Placed on GPU 3 first because it was the empty card. That was the wrong read:
the seat is ~800 MiB, under 1% of a 96 GB card, so the question was never "where
does it fit" but "whose headroom is cheapest to spend".
vLLM sizes its KV cache as a fraction of TOTAL VRAM, not free VRAM. A resident
tenant on an otherwise-clean card therefore does not cost its own megabytes — it
costs the profiling margin of whatever full-size seat lands there later, and
flash-next needs 93 GiB of 96. A 96 GB card at 2 MiB can still take that; the
same card at 922 MiB is one where the next big seat needs its utilization
hand-trimmed, which this repo's flash-next history shows is both thin and silent
when it goes wrong.
Committed utilization per card is the number that governs, not free bytes:
GPU 0 0.40 + 0.48 = 0.88 ~13 GB spare <- moved here
GPU 1 0.52+0.24+0.10+0.055+0.03+0.03 = 0.975 ~4.3 GB
GPU 2 0.96 ~1.8 GB
GPU 3 - kept empty as reserve
GPU 3 is back to 2 MiB / 97,247 MiB free and is now documented as a deliberate
reserve rather than a spare.
Post-move n=5 on the same clip: 0.68 / 0.54 / 0.54 / 0.52 / 0.53 s, median 0.54 s
against 0.50 s on GPU 3. The spreads overlap at this sample size and no difference
is claimed; the GPU 3 figure was taken on an idle card and is now noted as a best
case, since the seat shares GPU 0 with the hot serving path. Silence control and
the gateway round-trip both re-verified after the move.
Also records both Parakeet endpoints (FV v3 on Blackwell, IRV v2 on a 3090) and
the four confounds that make them not an A/B pair, sent to tts-dev for the bench.
Retargets the existing sherpa-onnx stack from irv-ml1 to fv-ml1's utility card
and puts it behind the gateway. GPU 3 was the only card with room: 0/1/2 carry
the vLLM seats at 84-95.5 GB of 96.
Changes:
- compose: pin GPU via `device_ids: ["3"]` (the dead on-host stub used
`count: all`, which would have handed a 0.6B ASR seat all four cards);
join traefik-net; port 8300; homepage href to the live FV address.
- .env.example: default to the v3 int8 model (25 European languages, 464 MiB)
rather than English-only v2; models to /tank/parakeet/models.
- app.py: warm the recognizer at startup before uvicorn accepts traffic.
The warmup is not an optimisation. ONNX Runtime's CUDA EP compiles and autotunes
lazily on the FIRST DECODE, and on sm_120 that measured 45.7s cold (reproduced at
45.1s on a second container) against ~0.50s warm. A 45s first request is
indistinguishable from a hang and LiteLLM's default timeout abandons it long
before it returns. Decoding 1s of silence at load moves the cost inside the
healthcheck's 300s start_period; first real request after restart is now 0.65s.
Verification, because "provider=cuda" in the log is only an echo of the env var:
ORT falls back to CPU silently and still returns correct text, so the service
being up and the transcript being right establishes nothing. The discriminator is
a process on GPU 3 (922 MiB), confirmed. Controls both directions — a known TTS
sentence transcribes near-exactly (positive), 3s of digital silence returns
empty (null). Warm throughput 0.50s median on an 8.52s clip, n=5, spread
0.47-0.65s, single-stream, one clip: a smoke measurement with its harness
stated, not a benchmark.
Gateway aliases `ext-stt` (engine-neutral, mirrors ext-tts) and `whisper-1`
(OpenAI-compatible drop-in) registered via POST /model/new, i.e. LiteLLM's
Postgres store where the ext-tts family already lives — no gateway restart, and
config.yaml is consequently not a complete picture of what the gateway serves.
Both verified end to end.
The aliases use a raw IP deliberately: ana-docker resolves no .internal names at
all (resolv.conf points at 1.1.1.1), and LiteLLM only reaches irv-ml1 through a
hand-pinned extra_hosts entry. A second hosts entry would mean recreating the
container and bouncing the gateway for every consumer.
Also records the svos_miranda plugin validation pass and its structural findings,
and notes that the irv-ml1 parakeet is still running — there are two now, and
retiring the old one is the operator's call.
The 2026-09-06 headscale cutover retired irv-ml1's wg0 tunnel IP 10.100.79.3
(now 10.6.110.50). Repointed all LIVE canonical refs to the DNS NAME so the next
move can't re-break them: homepage.href/siteMonitor labels across 25 stack
composes, load-bearing env defaults (asset-engine INFERENCE_HOST, open-webui
AUDIO_TTS_OPENAI_API_BASE_URL, skaldsong SKALDSONG_TTS_BASE_URL, zonos-gateway
ZONOS_URL, dia), homepage services.yaml manual cards (Voice Design Studio,
IRV-ML1), and servers/irv-ml1/ssh-target. Updated the stale 'WG tunnel' comment
to the mesh reality.
