57 lines
2.5 KiB
Bash
57 lines
2.5 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/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
FAST_MODEL_URL=https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
FAST_MODEL_SHA256=40fac4050e940397dbf13087afd50f4734a11805bf9d65ef8ddd7483470e6199
|
|
MEDIUM_MODEL_FILE=qwen/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
MEDIUM_MODEL_URL=https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
MEDIUM_MODEL_SHA256=40fac4050e940397dbf13087afd50f4734a11805bf9d65ef8ddd7483470e6199
|
|
LONG_MODEL_FILE=qwen/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
LONG_MODEL_URL=https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/resolve/main/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
LONG_MODEL_SHA256=40fac4050e940397dbf13087afd50f4734a11805bf9d65ef8ddd7483470e6199
|
|
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
|
|
# FAST/LONG use the MTP tensor embedded in the IQ4-MIX GGUF. A separate
|
|
# draft-model artifact is neither downloaded nor passed to llama-server.
|
|
|
|
FAST_CONTEXT=76800
|
|
MEDIUM_CONTEXT=94208
|
|
LONG_CONTEXT=131072
|
|
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
|