942 lines
26 KiB
YAML
942 lines
26 KiB
YAML
name: mike-ai
|
|
|
|
# One project and one command. Fach-MCPs remain in their own source file, but
|
|
# Compose loads them into this same stack instead of a second project.
|
|
include:
|
|
- path: platform/mcp/compose.yaml
|
|
|
|
x-llama-common: &llama-common
|
|
image: ${LLAMA_IMAGE:-mike-ai/llama.cpp:local}
|
|
restart: "no"
|
|
profiles: [inference]
|
|
gpus: all
|
|
ipc: host
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=1g,mode=1777
|
|
volumes:
|
|
- "${MODEL_DIR:-/srv/mike-ai/models}:/models:ro"
|
|
environment:
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
dns: ["${AI_DNS:-1.1.1.1}"]
|
|
networks:
|
|
inference:
|
|
aliases: [llama-upstream]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
healthcheck:
|
|
test: [CMD, curl, -fsS, "http://127.0.0.1:8080/health"]
|
|
interval: 10s
|
|
timeout: 5s
|
|
retries: 60
|
|
start_period: 30s
|
|
|
|
services:
|
|
wireguard-gateway:
|
|
build: ./platform/docker/wireguard-gateway
|
|
image: mike-ai/wireguard-gateway:local
|
|
container_name: mike-ai-wireguard-gateway
|
|
restart: unless-stopped
|
|
cap_add: [NET_ADMIN]
|
|
devices:
|
|
- /dev/net/tun:/dev/net/tun
|
|
sysctls:
|
|
net.ipv4.ip_forward: "1"
|
|
net.ipv4.conf.all.src_valid_mark: "1"
|
|
net.ipv6.conf.all.forwarding: "1"
|
|
read_only: true
|
|
tmpfs:
|
|
- /run:size=16m,mode=0755
|
|
- /tmp:size=16m,mode=1777
|
|
volumes:
|
|
- "${WIREGUARD_CONFIG_FILE:-/etc/mike-ai/wireguard/fritz-athena.conf}:/run/secrets/fritz-athena.conf:ro"
|
|
# Namespace-sharing services cannot publish ports themselves. The owner
|
|
# must keep these bindings so a gateway recreation cannot hide their UIs.
|
|
ports:
|
|
- "8099:8099"
|
|
- "9443:9443"
|
|
networks:
|
|
frontend:
|
|
ipv4_address: 172.30.10.254
|
|
tools:
|
|
ipv4_address: 172.30.40.254
|
|
tools-egress:
|
|
ipv4_address: 172.30.50.254
|
|
security_opt: ["no-new-privileges:true"]
|
|
healthcheck:
|
|
test: [CMD, /usr/local/sbin/mike-ai-wireguard-healthcheck]
|
|
interval: 10s
|
|
timeout: 3s
|
|
retries: 12
|
|
start_period: 10s
|
|
|
|
llama-fast:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-fast
|
|
labels:
|
|
com.mike-ai.llama-profile: fast
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${FAST_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
# CUDA0 remains the exclusive text-model device. The projector is kept
|
|
# on the secondary card so vision does not consume the 5080 context
|
|
# budget.
|
|
MTMD_BACKEND_DEVICE: CUDA1
|
|
command:
|
|
- --model
|
|
- "/models/${FAST_MODEL_FILE:?FAST_MODEL_FILE is required}"
