GGUF weights for Zyphra/ZONOS2, ready to run
with zonos2.cpp — a standalone ggml/GGUF C++
port of ZONOS2. The entire pipeline (ECAPA speaker encoder → MoE backbone → DAC
vocoder) runs as native C++ linking only libggml + gguf: no Python, no PyTorch, no
CUDA-only kernels at inference time. CPU and CUDA use the same files.
ZONOS2 is Zyphra's latest text-to-speech model, trained on more than 6 million hours of
varied multilingual speech, delivering expressiveness and quality on par with—or even
surpassing—top TTS providers at low latency with MoE. ZONOS2 excels at high-fidelity and
naturalistic voice cloning.
ECAPA-TDNN speaker encoder (wav → x-vector, for voice cloning)
All quants keep the spine (attention + router + embeddings + speaker projection) at
F16 and quantize only the MoE expert matrices — the layout that holds quality far below
the usual quant floor. The full-precision spine is the single biggest quality lever; pick
the expert precision (Q8 → Q4) that fits your VRAM.
Benchmarks
Build
Size
bpw
KLD ↓
Top-1 ↑
WER ↓
SpkSim ↑
UTMOS ↑
F16 (ref)
15.3 GB
16.0
—
—
2.79
66.75
4.40
Q8_0
8.5 GB
8.50
0.002
96.5%
2.87
66.30
4.40
Q6_K
6.8 GB
6.56
0.007
92.9%
3.07
66.12
4.40
Q5_K
5.8 GB
5.50
0.025
86.3%
2.98
66.30
4.40
Q4_K
4.9 GB
4.50
0.072
76.9%
3.00
64.54
4.36
KLD (mean) and Top-1 measure how closely each quant tracks the F16 backbone's
per-frame logits, scored with the zonos2-perplexity tool in zonos2.cpp.
WER (Qwen3-ASR word-error rate), SpkSim (clone speaker similarity), and
UTMOS (predicted MOS) are end-to-end on the
Zyphra/ZTT1-Eval Clean English set;
lower WER and higher SpkSim/UTMOS are better.
Although the logit metrics (KLD, Top-1) degrade steadily as the experts shrink, the
audio quality holds nearly flat down to Q4_K — WER, speaker similarity, and UTMOS stay
within eval noise of F16. The F16 spine keeps the model on-distribution, so the smaller
quants spend their error budget on inaudible logit jitter rather than audible artifacts.
Q8_0 is the recommended default (effectively lossless); Q4_K is a strong choice when
VRAM is tight.
Quick Start
Build zonos2.cpp (CPU, or -DGGML_CUDA=ON for
NVIDIA), then turn text into a waveform with one command:
Add --spk voice.mp3 (with spk-encoder.gguf) to clone a voice from a reference clip.
HTTP server
zonos2-server mirrors the reference FastAPI — low-latency streaming PCM, an OpenAI
/v1/audio/speech endpoint, in-process voice cloning, and a browser UI:
