Skip to content
Audio

Nemotron ASR Streaming Farsi Podcast

стриминговая расшифровка разговорного персидского

2 min readupdated

Модель от MohammadGholizadeh для ASR на персидском (fa). Это дообучение mehdi-hf/nemotron-asr-streaming-farsi на 365,3 часах подкастной речи; архитектура та же — cache-aware FastConformer + RNN-T на 600M параметров, токенизатор на 1 024 элемента и промпт fa-IR.

Ключевые особенности

  • Заточена под подкасты и разговорную речь: WER 12,3% против 16,4% у базовой модели при look-ahead 1,12 с.
  • Работает потоково чанками или с целыми файлами; доступны 1,12 с и 0,32 с look-ahead.
  • Пишет в разговорном стиле (رو, هستن, می‌خوایم), без пунктуации, числа словами, без латиницы.
  • Интерфейс как у базовой модели: NeMo checkpoint .nemo, промпт fa-IR обязателен.
  • Для формальной/читаемой речи хуже базовой: FLEURS 12,91% против 8,82% WER.

Пример

from huggingface_hub import hf_hub_download
from nemo.collections.asr.models import ASRModel

path = hf_hub_download("MohammadGholizadeh/nemotron-asr-streaming-farsi-podcast",
                       "nemotron-asr-streaming-farsi-podcast.nemo")
model = ASRModel.restore_from(path).eval()
model.encoder.set_default_att_context_size([56, 13])  # 1.12 s; [56, 3] для 0.32 s
model.set_inference_prompt("fa-IR")
ПодробнееПотоковый WER на подкастах: 12,31% при 1,12 с look-ahead и 13,54% при 0,32 с; у базовой — 16,44% и 18,08%.
Обучение: 365,3 ч, 5 000 шагов, LR 3e-5, 1× NVIDIA GH200 (96 GB), 2 ч 46 мин. В репозитории только NeMo checkpoint, без весов Transformers.
Ограничения: тест — 5 эпизодов, возможны пропуски окончаний и циклы повторения на длинных формальных фразах; на новых ведущих выигрыш может быть меньше.
Discuss Telegram 1 35

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 2.3 GB · ready files3.1 GB
  • BF16weights 1.2 GB1.8 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 · F32 — 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.

    # MohammadGholizadeh/nemotron-asr-streaming-farsi-podcast
    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 MohammadGholizadeh/nemotron-asr-streaming-farsi-podcast --local-dir ~/models/nemotron-asr-streaming-farsi-podcast
    # 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

  1. nemotron-3.5-asr-streaming-0.6bfamily root
  2. nemotron-asr-streaming-farsi
  3. nemotron-asr-streaming-farsi-podcast

New posts about versions of nemotron-3.5-asr-streaming-0.6b (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 1All on Hugging Face

About the model

MohammadGholizadeh/nemotron-asr-streaming-farsi-podcast
Author
MohammadGholizadeh
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

351

Ask AI about this model

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

Comments