|
|
- --mmproj
|
|
- "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
|
|
- --mmproj-offload
|
|
- --mmproj-device
|
|
- CUDA1
|
|
- --alias
|
|
- qwen-fast
|
|
- --ctx-size
|
|
- "${FAST_CONTEXT:-76800}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
# Keep the cross-chat prefix cache explicit. On-disk slot restore stays
|
|
# disabled until the current upstream restore regressions are fixed.
|
|
- --cache-prompt
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${FAST_BATCH_SIZE:-64}"
|
|
- --ubatch-size
|
|
- "${FAST_UBATCH_SIZE:-32}"
|
|
- --parallel
|
|
- "${FAST_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
- "0.2"
|
|
- --top-p
|
|
- "0.8"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0
|
|
- --split-mode
|
|
- none
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "2"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
|
|
llama-medium:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-medium
|
|
labels:
|
|
com.mike-ai.llama-profile: medium
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${MEDIUM_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
MTMD_BACKEND_DEVICE: CUDA1
|
|
command:
|
|
- --model
|
|
- "/models/${MEDIUM_MODEL_FILE:?MEDIUM_MODEL_FILE is required}"
|
|
- --mmproj
|
|
- "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
|
|
- --mmproj-offload
|
|
- --mmproj-device
|
|
- CUDA1
|
|
- --alias
|
|
- qwen-medium
|
|
- --ctx-size
|
|
- "${MEDIUM_CONTEXT:-160000}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
- --cache-prompt
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${MEDIUM_BATCH_SIZE:-2048}"
|
|
- --ubatch-size
|
|
- "${MEDIUM_UBATCH_SIZE:-128}"
|
|
- --parallel
|
|
- "${MEDIUM_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
# No fixed --reasoning-budget here: the router supplies a real budget
|
|
# per request from the client's reasoning_effort selection.
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
# Qwen3.8's official thinking-mode sampler. The former 0.2 setting was
|
|
# overly deterministic and could lock reasoning into verbatim loops.
|
|
- "1.0"
|
|
- --top-p
|
|
- "0.95"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0,CUDA1
|
|
- --main-gpu
|
|
- "0"
|
|
- --split-mode
|
|
- layer
|
|
- --tensor-split
|
|
- "${MEDIUM_TENSOR_SPLIT:-85,15}"
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "3"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
- --spec-draft-p-min
|
|
- "0.05"
|
|
|
|
# Beta 1 keeps the complete GSQ-RCO text model and KV cache on the RTX 5080.
|
|
# Only the multimodal projector runs on the RTX 3060. The measured hard
|
|
# boundary is 196608 tokens; 192K deliberately retains runtime headroom.
|
|
llama-beta1:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-beta1
|
|
labels:
|
|
com.mike-ai.llama-profile: beta1
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${BETA1_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
MTMD_BACKEND_DEVICE: CUDA1
|
|
command:
|
|
- --model
|
|
- "/models/${BETA1_MODEL_FILE:?BETA1_MODEL_FILE is required}"
|
|
- --mmproj
|
|
- "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
|
|
- --mmproj-offload
|
|
- --mmproj-device
|
|
- CUDA1
|
|
- --alias
|
|
- qwen-beta-1
|
|
- --ctx-size
|
|
- "${BETA1_CONTEXT:-192000}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
- --cache-prompt
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${BETA1_BATCH_SIZE:-2048}"
|
|
- --ubatch-size
|
|
- "${BETA1_UBATCH_SIZE:-128}"
|
|
- --parallel
|
|
- "${BETA1_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
- "1.0"
|
|
- --top-p
|
|
- "0.95"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0
|
|
- --main-gpu
|
|
- "0"
|
|
- --split-mode
|
|
- none
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "3"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
- --spec-draft-p-min
|
|
- "0.05"
|
|
|
|
llama-large:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-large
|
|
labels:
|
|
com.mike-ai.llama-profile: large
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${LARGE_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
MTMD_BACKEND_DEVICE: CUDA1
|
|
command:
|
|
- --model
|
|
- "/models/${LARGE_MODEL_FILE:?LARGE_MODEL_FILE is required}"
|
|
- --mmproj
|
|
- "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
|
|
- --mmproj-offload
|
|
- --mmproj-device
|
|
- CUDA1
|
|
- --alias
|
|
- qwen-large
|
|
- --ctx-size
|
|
- "${LARGE_CONTEXT:-192000}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
- --cache-prompt
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${LARGE_BATCH_SIZE:-2048}"
|
|
- --ubatch-size
|
|
- "${LARGE_UBATCH_SIZE:-128}"
|
|
- --parallel
|
|
- "${LARGE_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
- "0.2"
|
|
- --top-p
|
|
- "0.8"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0,CUDA1
|
|
- --main-gpu
|
|
- "0"
|
|
- --split-mode
|
|
- layer
|
|
- --tensor-split
|
|
- "${LARGE_TENSOR_SPLIT:-86,14}"
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "3"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
