67 lines
3.0 KiB
Bash
67 lines
3.0 KiB
Bash
# Copy to /root/mike-ai-install.env, edit, chmod 600, then run:
|
|
# sudo ./install.sh --config /root/mike-ai-install.env
|
|
|
|
AI_HOSTNAME=ki-host
|
|
ADMIN_USER=mike
|
|
AI_BIND_ADDRESS=10.77.0.2
|
|
MODEL_DIR=/srv/mike-ai/models
|
|
|
|
# Installing a new NVIDIA driver can require one reboot. In that case this
|
|
# installer exits with code 20; rerun the same command after reboot.
|
|
INSTALL_NVIDIA_DRIVER=true
|
|
# Optional: auf einen im offiziellen NVIDIA-Repository vorhandenen Hauptzweig
|
|
# festlegen (z. B. 610). Leer lassen, um dem aktuellen stabilen Zweig zu folgen.
|
|
NVIDIA_DRIVER_BRANCH=
|
|
NVIDIA_MIN_DRIVER_MAJOR=570
|
|
TEXT_GPU_DEVICES=0
|
|
SECONDARY_GPU_DEVICES=1
|
|
|
|
# WireGuard client. The home peer must route 10.77.0.2/32 back to this host.
|
|
WIREGUARD_ENABLE=true
|
|
WG_INTERFACE=wg0
|
|
WG_ADDRESS=10.77.0.2/32
|
|
WG_HOME_SUBNET=192.168.1.0/24
|
|
WG_PEER_PUBLIC_KEY=REPLACE_WITH_HOME_WIREGUARD_PUBLIC_KEY
|
|
WG_PEER_ENDPOINT=home.example.net:51820
|
|
WG_PEER_PRESHARED_KEY_FILE=
|
|
WG_DNS=192.168.1.1
|
|
WG_ROUTE_AI_INTERNET=true
|
|
|
|
# Exact model artifacts. Never put access tokens in these URLs.
|
|
FAST_MODEL_FILE=qwen-mix/Qwen3.8-27B-IQ4-MIX.gguf
|
|
FAST_MODEL_URL=https://huggingface.co/vmarcelo/Qwen3.8-27B-MIX_GGUF/resolve/main/Qwen3.8-27B-IQ4-MIX.gguf
|
|
FAST_MODEL_SHA256=54879ae8738d5938f46cb3b8cbf16bf42b8c85b7d68d7c73f062b612ec183e36
|
|
MEDIUM_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
MEDIUM_MODEL_URL=https://huggingface.co/jpetrina/Qwen3.8-27B-IQ4_XS-pure-GGUF/resolve/main/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
MEDIUM_MODEL_SHA256=ea5a3c45d407f9b9e5d2c0d647f0ea600f486f6b86b92b56d0823ba073dae675
|
|
LARGE_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
LARGE_MODEL_URL=https://huggingface.co/jpetrina/Qwen3.8-27B-IQ4_XS-pure-GGUF/resolve/main/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
LARGE_MODEL_SHA256=ea5a3c45d407f9b9e5d2c0d647f0ea600f486f6b86b92b56d0823ba073dae675
|
|
ULTRA_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
ULTRA_MODEL_URL=https://huggingface.co/jpetrina/Qwen3.8-27B-IQ4_XS-pure-GGUF/resolve/main/qwen3.8-27b-IQ4_XS-pure.gguf
|
|
ULTRA_MODEL_SHA256=ea5a3c45d407f9b9e5d2c0d647f0ea600f486f6b86b92b56d0823ba073dae675
|
|
EXPERIMENTAL_MODEL_FILE=qwen/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
EXPERIMENTAL_MODEL_URL=https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
EXPERIMENTAL_MODEL_SHA256=40fac4050e940397dbf13087afd50f4734a11805bf9d65ef8ddd7483470e6199
|
|
VISION_PROJECTOR_FILE=qwen/mmproj-BF16.gguf
|
|
VISION_PROJECTOR_URL=https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/mmproj-BF16.gguf
|
|
VISION_PROJECTOR_SHA256=83ee4f4f205fa514161778c41df1ea14144faa0f713510893b63c2395f5c2d53
|
|
# All standard profiles use the MTP tensor embedded in their GGUF. A separate
|
|
# draft-model artifact is neither downloaded nor passed to llama-server.
|
|
|
|
FAST_CONTEXT=76800
|
|
MEDIUM_CONTEXT=160000
|
|
MEDIUM_TENSOR_SPLIT=90,10
|
|
LARGE_CONTEXT=192000
|
|
LARGE_TENSOR_SPLIT=86,14
|
|
ULTRA_CONTEXT=262144
|
|
ULTRA_TENSOR_SPLIT=80,20
|
|
EXPERIMENTAL_CONTEXT=76800
|
|
LLAMA_THREADS=6
|
|
LLAMA_THREADS_BATCH=6
|
|
OPENWEBUI_IMAGE=ghcr.io/open-webui/open-webui:v0.9.5
|
|
OPENWEBUI_ENABLE_SIGNUP=false
|
|
OPENWEBUI_ENABLE_FOLLOW_UP_GENERATION=false
|
|
PIPER_TTS_VERSION=1.6.0
|
|
PIPER_VOICE=de_DE-thorsten-high
|