Move the ~22-service flat "AI Systems" group off the Main tab into a new
four-tab layout (Main / AI / Infrastructure / Toolchain). The AI tab sorts
the inference fleet by function into seven groups:
AI - Inference gen, char-rp, char-rp-reasoning, Granite summarizer
AI - Eval & Retrieval Selene, Skywork Reward, Qwen3 rerank/embed, image-bench
AI - Gateways & Chat LiteLLM, Asset Engine, Gateway Chat, Open WebUI, ...
AI - Speech (TTS) Chatterbox Fast, Kokoro, mOrpheus
AI - Audio Tools Parakeet ASR, YT Voice Clipper
AI - Image & Media ComfyUI, Arbo
AI - Dormant stopped rollback seats + retired auditions
Relabel each stack's homepage.group so canonical stacks/ matches the live
containers on ana-ml2, ana-docker, and irv-ml1. Dormant stacks were refreshed
with `docker compose up --no-start` so they carry the new label while staying
stopped (compose-start rollback preserved). settings.yaml drives tab/order/
columns; services.yaml and README updated to the new scheme.
Voxtral final fix (8th iteration):
* The bundled voxtral_tts.yaml hardcodes gpu_memory_utilization: 0.8
on the language_model stage — overrides the CLI flag. Mounted a
patched copy (0.4) at /etc/voxtral/voxtral_tts.yaml and pointed
--stage-configs-path there.
* With Kyutai stopped to free 5 GB on the 3090, both stages fit
(target 9.4 + 2.4 GB ≈ 11.8 GB; 17 GB free post-kyutai-stop).
* Voxtral now healthy on GPU 0 — bench: 1.9-2.7 s TTFB, real WAV.
Fish s2-pro optimization (per-request sweep, no model swap):
* `streaming: true` in request body drops TTFB from 7.7 s → 0.026 s
(300×). Total time goes up ~1 s (chunked HTTP overhead) but
perceived latency = TTFB. Use stream:true for any interactive use.
* `latency: "balanced"` actually slower than default — bad name; skip.
* `use_memory_cache: "on"` no measurable benefit.
* `chunk_length: 100` (default 200) no TTFB benefit non-streaming.
* Server-side `--half` (fp16 inference) added via compose `command`
override — passes through start_server.sh's $@ unchanged into
api_server.py. Should reduce total time too. Validation pending
the post-restart bench.
Kyutai stopped to free GPU 0 budget — the bench numbers earlier
(3.4 s avg) were unimpressive vs Voxtral's 2.3 s in the same
multilingual slot. Kept the stack files for future re-deploy if
needed; just the running container is gone.
Three fixes from the second-wave deploy attempts:
* voxtral: vllm/vllm-omni doesn't publish a `latest` tag — pull
failed with "manifest unknown". Pinned VOXTRAL_VLLM_TAG to v0.18.0
(released 2026-03-29, the day after the Voxtral 4B TTS release —
first cut with Voxtral support).
* kyutai-tts: NillPointer wrapper exposes ONLY /health (root) and
POST /v1/audio/speech. No /v1/models, no /v1/audio/voices —
those return 404. Verified by /openapi.json against the live
container. Compose healthcheck + playbook wait + verify steps
all repointed at the actual paths. POST /v1/audio/speech is now
smoke-tested with a RIFF WAV assertion (same pattern as fish-s2).
* fish-s2: added FISH_S2_MODEL env var so the model variant is
swappable via .env without rebuilding. Both s2-pro (default) and
s1-mini are pre-pulled into the bind-mount; LLAMA_CHECKPOINT_PATH
+ DECODER_CHECKPOINT_PATH now use ${FISH_S2_MODEL:-s2-pro}.
s1-mini was originally gated on fishaudio's HF org (401), but
niobures/OpenAudio-S1 mirrors the same files openly — pulled
from there via a one-shot snapshot_download.
Fourth fish-s2 attempt got past build + checkpoints, then container
crashlooped silently again. Diagnosis: the upstream docker/Dockerfile
is multi-stage with `webui` and `server` targets; without specifying
a target, docker builds the LAST stage (webui — gradio-only, no
start_server.sh, no API server). start_server.sh is the entrypoint
script that lives only in the `server` stage.
Confirmed by `cat /app/start_server.sh` inside the built image:
"No such file or directory."