|
|
# Text-only maximum-context profile. This exact IQ4_XS-pure / 256K / 80:20
|
|
# combination completed the 220K fill test on RTX 5080 + RTX 3060.
|
|
# Deliberately no vision projector: Ultra prioritizes maximum usable context.
|
|
llama-ultra:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-ultra
|
|
labels:
|
|
com.mike-ai.llama-profile: ultra
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${ULTRA_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
command:
|
|
- --model
|
|
- "/models/${ULTRA_MODEL_FILE:?ULTRA_MODEL_FILE is required}"
|
|
- --alias
|
|
- qwen-ultra
|
|
- --ctx-size
|
|
- "${ULTRA_CONTEXT:-262144}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
- --cache-prompt
|
|
- --cache-reuse
|
|
- "${LLAMA_CACHE_REUSE:-256}"
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${ULTRA_BATCH_SIZE:-2048}"
|
|
- --ubatch-size
|
|
- "${ULTRA_UBATCH_SIZE:-128}"
|
|
- --parallel
|
|
- "${ULTRA_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
- "0.2"
|
|
- --top-p
|
|
- "0.8"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0,CUDA1
|
|
- --main-gpu
|
|
- "0"
|
|
- --split-mode
|
|
- layer
|
|
- --tensor-split
|
|
- "${ULTRA_TENSOR_SPLIT:-80,20}"
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "2"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
|
|
# Deliberately less refusal-prone weight-level ablation. It remains behind
|
|
# the same authenticated router, tool permissions and confirmation guards as
|
|
# every other profile; "uncensored" never means unrestricted tool access.
|
|
llama-uncensored:
|
|
<<: *llama-common
|
|
container_name: mike-ai-llama-uncensored
|
|
labels:
|
|
com.mike-ai.llama-profile: uncensored
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${UNCENSORED_GPU_DEVICES:-0,1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
MTMD_BACKEND_DEVICE: CUDA1
|
|
command:
|
|
- --model
|
|
- "/models/${UNCENSORED_MODEL_FILE:?UNCENSORED_MODEL_FILE is required}"
|
|
- --mmproj
|
|
- "/models/${UNCENSORED_PROJECTOR_FILE:?UNCENSORED_PROJECTOR_FILE is required}"
|
|
- --mmproj-offload
|
|
- --mmproj-device
|
|
- CUDA1
|
|
- --alias
|
|
- qwen-uncensored
|
|
- --ctx-size
|
|
- "${UNCENSORED_CONTEXT:-80000}"
|
|
- --flash-attn
|
|
- "on"
|
|
- --cache-type-k
|
|
- q4_0
|
|
- --cache-type-v
|
|
- q4_0
|
|
- --cache-prompt
|
|
- --cache-ram
|
|
- "${LLAMA_CACHE_RAM_MIB:-32768}"
|
|
- --threads
|
|
- "${LLAMA_THREADS:-6}"
|
|
- --threads-batch
|
|
- "${LLAMA_THREADS_BATCH:-6}"
|
|
- --batch-size
|
|
- "${UNCENSORED_BATCH_SIZE:-2048}"
|
|
- --ubatch-size
|
|
- "${UNCENSORED_UBATCH_SIZE:-128}"
|
|
- --parallel
|
|
- "${UNCENSORED_PARALLEL_SLOTS:-1}"
|
|
- --kv-unified
|
|
- --jinja
|
|
- --reasoning
|
|
- auto
|
|
- --reasoning-preserve
|
|
- --host
|
|
- 0.0.0.0
|
|
- --port
|
|
- "8080"
|
|
- --metrics
|
|
- --fit
|
|
- "off"
|
|
- --n-gpu-layers
|
|
- all
|
|
- --no-mmap
|
|
- --no-ui
|
|
- --temperature
|
|
- "0.2"
|
|
- --top-p
|
|
- "0.8"
|
|
- --top-k
|
|
- "20"
|
|
- --device
|
|
- CUDA0,CUDA1
|
|
- --main-gpu
|
|
- "0"
|
|
- --split-mode
|
|
- layer
|
|
- --tensor-split
|
|
- "${UNCENSORED_TENSOR_SPLIT:-90,10}"
|
|
- --spec-type
|
|
- draft-mtp
|
|
- --spec-draft-n-max
|
|
- "${UNCENSORED_MTP_MAX:-2}"
|
|
- --spec-draft-p-min
|
|
- "0.10"
|
|
- --spec-draft-type-k
|
|
- f16
|
|
- --spec-draft-type-v
|
|
- f16
|
|
|
|
profile-controller:
|
|
build: ./platform/docker/profile-controller
|
|
image: mike-ai/profile-controller:local
|
|
container_name: mike-ai-profile-controller
|
|
restart: unless-stopped
|
|
read_only: true
|
|
tmpfs: ["/tmp:size=16m"]
|
|
volumes:
|
|
- /var/run/docker.sock:/var/run/docker.sock
|
|
environment:
|
|
CONTROLLER_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
|
|
ALLOWED_PROFILES: fast,medium,beta1,large,ultra,uncensored
|
|
IMAGE_WORKER: image
|
|
networks: [control]
|
|
security_opt: ["no-new-privileges:true"]
|
|
healthcheck:
|
|
test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8090/health', timeout=2)"]
|
|
interval: 10s
|
|
timeout: 3s
|
|
retries: 10
|
|
|
|
router:
|
|
build:
|
|
context: .