Left as-is: README curl-examples and .env.example comments (docs), and historical
mentions in CLAUDE.md/persistent-memory. NOTE: applying the label repoints to the
RUNNING irv-ml1 containers needs a recreate per service (labels read at creation);
deployed .env values are separate from these canonical defaults.
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.
Operator consolidation (2026-06-18): give ComfyUI the full 48 GB A6000 and move the
audio/TTS zoo to the 3090.
- comfyui: NVIDIA_VISIBLE_DEVICES all -> 1 (A6000 only), + DISABLE_UPGRADES=true to
pin torch at 2.12.1+cu129 so the mmartial boot script stops auto-upgrading it and
the compiled SageAttention kernels stay matched (comfy-dev torch-pin, approved).
- parakeet: NVIDIA_VISIBLE_DEVICES all -> 0 (3090).
Other GPU reassignments are deployment-side (not repo compose): chatterbox-fast via
its .env CBF_GPU_DEVICES=0; vibevoice device_ids ["1"]->["0"] (deployed from
/worktank/vibevoice/build); yt-voice-clipper worker via its override. dia2-2b,
ace-step, csm-expressiva downed (stale/unused).
Result: A6000 = ComfyUI alone (48.3 GB free); 3090 = chatterbox + parakeet + the
on-demand audio (vibevoice/ytvc/kokoro). SageAttention rebuilt against the pinned
torch; OOM cmdline (COMFY_CMDLINE_EXTRA) preserved; /object_info still lists the 9
acceleration nodes.
- parakeet/compose.yaml: healthcheck was using curl which isn't in the
image (only wget is, via apt). 2,190 failing checks — switched to
`wget -q -O /dev/null`, container went healthy on recreate.
- qwen3-tts/.env.example: variant annotation was reversed. The upstream
wrapper's runtime error is unambiguous: voice cloning requires the
-Base variant, not -CustomVoice. Corrected the comment block and
flipped the default to Qwen/Qwen3-TTS-12Hz-1.7B-Base.
- qwen3-tts/README.md: 0.6B switch snippet now suffixes -Base too,
since plain `Qwen/Qwen3-TTS-12Hz-0.6B` isn't published on HF.
The Shadowfita FastAPI wrapper hit two unfixed upstream bugs on the
first real /transcribe call — chunker return-shape mismatch (open
issue #16) and a `torchaudio.tensor` that doesn't exist (open #10).
Rather than babysit someone else's half-tested code, switched to
sherpa-onnx with the prebuilt int8 Parakeet-TDT tarball from k2-fsa,
and wrote our own ~60-line FastAPI wrapper.
Moving parts now owned in-tree:
Dockerfile CUDA 12.8 + cuDNN 9 runtime base, installs
sherpa-onnx==1.12.39+cuda12.cudnn9 + fastapi +
soundfile + libasound2 (sherpa-onnx links to ALSA
at load time even when we never touch a mic).
app.py OfflineRecognizer.from_transducer() once at startup;
/transcribe and /v1/audio/transcriptions both accept
multipart uploads and return {"text": ...}.
entrypoint.sh Idempotent model download to /models on first run
(~400 MB int8 tarball), then exec uvicorn.
Smoke test: 0.wav (bundled in the tarball, The House of the Seven
Gables excerpt) transcribes cleanly in ~1.2s on GPU.
PARAKEET_MODEL_URL in .env lets you swap to the v3 (25-language)
tarball without touching any other files. Wipe *.onnx + tokens.txt
from the models dir and the entrypoint re-downloads.
Two new speech stacks on irv-ml1, both on the /worktank/<stack>/
pattern, no tnet (irv-ml1 is local-endpoints-only for now).
parakeet — ASR via Shadowfita/parakeet-tdt-0.6b-v2-fastapi:
- docker buildx git context pinned to SHA 31c5652; no source
vendored. Rebuild on SHA bump.
- GPU-capable FastAPI + Silero VAD + WS streaming.
- API: POST /transcribe, WS /ws/transcribe, GET /healthz. Not the
literal OpenAI `/v1/audio/transcriptions` path — note in README.
- HF cache at /worktank/parakeet/models/ (excluded from restic).
- Build ~158s first time; steady-state start ~40s.
cosyvoice — TTS via neosun/cosyvoice:v1.3.2 shipping
Fun-CosyVoice3-0.5B-2512 (CosyVoice 3, chosen over v2 for the
expanded 5,000-hour instruction-following data covering emotions,
speed, tones, dialects, accents, role-playing; ~150ms streaming
TTFB matches v2). API: /v1/audio/speech (OpenAI drop-in),
/v1/voices/create (cloning), /health.
- Host port 8190 (container 8188; host 8188 already taken by comfyui).
- /worktank/cosyvoice/{voices,input,output}/; voices include in
restic (precious — reproducing a clone needs the original ref
audio), input+output excluded (scratch).
- Model weights (~2-3 GB) live inside image layer; re-download on
tag bump, persist across `compose up -d`.
Both healthy on first deploy.
vh