Skip to content
Audio

NVIDIA Nemotron 3.5 Transcribe

ASR на 40 языках с управляемым стилем

1 min readupdated

Открытая офлайн-модель NVIDIA для распознавания речи и аудио: транскрипция, перевод, диаризация, таймкоды и ответы на вопросы по записи. Гибридная MoE-архитектура Mamba-2/Transformer: 34B параметров, около 4.3B активных вместе с аудиоэнкодером, контекст 65 536 токенов и записи до часа.

Ключевые особенности

  • 40 языков, включая русский и украинский; транскрипция в verbatim или читаемом виде.
  • Управление стилем промптом: контекстные термины, спикеры, таймкоды, перевод речи.
  • Длинные записи до 1 часа и контекст 65 536 токенов.
  • Варианты BF16, FP8 и NVFP4; запуск через vLLM/NeMo Speech или Transformers.
  • Режим рассуждения включается через chat template для аудио-вопросов.

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()
r = client.chat.completions.create(model="nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B", temperature=0,
    messages=[{"role": "user", "content": [{"type": "text", "text": "Transcribe the following audio."},
    {"type": "audio_url", "audio_url": {"url": f"data:audio/wav;base64,{audio}"}}]}])
print(r.choices[0].message.content)
ПодробнееБенчмарки: средний cpWER 15.80 на мультиспикерном ASR; WER 4.71 на английском Open ASR Leaderboard и 4.62 на мультиязычном.
Запуск: 1× H100 80 ГБ; есть BF16/FP8/NVFP4.
Ограничение: офлайн-модель, не потоковый ASR; оптимизации NIM пока не в открытом релизе, vLLM-интеграция примерно в 2 раза медленнее.
Примеры
Discuss Telegram 0 109

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 63 GB · ready files70 GB
  • FP8 / INT8recommendedweights 31 GB35 GB
  • 4-bitweights 18 GB20 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.

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.

    # jingjingji007/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
    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 jingjingji007/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B --local-dir ~/models/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
    # 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 NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

jingjingji007/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
Author
jingjingji007
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

1090

Ask AI about this model

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

Comments