|
|
dockerfile: platform/docker/router/Dockerfile
|
|
image: mike-ai/profile-router:local
|
|
container_name: mike-ai-router
|
|
restart: unless-stopped
|
|
read_only: true
|
|
tmpfs: ["/tmp:size=256m"]
|
|
volumes:
|
|
- ./router/router_profiles.json:/etc/mike-ai/router-profiles.json:ro
|
|
- ./config/global-system-policy.txt:/etc/mike-ai/global-system-policy.txt:ro
|
|
- router-state:/var/lib/mike-ai-profile-router
|
|
- router-images:/data/images
|
|
environment:
|
|
ROUTER_HOST: 0.0.0.0
|
|
ROUTER_PORT: "8081"
|
|
ROUTER_AUTH_MODE: required
|
|
ROUTER_API_KEY: "${ROUTER_API_KEY:?ROUTER_API_KEY is required}"
|
|
ROUTER_PROFILES_FILE: /etc/mike-ai/router-profiles.json
|
|
ROUTER_STATE_FILE: /var/lib/mike-ai-profile-router/state.json
|
|
ROUTER_MAX_CONCURRENT_REQUESTS: "16"
|
|
UPSTREAM_URL: http://llama-upstream:8080
|
|
PROFILE_CONTROL_URL: http://profile-controller:8090
|
|
PROFILE_CONTROL_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
|
|
SWITCH_TIMEOUT: "600"
|
|
REQUEST_TIMEOUT: "600"
|
|
# Last-resort guard for every OpenAI-compatible client. Without a
|
|
# request limit llama.cpp uses n_predict=-1 and a reasoning loop can
|
|
# consume the complete context before yielding visible output.
|
|
MAX_GENERATION_TOKENS: "8192"
|
|
DEFAULT_REASONING_EFFORT: "${DEFAULT_REASONING_EFFORT:-off}"
|
|
GLOBAL_SYSTEM_POLICY_FILE: /etc/mike-ai/global-system-policy.txt
|
|
IMAGE_DIR: /data/images
|
|
IMAGE_WORKER_URL: http://image-worker:8086
|
|
IMAGE_WORKER_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
|
|
IMAGE_MODEL_NAME: FLUX.2-klein-4B
|
|
CHAT_IMAGE_ALLOW_REMOTE_URLS: "false"
|
|
ENABLE_IMAGE_GENERATION: "true"
|
|
ENABLE_TTS: "true"
|
|
# Stable OpenAI compatibility names remain piper/alloy because an
|
|
# existing Open WebUI database persists those values. The gateway maps
|
|
# alloy to XTTS speaker Annmarie Nele and automatically falls back to
|
|
# Piper if XTTS is unavailable, busy or returns an error.
|
|
TTS_WORKER_URL: http://tts-gateway:8085
|
|
TTS_MODEL: piper
|
|
TTS_VOICES: alloy
|
|
TTS_DEFAULT_VOICE: alloy
|
|
ENABLE_STT: "true"
|
|
STT_WORKER_URL: http://whisper:8084
|
|
STT_TIMEOUT: "300"
|
|
networks: [frontend, control, inference]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
