vh
051cb1549a
Build + container + /health all came up clean on the re-run; only the
voices-endpoint verify failed. The check greped the response body for
"voices"/"voice"/alloy/Carter — but VibeVoice's actual response shape
is OpenAI list-format `{"object":"list","data":[...]}`, which contains
none of those substrings. On a fresh install the data array is also
empty (voices live at /worktank/vibevoice/voices/ and the user seeds
them).
Switched the check to parse the JSON and assert the shape (object="list",
data is a list). Robust against empty voices, robust against future
schema additions.
106 lines
4.2 KiB
YAML
106 lines
4.2 KiB
YAML
# Deploy VibeVoice 1.5B (long-form) to irv-ml1.
|
|
#
|
|
# Builds the image locally from groxaxo/VibeVoice-FastAPI1 via docker
|
|
# buildx git URL context. ~12 min cold build (CUDA 12.8 + torch 2.8 +
|
|
# flash-attn). First start downloads VibeVoice-1.5B (~7 GB) into the
|
|
# bind-mounted HF cache. Generous /healthz wait deadline accommodates
|
|
# both.
|
|
#
|
|
# Usage:
|
|
# scripts/elway irv-ml1 --playbook playbooks/deploy-vibevoice.yaml
|
|
#
|
|
# Idempotent — every step is creates-/when-gated; rerun is safe.
|
|
|
|
vars:
|
|
compose_dir: /opt/docker/compose/vibevoice
|
|
voices_dir: /worktank/vibevoice/voices
|
|
cache_dir: /worktank/vibevoice/cache
|
|
host_port: "8194"
|
|
|
|
steps:
|
|
# ── host-side dirs ──────────────────────────────────────────────────
|
|
|
|
- name: Ensure /worktank/vibevoice root exists (one-time, sudo)
|
|
shell: mkdir -p /worktank/vibevoice
|
|
sudo: true
|
|
creates: /worktank/vibevoice
|
|
|
|
- name: Chown /worktank/vibevoice to lkraven
|
|
shell: chown lkraven:lkraven /worktank/vibevoice
|
|
sudo: true
|
|
when: '[ "$(stat -c %U /worktank/vibevoice)" != lkraven ]'
|
|
|
|
- name: Ensure voices dir exists
|
|
shell: mkdir -p {{ voices_dir }}
|
|
creates: "{{ voices_dir }}"
|
|
|
|
- name: Ensure cache dir exists
|
|
shell: mkdir -p {{ cache_dir }}
|
|
creates: "{{ cache_dir }}"
|
|
|
|
- name: Ensure compose dir exists
|
|
shell: mkdir -p {{ compose_dir }}
|
|
creates: "{{ compose_dir }}"
|
|
|
|
# ── deploy compose files ────────────────────────────────────────────
|
|
|
|
- name: Upload compose.yaml
|
|
upload:
|
|
src: stacks/vibevoice/compose.yaml
|
|
dest: "{{ compose_dir }}/compose.yaml"
|
|
mode: "0644"
|
|
|
|
- name: Seed .env from template (only if absent)
|
|
upload:
|
|
src: stacks/vibevoice/.env.example
|
|
dest: "{{ compose_dir }}/.env"
|
|
mode: "0644"
|
|
when: "[ ! -f {{ compose_dir }}/.env ]"
|
|
|
|
# ── build + bring up ────────────────────────────────────────────────
|
|
|
|
- name: docker compose build (~12 min first time; cached after)
|
|
# --progress=plain stops the carriage-return TUI from littering the log
|
|
# with garbled redraws. The grep drops pip's Downloading / Collecting /
|
|
# Requirement / progress-bar lines (~25% of the log) while keeping
|
|
# buildkit step transitions, DONE/CACHED/ERROR markers, and any apt /
|
|
# build-stage messages worth seeing. set -o pipefail so a build
|
|
# failure isn't swallowed by the grep return code.
|
|
shell: |
|
|
set -o pipefail
|
|
cd {{ compose_dir }} && docker compose build --progress=plain 2>&1 \
|
|
| grep -vE '^#[0-9]+ [0-9.]+ (Downloading|Collecting|Requirement|Using cached|Installing collected|Successfully (installed|built)|Saved /|━|Resolved|Prepared|Built)'
|
|
|
|
- name: docker compose up -d
|
|
shell: cd {{ compose_dir }} && docker compose up -d
|
|
|
|
- name: Wait for /health to respond (allow ~20 min for model download + warmup)
|
|
shell: |
|
|
for i in $(seq 1 240); do
|
|
curl -sf -o /dev/null --max-time 3 http://localhost:{{ host_port }}/health && exit 0
|
|
sleep 5
|
|
done
|
|
exit 1
|
|
changed_when: "false"
|
|
|
|
verify:
|
|
- name: /health returns 200
|
|
shell: curl -sf -o /dev/null http://localhost:{{ host_port }}/health
|
|
changed_when: "false"
|
|
|
|
- name: /v1/audio/voices returns OpenAI list-shape JSON
|
|
# Response is `{"object":"list","data":[...]}` per OpenAI's
|
|
# /v1/audio/voices contract. `data` may be empty when no voice
|
|
# files are seeded yet at /worktank/vibevoice/voices/ — the
|
|
# wrapper is still healthy. Don't grep for sample voice names;
|
|
# the seed set is user-controlled and a fresh install legitimately
|
|
# ships with zero voices.
|
|
shell: |
|
|
curl -sf http://localhost:{{ host_port }}/v1/audio/voices \
|
|
| python3 -c "import json,sys; d=json.load(sys.stdin); assert d.get('object')=='list' and isinstance(d.get('data'), list)"
|
|
changed_when: "false"
|
|
|
|
- name: Container is running
|
|
shell: docker inspect vibevoice --format '{{.State.Status}}' | grep -q running
|
|
changed_when: "false"
|