mAi
c81c145163
Go daemon listening on :8770 that fronts mvoice (8766), whisper-server
(8178), ollama (11434), comfyui (8188) behind a single /v1 façade.
What this MVP does:
- Loads config/consumers.yaml: routing table, per-consumer URL + health +
paths + vram_resident_mib + can_coexist_with + load/unload routes.
- Background health probe (5s) on every consumer; refuses fast with a
structured 503 if the last probe failed (no Felix-Banholzer-style
silent fallback).
- POST /v1/{tts,stt,llm,image} proxies the request body + Content-Type
to the routed consumer's path and streams the response back.
- GET /audio/* proxies to audio_proxy consumer (wa.sh fetches its WAV
this way).
- GET /v1/status exposes live GPU sample (nvidia-smi every 2s),
per-consumer health/loaded/gpu_resident_mib/active/total_requests,
scheduler stats.
- GET /healthz, GET / — broker liveness.
The Scheduler interface is in place but the implementation is
'Passthrough' — every job runs immediately, no lock, no queue. Schritt 4
replaces it with a serialising mutex; Schritt 5 adds VRAM-pressure
eviction. The interface boundary means server.go stays unchanged.
Out of scope here:
- Schritt 3: wa.sh migration (parallel work in mAi).
- Schritt 4: queue + global GPU lock.
- Schritt 5: nvidia-smi-driven LRU eviction.
Tests: config validation (good/bad), proxy forwards body, audio proxy
streams bytes, unhealthy consumer returns 503, /v1/status JSON shape.
Refs: m/mGPUmanager#1
89 lines
1.9 KiB
YAML
89 lines
1.9 KiB
YAML
listen: 127.0.0.1:8770
|
|
|
|
gpu:
|
|
total_mib: 16376 # RTX 4070 Ti SUPER
|
|
reserved_mib: 1024 # headroom for system/desktop
|
|
poll_interval_seconds: 2
|
|
|
|
routing:
|
|
tts: mvoice
|
|
stt: mvoice # whisper-server is alternative if explicitly requested
|
|
llm: ollama
|
|
image: comfyui
|
|
|
|
# Audio download proxy: GET /audio/* forwards to this consumer.
|
|
audio_proxy: mvoice
|
|
|
|
consumers:
|
|
mvoice:
|
|
url: http://localhost:8766
|
|
health:
|
|
method: GET
|
|
path: /api/health
|
|
paths:
|
|
tts:
|
|
method: POST
|
|
path: /api/synthesize
|
|
stt:
|
|
method: POST
|
|
path: /api/transcribe
|
|
vram_resident_mib: 2800
|
|
load:
|
|
method: POST
|
|
path: /api/admin/load
|
|
unload:
|
|
method: POST
|
|
path: /api/admin/unload
|
|
can_coexist_with: [whisper-server, ollama]
|
|
priority: 3
|
|
max_concurrency: 1
|
|
|
|
whisper-server:
|
|
url: http://localhost:8178
|
|
health:
|
|
method: GET
|
|
path: /
|
|
paths:
|
|
stt:
|
|
method: POST
|
|
path: /inference
|
|
vram_resident_mib: 2050
|
|
# No HTTP unload; mGPUmanager evicts via systemd restart (Schritt 5).
|
|
systemd_unit: whisper-server.service
|
|
can_coexist_with: [mvoice, ollama]
|
|
priority: 2
|
|
max_concurrency: 1
|
|
|
|
ollama:
|
|
url: http://localhost:11434
|
|
health:
|
|
method: GET
|
|
path: /api/tags
|
|
paths:
|
|
llm:
|
|
method: POST
|
|
path: /api/generate
|
|
# Ollama runs its own LRU keep_alive; we don't track resident VRAM.
|
|
vram_managed: true
|
|
can_coexist_with: [mvoice, whisper-server]
|
|
priority: 2
|
|
max_concurrency: 1
|
|
|
|
comfyui:
|
|
url: http://localhost:8188
|
|
health:
|
|
method: GET
|
|
path: /system_stats
|
|
paths:
|
|
image:
|
|
method: POST
|
|
path: /prompt
|
|
vram_resident_mib: 13000
|
|
unload:
|
|
method: POST
|
|
path: /api/free
|
|
body: '{"unload_models":true,"free_memory":true}'
|
|
can_coexist_with: []
|
|
priority: 1
|
|
max_concurrency: 1
|