Skip to content
Audio

balbal-v2e-zellige

распознавание марокканского дариджа на CPU

1 min readupdated

Экспериментальный превью-релиз и первая арабская модель линейки balbal. Это ASR для марокканского дариджа на базе FastConformer-CTC (307M параметров), инициализированная из Sabeq-ASR, с KenLM-декодированием. Демо доступно в браузере: erayy.com/zellige.

Ключевые особенности

  • 307M параметров; около 100× быстрее реального времени на 8 потоках CPU, включая KenLM-декодирование.
  • На тесте Casablanca MA: 42.85% WER / 14.11% CER; среднее по трём наборам — 44.38 / 17.74.
  • Ниже WER/CER, чем у более крупных открытых моделей Cohere Transcribe Arabic, Jais-ASR и MoulSot v0.3 на всех трёх тестах.
  • Архитектура: FastConformer-CTC, 17 слоёв, d_model 512, 80 mel, 8× subsampling; looped refiner с 3 проходами и product-key memory.
  • Выдаёт текст арабской графикой с таймстемпами по словам; французские слова тоже пишутся арабской графикой.
  • Есть PyTorch и ONNX-веса; лицензия CC BY-NC-SA 4.0 — только некоммерческое использование.

Пример

import balbal
model = balbal.load("balbal-v2e-zellige")
print(model.transcribe("speech.wav")["text"])
ПодробнееСредний WER/CER: 44.38 / 17.74%; Casablanca MA: 42.85 / 14.11%.
CPU: ~100× быстрее реального времени на 8 потоках AMD Ryzen AI Max+ 395 с KenLM.
Ограничения: только марокканский дариджа; другие диалекты и MSA работают хуже; аудио режется на 30-секундные фрагменты; стриминга пока нет.
Лицензия весов CC BY-NC-SA 4.0, производные — под той же лицензией.
Discuss Telegram 3 37

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.2 GB · ready files1.8 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 · 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.

    # erayyapagci/balbal-v2e-zellige
    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 erayyapagci/balbal-v2e-zellige --local-dir ~/models/balbal-v2e-zellige
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to balbal-v2e-zellige
  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 Sabeq-ASR (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

erayyapagci/balbal-v2e-zellige
Author
erayyapagci
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

373

Ask AI about this model

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

Comments