Zyphra/ZONOS3-Mini
AI model · https://huggingface.co/Zyphra/ZONOS3-Mini
First seen by Oriel on .
AI summary
ZONOS3-Mini is an Apache 2.0 text-to-speech model offering multilingual speech, voice cloning, fine-grained delivery controls, and real-time local inference with about 100 million active parameters.
- The model supports 43 languages, generates up to 10 minutes of 44.1 kHz audio, and accepts reference clips up to 60 seconds.
- Its 800M-parameter latent MoE backbone activates about 100M parameters per token, using 64 experts with top-2 routing across 11 layers.
- Deployment options include WebGPU browsers, SGLang-Omni on NVIDIA GPUs, MLX on Apple Silicon, and C++/ggml runtimes for CPU, CUDA, Metal, or Vulkan.
- SGLang-Omni exposes an OpenAI-compatible /v1/audio/speech endpoint supporting streaming and transcript-free voice cloning from reference audio.
- On one H100 80GB, evaluation reports 80 ms median time to first audio at eight concurrent requests and peak throughput of 457 audio-seconds per second.
#text-to-speech #voice cloning #mixture of experts #webgpu
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 496, latest observation ; +404 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 69,946, latest observation ; +69,090 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | Zyphra/ZONOS3-Mini | https://huggingface.co/Zyphra/ZONOS3-Mini Read discussion |