Skip to content
Language models

ReDimNet2-ONNX

fast speaker verification without PyTorch

2 min readupdated

An unofficial ONNX build of all 20 checkpoints from the ReDimNet2 model family (by PalabraAI, Interspeech 2026) for running on CPU and GPU via ONNX Runtime with zero PyTorch dependencies. The model takes 16 kHz mono audio and outputs a 192-dimensional embedding (a numerical voice vector), with mel-spectrogram calculation baked straight into the graph. You can compare recordings using the cosine similarity of the resulting vectors. This repo is ideal for developers who need lightweight, fast voice recognition or verification in production.

Key features

  • Full set of weights: all 20 variants are available, ranging from the compact B0 (1.1M parameters) to the large B6 (12.5M), trained on VoxCeleb2, VoxBlink2, and CN-Celeb2.
  • CPU speedup: FP32 graphs run 1.8–2.6x faster than the original PyTorch implementation while delivering virtually identical outputs (relative difference ~1e-6).
  • GPU optimization: FP16 versions are 21–42% smaller and run up to 2x faster than FP32 on long recordings or batches.
  • Built-in preprocessing: the graph handles mel-spectrogram computation and volume normalization on its own, supporting dynamic audio lengths and batch sizes.

What you can do

  • Voice authentication: integrate speaker identity verification into a server-side service or bot without dragging in a heavy PyTorch environment.
  • Speaker search and clustering: extract embeddings from audio recordings and group speech segments by voice using cosine similarity.
  • Local CPU inference: use lightweight models like b0-vox2-lm or b3-vox2-lm to process short phrases quickly on a standard processor.

Example

import numpy as np, onnxruntime as ort, soundfile as sf, soxr
from huggingface_hub import hf_hub_download

path = hf_hub_download("lzrhahahahaha/ReDimNet2-ONNX", "fp32/b6-vox2-lm.onnx")
model = ort.InferenceSession(path, providers=["CPUExecutionProvider"])
audio, rate = sf.read("first.wav", dtype="float32", always_2d=True)
audio = soxr.resample(audio.mean(axis=1), rate, 16000)
emb = model.run(None, {"input_values": audio[None]})[0][0]  # вектор [192]
Read moreDetails
Checkpoints with the lm suffix (after large-margin fine-tuning) are recommended; the minimum clip length is 720 samples (45 ms), though accuracy is noticeably higher on segments lasting a few seconds.
In benchmarks (Intel Core i5-14400 / RTX 4070 SUPER), processing a 7–9 s clip with B6-LM takes 361.1 ms on CPU (compared to 762.5 ms in PyTorch) and 21.1 ms on GPU in FP16.
Limitations: when running in batches, all clips in a single batch must be the exact same length (there is no padding mask), and for cnc2 checkpoints on CPU, you should enable the session.set_denormal_as_zero=1 option.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 GB
  • 4-bitweights 0.0 GB0.6 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

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # lzrhahahahaha/ReDimNet2-ONNX — A30, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve lzrhahahahaha/ReDimNet2-ONNX --max-model-len 8192 --quantization fp8
    Add more LoRAs to ReDimNet2-ONNX
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of ReDimNet2-ONNX (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

lzrhahahahaha/ReDimNet2-ONNX
Author
lzrhahahahaha
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Language models

Emotion-S1

Emotion-S1