Skip to content
Audio

balbal-v2-manas

турецкий и английский ASR быстро на CPU

2 min read

balbal-v2-manas — большая версия ASR-модели от erayyapagci для турецкого и английского языков. Это FastConformer-CTC на 313,3 млн параметров с looped refiner и product-key акустической памятью, декодирование — 6-gram token KenLM. Обучена примерно на 7 000 часах аудио и рассчитана на CPU.

Ключевые особенности

  • Поддерживает турецкий и английский; обучена примерно на 7 000 часах аудио.
  • Работает на CPU: 123× быстрее реального времени на 8 потоках.
  • FastConformer-CTC, 18 слоёв, плюс looped refiner с 3 проходами и product-key акустическая память.
  • Выдаёт текст с регистром, пунктуацией и таймстемпами по словам; числа — цифрами через ITN.
  • Есть mem-efficient вариант: память сокращена с 1 000 000 до 250 000 слотов, но только для KenLM-декодера.

Пример

import balbal
model = balbal.load("balbal-v2-manas")
print(model.transcribe("speech.wav")["text"])
ПодробнееMu-Bench Turkish: 8,20% WER / 14,89% sigWER (официальный harness, GPT-4.1 judge); FLEURS Turkish: 4,06% WER; English Open ASR Leaderboard: 8,67% WER в среднем.
313,3 млн параметров, из них 192,0 млн — таблица памяти; PyTorch-веса 1,2 ГБ, быстрый ONNX-граф 353 МБ; Linux x86-64, Python 3.11/3.12.
Ограничения: на очень коротких клипах больше ошибок и возможно смешение турецкого с английским; длинное аудио режется на 30-секундные чанки, диаризации нет.
Лицензия balbal Community License 1.1: бесплатно для исследований и некоммерческого использования; малому бизнесу — при условиях и с «Powered by balbal».
Discuss Telegram 16 201

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-v2-manas
    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-v2-manas --local-dir ~/models/balbal-v2-manas
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to balbal-v2-manas
  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 stt_en_fastconformer_ctc_large (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 2All on Hugging Face
Fine-tunes 6All on Hugging Face

About the model

erayyapagci/balbal-v2-manas
Author
erayyapagci
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

20116

Ask AI about this model

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

Comments