stacks: add dia + zonos to the TTS bench

dia: Nari Labs dialogue TTS (Dia 1.6B / Dia2-1B / Dia2-2B) via
devnen/Dia-TTS-Server — OpenAI-compat, fills the multi-speaker
dialogue-scene slot for skaldsong. Port 8200 on irv-ml1.

zonos: Zyphra Zonos-v0.1 (Apache-2.0, 44kHz, emotion sliders) via the
official Gradio interface. Audition surface only — no OpenAI-compat
endpoint yet (needs the FastAPI fork to become skaldsong-pluggable).
Port 8199 on irv-ml1.

Both follow the chatterbox/fish-s2 convention: local image built from a
pinned wrapper SHA via buildx git-context, .env-driven port/GPU, python
healthcheck, homepage labels.
This commit is contained in:
vh
2026-05-31 10:53:27 -07:00
parent 53157b193d
commit 666f7f31e0
6 changed files with 356 additions and 0 deletions
+44
View File
@@ -0,0 +1,44 @@
# Zonos-v0.1 stack tunables. Copy to `.env` on irv-ml1 before deploying.
# ── build pin ────────────────────────────────────────────────────────
# SHA of Zyphra/Zonos. Use the FULL 40-char SHA — docker buildx's git
# source resolver doesn't accept short hashes. `main` works but is NOT
# reproducible; pin a SHA before you rely on this. Latest commits:
# https://github.com/Zyphra/Zonos/commits/main
ZONOS_SHA=main
# Local image tag — bump when you change build context to force a
# fresh layer build.
ZONOS_TAG=v1
# ── network ──────────────────────────────────────────────────────────
# Host port. Container's Gradio listens on 7860 internally.
# Reserved on irv-ml1: 8188 ComfyUI, 8190 CosyVoice, 8191 Qwen3-TTS,
# 8192 IndexTTS-2, 8193 Kokoro, 8194 VibeVoice, 8195 Fish-S2,
# 8196 Chatterbox, 8197 Voxtral, 8198 Kyutai, 8200 Dia, 8765 Parakeet.
# 8199 picked here.
ZONOS_PORT=8199
# Bind address. 0.0.0.0 exposes on all interfaces (incl. WG tunnel
# interface 10.100.79.3); 127.0.0.1 restricts to local-only.
ZONOS_BIND=0.0.0.0
# ── runtime / GPU ────────────────────────────────────────────────────
# Devices visible inside the container. "0" pins to the RTX 3090
# (24 GB) — plenty for Zonos (~6 GB). Both the 3090 and the A6000 are
# Ampere, so either satisfies the hybrid variant's "3000-series or
# newer" requirement; "1" selects the A6000.
ZONOS_GPU_DEVICES=0
# Model variant is chosen in the Gradio UI / SDK, not via env:
# Zyphra/Zonos-v0.1-transformer — default, ~3.6 GB, no special kernels
# Zyphra/Zonos-v0.1-hybrid — Mamba-SSM hybrid; needs Ampere+ and
# extra build deps (mamba-ssm,
# causal-conv1d) baked by the image
# Start with the transformer variant unless you specifically want hybrid.
# ── persistent storage on the host ───────────────────────────────────
# HuggingFace cache — Zonos weights land here on first start.
# Bind-mounted so they survive container recreate. Excluded from restic
# (regenerable from HF).
ZONOS_CACHE_DIR=/worktank/zonos/cache
+65
View File
@@ -0,0 +1,65 @@
# Zonos
Zyphra's expressive multilingual open-weight TTS
([Zyphra/Zonos](https://github.com/Zyphra/Zonos)) — 44 kHz output,
zero-shot voice cloning, and explicit emotion/conditioning controls —
served via the **official repo's Gradio interface**.
Models ([`Zonos.from_pretrained()`](https://github.com/Zyphra/Zonos)):
| Variant | HF repo | Notes |
|---|---|---|
| transformer | [Zyphra/Zonos-v0.1-transformer](https://huggingface.co/Zyphra/Zonos-v0.1-transformer) | ~3.6 GB, no special kernels (default) |
| hybrid | [Zyphra/Zonos-v0.1-hybrid](https://huggingface.co/Zyphra/Zonos-v0.1-hybrid) | Mamba-SSM; needs Ampere+ GPU + extra build deps |
**Server:** irv-ml1 (Irvine, WireGuard-only)
**Port:** 8199 (container Gradio listens on 7860)
**GPUs:** pins to device 0 (RTX 3090) by default; ~6 GB VRAM
**Image:** `local/zonos:v1` — built locally from a pinned git SHA of the
upstream repo via docker buildx's git URL context
**Upstream:** [Zyphra/Zonos](https://github.com/Zyphra/Zonos) (Apache-2.0),
trained on 200k+ hours; multilingual EN/JA/ZH/FR/DE
## Why this stack exists
A different *control surface* from the rest of the bench: Zonos exposes