# The entrypoint fixes ownership of fresh named volumes and immediately
|
|
# drops to uid/gid 10002 via gosu before starting the router. Without this
|
|
# narrowly scoped capabilities a clean installation cannot initialize the
|
|
# volumes and then switch to its unprivileged runtime identity.
|
|
cap_add: [CHOWN, SETUID, SETGID]
|
|
healthcheck:
|
|
test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8081/health', timeout=2)"]
|
|
interval: 5s
|
|
timeout: 3s
|
|
retries: 24
|
|
start_period: 5s
|
|
depends_on:
|
|
wireguard-gateway:
|
|
condition: service_healthy
|
|
profile-controller:
|
|
condition: service_healthy
|
|
piper:
|
|
condition: service_healthy
|
|
tts-gateway:
|
|
condition: service_healthy
|
|
whisper:
|
|
condition: service_healthy
|
|
|
|
image-worker:
|
|
build:
|
|
context: platform/docker/image-worker
|
|
args:
|
|
DIFFUSERS_VERSION: ${DIFFUSERS_VERSION:-0.40.0}
|
|
TRANSFORMERS_VERSION: ${TRANSFORMERS_VERSION:-5.15.1}
|
|
ACCELERATE_VERSION: ${ACCELERATE_VERSION:-1.14.0}
|
|
HF_HUB_VERSION: ${HF_HUB_VERSION:-1.28.0}
|
|
image: mike-ai/image-worker:local
|
|
container_name: mike-ai-image-worker
|
|
restart: "no"
|
|
profiles: [image]
|
|
labels:
|
|
com.mike-ai.image-worker: image
|
|
gpus: all
|
|
read_only: true
|
|
tmpfs: ["/tmp:size=1g,mode=1777"]
|
|
volumes:
|
|
- "${FLUX_MODEL_DIR:-/data/models/FLUX.2-klein-4B}:/models/FLUX.2-klein-4B:ro"
|
|
- router-images:/data/images
|
|
environment:
|
|
NVIDIA_VISIBLE_DEVICES: ${IMAGE_GPU_DEVICES:-1}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
WORKER_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
|
|
FLUX_MODEL_DIR: /models/FLUX.2-klein-4B
|
|
IMAGE_DIR: /data/images
|
|
networks: [inference]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
healthcheck:
|
|
test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8086/health', timeout=2)"]
|
|
interval: 5s
|
|
timeout: 3s
|
|
retries: 12
|
|
|
|
piper:
|
|
build:
|
|
context: platform/docker/piper
|
|
args:
|
|
PIPER_TTS_VERSION: ${PIPER_TTS_VERSION:-1.6.0}
|
|
image: mike-ai/piper:local
|
|
container_name: mike-ai-piper
|
|
restart: unless-stopped
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=256m,mode=1777
|
|
volumes:
|
|
- piper-data:/data
|
|
environment:
|
|
PIPER_DATA_DIR: /data
|
|
PIPER_VOICE: ${PIPER_VOICE:-de_DE-thorsten-high}
|
|
PIPER_VOICE_ALIAS: alloy
|
|
PIPER_HOST: 0.0.0.0
|
|
PIPER_PORT: "8085"
|
|
PIPER_MAX_TEXT_CHARS: "8000"
|
|
networks: [frontend]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
cap_add: [CHOWN, SETUID, SETGID]
|
|
healthcheck:
|
|
test: [CMD, curl, -fsS, "http://127.0.0.1:8085/status"]
|
|
interval: 10s
|
|
timeout: 5s
|
|
retries: 30
|
|
start_period: 120s
|
|
|
|
xtts:
|
|
image: ${XTTS_IMAGE:-ghcr.io/coqui-ai/xtts-streaming-server:latest-cuda121@sha256:f7fb3b1f9d4bc88af94da1b5959d8002f1e0b003c97557164034eb8a29f01b90}
|
|
container_name: mike-ai-xtts
|
|
