Skip to content
Audio

Tynda-ASR-kk-ru

казахский и русский ASR на базе Nemotron

1 min readupdated

Это файнтюн NVIDIA Nemotron-3.5-Transcribe-30B-A4B от Raushan Baisalova (University of Waterloo): к базовой модели добавлен казахский, русский сохранён. Около 33,7 млрд параметров, обучение — на публичных данных, примерно 1470 часов речи.

Ключевые особенности

  • Казахский и русский в одной ASR-модели; для казахского карточка рекомендует prompt с пунктуацией.
  • LoRA r=64, α=128 на attention, Mamba и MLP-проекциях, плюс полный ASR-encoder и audio projection.
  • Обучение: KSC2 train (~1154 ч), FLEURS kk train (×3), Golos train (~249 ч); тестовые сплиты не использовались.
  • Сохранены CTC word-timestamps и diarization branch от базовой модели.
  • WER/CER на FLEURS: kk 12,23/4,47%, ru 9,21/6,37%.

Пример

import base64
from openai import OpenAI
client = OpenAI(base_url='http://127.0.0.1:8000/v1', api_key='EMPTY')
audio = base64.b64encode(open('audio.wav', 'rb').read()).decode()
prompt = 'Audio transcription. Language: Kazakh. Wording: verbatim. Formatting: punctuation and capitalization. Output: transcript only.'
r = client.chat.completions.create(model='tynda', temperature=0, max_tokens=2048,
    messages=[{'role': 'user', 'content': [{'type': 'text', 'text': prompt}, {'type': 'audio_url', 'audio_url': {'url': f'data:audio/wav;base64,{audio}'}}]}])
print(r.choices[0].message.content)
ПодробнееWER/CER на FLEURS test: kk — 12,23/4,47%, ru — 9,21/6,37% (нормализованные метрики).
BF16-веса занимают 63 ГБ; нужен один GPU на 80–141 ГБ (H100 / H200 / RTX PRO 6000).
Ограничения: разговорный казахский с хезитациями, русские данные обучались без пунктуации, числа обычно словами, code-switching отдельно не тренировали.
Discuss Telegram 1 164

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # baisalovar/Tynda-ASR-kk-ru
    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 baisalovar/Tynda-ASR-kk-ru --local-dir ~/models/Tynda-ASR-kk-ru
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Tynda-ASR-kk-ru
  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 Tynda-ASR-kk-ru (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

baisalovar/Tynda-ASR-kk-ru
Author
baisalovar
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

1641

Ask AI about this model

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

Comments