Skip to content
Audio

omniASR-CTC-3B

a large multilingual speech recognition model

2 min readupdated

This model transcribes spoken audio into text across multiple languages. It is based on the omniASR_CTC_3B checkpoint from Facebook Research's OmniLingual project, converted from the fairseq2 framework into the transformers library format using the Wav2Vec2ForCTC architecture (a speech recognition model that handles variable-length audio). Unlike the original release, this port lets you skip setting up fairseq2 and run inference with standard Hugging Face tools. It is a solid fit for developers building voice services, transcription tools, and multilingual audio processing pipelines.

Key features

  • Packs 3.08 billion parameters across 60 encoder layers, with a hidden size of 2048 and 32 attention heads.
  • Supports multilingual transcription and outputs CTC logits over a 9,812-token SentencePiece vocabulary.
  • Native transformers compatibility: model loading and decoding work out of the box via standard AutoProcessor and Wav2Vec2ForCTC classes.
  • The author notes that full output parity verification against the original fairseq2 checkpoint has not been performed.

What you can do

  • Integrate it into transcription pipelines for multilingual podcasts, interviews, and voice notes.
  • Use it as a baseline ASR backend in transformers-based projects without tying your stack to the fairseq2 ecosystem.
  • Benchmark its recognition accuracy on your target languages and compare the results against the original OmniLingual build.

Example

from transformers import Wav2Vec2ForCTC, AutoProcessor
import torch, torchaudio

processor = AutoProcessor.from_pretrained("nolimitsxl/omniASR-CTC-3B")
model = Wav2Vec2ForCTC.from_pretrained("nolimitsxl/omniASR-CTC-3B").eval()
waveform, sr = torchaudio.load("audio.wav")
inputs = processor(waveform.squeeze().numpy(), sampling_rate=16_000, return_tensors="pt")
print(processor.decode(torch.argmax(model(**inputs).logits, dim=-1)[0]))
Read moreDetails
Architecture: Wav2Vec2ForCTC, 60 layers, FFN intermediate 8192.
Input audio sample rate: 16,000 Hz.
Source: port of the omniASR_CTC_3B checkpoint from facebookresearch/omnilingual-asr.
Limitation: parity verification was not run during conversion.
Discuss Telegram 0 25

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 11 GB · ready files13 GB
  • BF16weights 5.7 GB6.8 GB
  • FP8 / INT8recommendedweights 2.9 GB3.7 GB
  • 4-bitweights 1.6 GB2.3 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 5080

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.

    # nolimitsxl/omniASR-CTC-3B
    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 nolimitsxl/omniASR-CTC-3B --local-dir ~/models/omniASR-CTC-3B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to omniASR-CTC-3B
  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 omniASR-CTC-3B (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

nolimitsxl/omniASR-CTC-3B
Author
nolimitsxl
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

250

Ask AI about this model

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

Comments

Audio

Prakriti v1

Prakriti v1