restart: unless-stopped
|
|
deploy:
|
|
resources:
|
|
reservations:
|
|
devices:
|
|
- driver: nvidia
|
|
device_ids:
|
|
- ${XTTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
|
|
capabilities: [gpu]
|
|
read_only: true
|
|
shm_size: 1g
|
|
tmpfs:
|
|
- /tmp:size=1g,mode=1777
|
|
- /root/.cache:size=2g,mode=0700
|
|
volumes:
|
|
- "${XTTS_CACHE_DIR:-/data/models/xtts-v2-cache}:/root/.local/share/tts"
|
|
environment:
|
|
COQUI_TOS_AGREED: "1"
|
|
NVIDIA_VISIBLE_DEVICES: ${XTTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
CUDA_VISIBLE_DEVICES: "0"
|
|
NUM_THREADS: "4"
|
|
networks: [frontend]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
healthcheck:
|
|
test: [CMD, curl, -fsS, "http://127.0.0.1/languages"]
|
|
interval: 10s
|
|
timeout: 5s
|
|
retries: 36
|
|
start_period: 240s
|
|
|
|
tts-gateway:
|
|
build:
|
|
context: platform/docker/tts-gateway
|
|
image: mike-ai/tts-gateway:local
|
|
container_name: mike-ai-tts-gateway
|
|
restart: unless-stopped
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=256m,mode=1777
|
|
environment:
|
|
TTS_GATEWAY_HOST: 0.0.0.0
|
|
TTS_GATEWAY_PORT: "8085"
|
|
XTTS_URL: http://xtts:80
|
|
PIPER_URL: http://piper:8085
|
|
TTS_VOICE_ALIAS: alloy
|
|
XTTS_SPEAKER: Annmarie Nele
|
|
TTS_DEFAULT_LANGUAGE: de
|
|
# Mixed-language clip stitching caused long pauses and unintelligible
|
|
# transitions. Keep full sentences in one stable German voice.
|
|
TTS_CODE_SWITCH_ENABLED: "false"
|
|
XTTS_QUEUE_TIMEOUT: "15"
|
|
XTTS_TIMEOUT: "120"
|
|
# Keep normal sentences intact for natural prosody. This is only the
|
|
# safety ceiling for unusually long sentences.
|
|
XTTS_CHUNK_CHARS: "420"
|
|
# XTTS occasionally inserts multi-second silence inside a phrase.
|
|
XTTS_INTERNAL_SILENCE_TRIGGER_MS: "650"
|
|
XTTS_INTERNAL_SILENCE_KEEP_MS: "220"
|
|
PIPER_TIMEOUT: "120"
|
|
networks: [frontend]
|
|
depends_on:
|
|
piper:
|
|
condition: service_healthy
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
healthcheck:
|
|
test: [CMD, curl, -fsS, "http://127.0.0.1:8085/status"]
|
|
interval: 10s
|
|
timeout: 5s
|
|
retries: 12
|
|
start_period: 10s
|
|
|
|
whisper:
|
|
build:
|
|
context: .
|
|
dockerfile: platform/docker/whisper/Dockerfile
|
|
args:
|
|
WHISPER_CPP_VERSION: ${WHISPER_CPP_VERSION:-v1.9.1}
|
|
image: mike-ai/whisper:local
|
|
container_name: mike-ai-whisper
|
|
restart: unless-stopped
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=2g,mode=1777
|
|
volumes:
|
|
- whisper-data:/models
|
|
environment:
|
|
WHISPER_HOST: 0.0.0.0
|
|
WHISPER_PORT: "8084"
|
|
WHISPER_CLI: /opt/whisper.cpp/build/bin/whisper-cli
|
|
WHISPER_MODEL: /models/ggml-small.bin
|
|
WHISPER_MODEL_URL: ${WHISPER_MODEL_URL:-https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin}
|
|
WHISPER_SERVER_PORT: "8085"
|
|
WHISPER_THREADS: ${WHISPER_THREADS:-8}
|
|
WHISPER_LANGUAGE: ${WHISPER_LANGUAGE:-de}
|
|
networks: [frontend, inference]
