asr = pipeline("automatic-speech-recognition", model="rebreda/moonshine-base-welsh-compact")
print(asr("clip.wav", generate_kwargs={"num_beams": 5})["text"])
ПодробнееWER: Common Voice 25.8, FLEURS Welsh 37.6, Siarad 76.8 (beam 5, без length cap).
Среднее по Common Voice, FLEURS Welsh, Arfor и Siarad — 40.1 против 43.5 у прошлой версии.
Английский ухудшился: 21.0 WER против 16.6; для чистого английского лучше базовая версия.
Вес 106 MB в fp16; обучалась на одной AMD Radeon 8060S (ROCm, bf16). Siarad — свободная речь с code-switching, для неё модель всё ещё слаба; длинные записи резать на короткие сегменты.
Discuss Telegram 0 23

Will it run?

Runs in F16

Fits without quantisation.

  • F16weights 0.1 GB · ready files0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 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 · F16 — 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.

    # rebreda/moonshine-base-welsh-compact
    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 rebreda/moonshine-base-welsh-compact --local-dir ~/models/moonshine-base-welsh-compact
    # 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 moonshine-base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 9All on Hugging Face
Fine-tunes 9All on Hugging Face

About the model

rebreda/moonshine-base-welsh-compact
Author
rebreda
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

230

Ask AI about this model

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

Comments