**emotion/conditioning sliders** plus pitch/rate conditioning and 44 kHz
output, where Chatterbox/Dia use inline tags and Fish uses
natural-language paralinguistic tags. Worth A/B-ing by ear against
Chatterbox-Turbo (the research that prompted this stack explicitly said
"benchmark Zonos against Chatterbox before choosing").
## ⚠️ Audition surface, not skaldsong-pluggable (yet)
The official repo ships a **Gradio WebUI + Python SDK only** — there is
**no OpenAI-compatible `/v1/audio/speech` endpoint**. So this stack is
for *auditioning quality*, not for wiring into skaldsong's engine
router as-is. To promote Zonos to a real engine slot we'd need either:
- the community FastAPI fork ([Zyphra/Zonos PR #73](https://github.com/Zyphra/Zonos/pull/73), adds REST + basic streaming), or
- a thin OpenAI-compat adapter in front of the Python SDK.
Both are a follow-up if Zonos earns a slot in the ear test.
## Deploy
```bash
# from this workstation (irv-ml1 is WG-only — routes via ana-wg):
scripts/deploy-stack.sh irv-ml1 zonos
# then on irv-ml1, first run builds the image from the pinned SHA:
# docker compose up -d --build
```
Open `http://10.100.79.3:8199/` for the Gradio UI. First boot pulls the
model (~3.6 GB) into `ZONOS_CACHE_DIR`; the 600 s `start_period` covers
it.
## Notes
- **Pin `ZONOS_SHA`** to a full 40-char commit before relying on this —
`.env.example` ships `main`, which is not reproducible.
- The upstream image bundles **espeak-ng** (required for Zonos's
phonemization) via its Dockerfile.
- `GRADIO_SERVER_NAME=0.0.0.0` is set so the in-container Gradio binds
all interfaces and the host port-map reaches it. If a future upstream
hardcodes `server_name`, the API fork above is the cleaner path.
+56
View File
@@ -0,0 +1,56 @@
# Zonos-v0.1 (Zyphra's expressive multilingual open-weight TTS) served
# via the OFFICIAL Zyphra/Zonos repo's Gradio interface.
#
# Why this stack exists alongside the other TTS:
# * Apache-2.0, 44 kHz output, zero-shot voice cloning from a
# 10-30 s sample, and explicit emotion/conditioning sliders —
# a different control surface from tag-based engines.
# * Multilingual (EN/JA/ZH/FR/DE), trained on 200k+ hours.
# * Cheap to run (~6 GB VRAM, transformer variant ~3.6 GB weights).
#
# CAVEAT — this is an AUDITION/EVAL surface, not a skaldsong-pluggable
# engine yet: the official repo ships a Gradio WebUI + Python SDK, NOT
# an OpenAI-compatible /v1/audio/speech endpoint. To wire Zonos into
# skaldsong's router we'd need the community FastAPI fork (Zyphra/Zonos
# PR #73) or a thin adapter. Bench it by ear first; promote later if it
# earns a slot.
#
# Image is built locally from the upstream repo's Dockerfile via docker
# buildx git-context. All tunables live in .env — edit that, not this.
services:
zonos:
image: local/zonos:${ZONOS_TAG}
build:
# Single Dockerfile at repo root; bundles espeak-ng (required for
# Zonos's eSpeak phonemization).
context: https://github.com/Zyphra/Zonos.git#${ZONOS_SHA}
dockerfile: Dockerfile
container_name: zonos
restart: unless-stopped
runtime: nvidia
ports:
- "${ZONOS_BIND:-0.0.0.0}:${ZONOS_PORT}:7860"
environment:
- NVIDIA_VISIBLE_DEVICES=${ZONOS_GPU_DEVICES:-0}
- NVIDIA_DRIVER_CAPABILITIES=compute,utility
# Make Gradio bind all interfaces inside the container so the host
# port-map reaches it (Gradio otherwise may bind 127.0.0.1 only).
- GRADIO_SERVER_NAME=0.0.0.0
- HF_HOME=/app/hf_cache
volumes:
- ${ZONOS_CACHE_DIR}:/app/hf_cache
healthcheck:
# No JSON status route on the Gradio app — just probe the root
# page for liveness. start_period covers the first-boot model pull.
test: ["CMD-SHELL", "python3 -c \"import urllib.request,sys; urllib.request.urlopen('http://127.0.0.1:7860/', timeout=5); sys.exit(0)\""]
interval: 30s
timeout: 10s
retries: 3
start_period: 600s
labels:
- homepage.group=AI Systems
- homepage.name=Zonos
- homepage.icon=mdi-waveform
- homepage.description=Expressive multilingual TTS + cloning, 44kHz (Gradio eval, irv-ml1)
- homepage.href=http://10.100.79.3:${ZONOS_PORT}