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
# The entrypoint supervises whisper-server after dropping it to uid 10004.
|
|
cap_add: [CHOWN, SETUID, SETGID, KILL]
|
|
healthcheck:
|
|
test: [CMD, curl, -fsS, "http://127.0.0.1:8084/status"]
|
|
interval: 10s
|
|
timeout: 5s
|
|
retries: 90
|
|
start_period: 20m
|
|
|
|
llama-dashboard:
|
|
build: ./platform/llama-dashboard
|
|
image: mike-ai/llama-dashboard:local
|
|
container_name: mike-ai-llama-dashboard
|
|
restart: unless-stopped
|
|
network_mode: "service:wireguard-gateway"
|
|
gpus: all
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=16m,mode=1777
|
|
volumes:
|
|
- /proc:/host/proc:ro
|
|
- /data:/host/data:ro
|
|
- /data/models:/host/models:ro
|
|
- /data/llama-dashboard:/var/lib/llama-dashboard
|
|
environment:
|
|
DASHBOARD_HOST: 0.0.0.0
|
|
DASHBOARD_PORT: "8099"
|
|
ROUTER_URL: http://router:8081
|
|
ROUTER_API_KEY: "${ROUTER_API_KEY:?ROUTER_API_KEY is required}"
|
|
HOST_PROC: /host/proc
|
|
HOST_DATA: /host/data
|
|
HOST_MODELS: /host/models
|
|
DASHBOARD_HISTORY_DB: /var/lib/llama-dashboard/history.sqlite3
|
|
DASHBOARD_HISTORY_INTERVAL: "15"
|
|
DASHBOARD_DETAIL_RETENTION_DAYS: "21"
|
|
NVIDIA_VISIBLE_DEVICES: all
|
|
NVIDIA_DRIVER_CAPABILITIES: compute,utility
|
|
depends_on:
|
|
wireguard-gateway:
|
|
condition: service_healthy
|
|
router:
|
|
condition: service_healthy
|
|
security_opt: ["no-new-privileges:true"]
|
|
cap_drop: [ALL]
|
|
healthcheck:
|
|
test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8099/health', timeout=2)"]
|
|
interval: 10s
|
|
timeout: 3s
|
|
retries: 12
|
|
start_period: 10s
|
|
|
|
portainer:
|
|
image: ${PORTAINER_IMAGE:-portainer/portainer-ce@sha256:511f3f06c96fe3b993ebeaafde311c1959cae73a7ef825dba6397d51b450dffa}
|
|
container_name: mike-ai-portainer
|
|
restart: unless-stopped
|
|
network_mode: "service:wireguard-gateway"
|
|
command: [--no-setup-token]
|
|
volumes:
|
|
- /var/run/docker.sock:/var/run/docker.sock
|
|
- portainer-data:/data
|
|
depends_on:
|
|
wireguard-gateway:
|
|
condition: service_healthy
|
|
security_opt: ["no-new-privileges:true"]
|
|
|
|
backup:
|
|
image: ${BACKUP_IMAGE:-offen/docker-volume-backup@sha256:19102d8e59eb1d598cf8c647c2b21100abaadc5a1c808ac643fa612e323c3013}
|
|
container_name: mike-ai-backup
|
|
restart: unless-stopped
|
|
environment:
|
|
BACKUP_CRON_EXPRESSION: "0 */5 * * *"
|
|
BACKUP_FILENAME: "athena-%Y-%m-%dT%H-%M-%S.tar.gz"
|
|
BACKUP_LATEST_SYMLINK: athena-latest.tar.gz
|
|
BACKUP_RETENTION_DAYS: "14"
|
|
BACKUP_PRUNING_PREFIX: athena-
|
|
volumes:
|
|
- /var/run/docker.sock:/var/run/docker.sock:ro
|
|
- /data/docker-backups:/archive
|
|
- /etc/mike-ai:/backup/etc-mike-ai:ro
|
|
- /opt/mike-ai/stack:/backup/stack:ro
|
|
- piper-data:/backup/volumes/piper-data:ro
|
|
- router-state:/backup/volumes/router-state:ro
|
|
- router-images:/backup/volumes/router-images:ro
|
|
- portainer-data:/backup/volumes/portainer-data:ro
|
|
security_opt: ["no-new-privileges:true"]
|
|
|
|
networks:
|
|
frontend:
|
|
internal: false
|
|
ipam:
|
|
config: [{subnet: 172.30.10.0/24}]
|
|
control:
|
|
internal: true
|
|
ipam:
|
|
config: [{subnet: 172.30.20.0/24}]
|
|
inference:
|
|
internal: true
|
|
ipam:
|
|
config: [{subnet: 172.30.30.0/24}]
|
|
tools:
|
|
external: true
|
|
name: mike-ai-tools
|
|
tools-egress:
|
|
external: true
|
|
name: mike-ai-tools-egress
|
|
|
|
volumes:
|
|
piper-data:
|
|
whisper-data:
|
|
router-state:
|
|
router-images:
|
|
portainer-data:
|
|
name: portainer_data
|
|
external: true
|