Upstream's compose.yml uses target: server on its server service —
doing the same here.
Second deploy attempt failed at build time:
failed to fetch anonymous token: ... ghcr.io/fishaudio/fish-speech ... 403 Forbidden
Root cause: dockerfile.dev is a thin two-line wrapper around
`FROM ghcr.io/fishaudio/fish-speech:${VERSION}`, which is a private
GHCR base image. Anonymous pulls 403, and we'd need GHCR auth to use
that path. The dev variant is meant for upstream's CI / fish-speech
contributors, not external consumers.
The REAL production path (from upstream's compose.base.yml) is to
build from `docker/Dockerfile` with build args BACKEND=cuda,
CUDA_VER=12.9.0, UV_EXTRA=cu129, UV_VERSION=0.8.15. That builds
everything from source — slower (15-20 min cold), but fully self-
contained.
irv-ml1's driver (595.58.03, CUDA 13.2 capable) is forward-compatible
with the 12.9 PyTorch wheels.
Took three iterations to find the right Dockerfile because:
1. First try: dockerfile (lowercase) — doesn't exist
2. Second try: dockerfile.dev — exists but pulls a private base
3. Third try: docker/Dockerfile — actual production path
First fish-s2 deploy attempt failed in step 9/11:
failed to read dockerfile: open dockerfile: no such file or directory
Upstream fishaudio/fish-speech ships:
* dockerfile.dev (lowercase, dev/test image)
* compose.yml + compose.base.yml (intended deploy path:
`docker compose --profile server up`)
There is no standalone production Dockerfile. The dockerfile.dev
image is what their own compose.yml builds from anyway, so building
against it directly is functionally equivalent to using their compose
profile — we just keep our own restart-policy / labels / bind-mount
conventions on the outer compose.
Comment in the build block now documents this so future-Claude doesn't
re-walk the path.
Adds the three premier 2026 TTS releases we missed during the original
fleet build-out (early April), all licensed for self-host:
* Fish Audio S2-Pro (port 8195, GPU 1 / A6000) — released 2026-03-09.
4B dual-AR (Slow + Fast) trained on 10M+ hours / 80+ languages.
Headline: 15,000+ paralinguistic / emotion tags via natural language
([laugh] [whispers] [super happy] etc.) — a step-function over
Chatterbox Turbo's 9 fixed tags. 91.61% paralinguistic win rate on
EmergentTTS-Eval. ~150 ms streaming TTFB, voice cloning, MIT-style
open. ~17 GB VRAM.
* Voxtral TTS (port 8197, GPU 1 / A6000) — Mistral, released 2026-03-28.
4B open-weight, 70 ms model latency, 9.7× realtime. 68.4% blind A/B
win rate vs ElevenLabs Flash v2.5 in cloning. 8 languages
(EN/FR/DE/ES/IT/PT/NL/HI). Served via vLLM-Omni (Mistral's partner
serving stack) — published Docker image, no local build. ~16 GB VRAM.
CC BY-NC license — personal/research use only; flagged in README.
* Kyutai TTS (port 8198, GPU 0 / 3090) — kyutai/tts-1.6b-en_fr.
Trained on 2.5M hours from the Moshi/Mimi team. Claimed 220 ms in
solo setup, 32 simultaneous streams under 350 ms on L40. Kyutai's
official deploy is Rust + websockets only; using NillPointer's
community OpenAI-compat wrapper to bridge to /v1/audio/speech so
it slots into the same bench harness. ~4-6 GB VRAM.
Each stack: compose.yaml (build context, env, volumes, healthcheck,
homepage label), .env.example (all tunables documented), README.md
(why it exists, headline numbers, API, deploy + hardware notes).
Playbooks at playbooks/deploy-{fish-s2,voxtral,kyutai-tts}.yaml are
idempotent in the same shape as the existing deploy-vibevoice /
deploy-chatterbox playbooks.
Port allocations on irv-ml1 after this lands: 8188 ComfyUI, 8190
CosyVoice, 8191 Qwen3-TTS, 8192 IndexTTS-2, 8193 Kokoro, 8194
VibeVoice, 8195 Fish, 8196 Chatterbox, 8197 Voxtral, 8198 Kyutai,
8765 Parakeet ASR.
vh