Skip to content

Instantly share code, notes, and snippets.

@casjay

casjay/llm.sh

Created July 10, 2025 00:05
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save casjay/9f483a38e27840d4f7074383ff6086a7 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save casjay/9f483a38e27840d4f7074383ff6086a7 to your computer and use it in GitHub Desktop.
Download ZIP
Raw
llm.sh
#!/usr/bin/env bash
# 🛣️ setup-stack.sh
# Fully automated, GPU-aware AI stack installer.
# Includes: Ollama, LocalAI, OpenWebUI, OpenDiffusion + VSCode LSP, unified config, stateful tracking, port resolution, bind mounts.
set -euo pipefail
# =========================
# 📁 Directory + Variables
# =========================
STACK_ROOT="/var/lib/localai"
MODELS_DIR="$STACK_ROOT/models"
LOG_DIR="/var/log/localai"
STATE_DIR="$STACK_ROOT/state"
CONFIG_DIR="$STACK_ROOT/config"
INSTALL_STATE_FILE="$STATE_DIR/install.state"
OPENWEBUI_DIR="$STACK_ROOT/openwebui"
OPENDIFFUSION_DIR="$STACK_ROOT/opendiffusion"
OLLAMA_DIR="$STACK_ROOT/ollama"
LOCALAI_CONFIG_FILE="$CONFIG_DIR/config.yaml"
mkdir -p "$MODELS_DIR" "$LOG_DIR" "$STATE_DIR" "$CONFIG_DIR" \
"$OPENWEBUI_DIR" "$OPENDIFFUSION_DIR" "$OLLAMA_DIR"
chmod -R a+rwx "$STACK_ROOT" "$LOG_DIR"
# =========================
# ⚙️ Flags
# =========================
DEBUG=false
DRY_RUN=false
UNINSTALL=false
UPDATE=false
STATUS=false
RAW=false
for arg in "$@"; do
case "$arg" in
--debug) DEBUG=true ;;
--dry-run) DRY_RUN=true ;;
--uninstall) UNINSTALL=true ;;
--update) UPDATE=true ;;
--status|--summary) STATUS=true ;;
--raw) RAW=true ;;
esac
done
log() { [[ "$DEBUG" == true ]] && echo "[DEBUG] $*" || true; }
run() { [[ "$DRY_RUN" == true ]] && echo "+ $*" || eval "$@"; }
require_cmd() { command -v "$1" >/dev/null || { echo "❌ Missing required command: $1"; exit 1; }; }
# =========================
# 🧠 Detection + Ports
# =========================
has_nvidia_gpu() { command -v nvidia-smi >/dev/null && nvidia-smi -L | grep -q GPU; }
has_amd_gpu() { lspci | grep -i amd | grep -qi vga; }
get_total_ram() { awk '/MemTotal/ {print int($2/1024/1024)}' /proc/meminfo; }
get_free_disk() { df --output=avail -BG "$STACK_ROOT" | tail -1 | tr -dc '0-9'; }
CPU_MODE=true
if has_nvidia_gpu || has_amd_gpu; then CPU_MODE=false; fi
find_free_port() {
local port=62000
while ss -tuln | grep -q ":$port"; do ((port++)); done
echo "$port"
}
# =========================
# 🔧 LocalAI Config
# =========================
generate_localai_config() {
cat >"$LOCALAI_CONFIG_FILE" <<EOF
server:
host: 0.0.0.0
port: $LOCALAI_PORT
workers: 4
threads: 8
timeout: 120
debug: false
gpu:
enabled: $(if [[ "$CPU_MODE" == false ]]; then echo true; else echo false; fi)
log_level: info
log_json: true
models-path: "$MODELS_DIR"
models:
- name: mistral
backend: llama
path: "$MODELS_DIR/mistral"
parameters:
temperature: 0.7
- name: llama3
backend: llama
path: "$MODELS_DIR/llama3"
parameters:
temperature: 0.7
EOF
echo "localai_config_generated=true" >> "$INSTALL_STATE_FILE"
}
# =========================
# 🧠 Ollama + Models
# =========================
install_ollama() {
if command -v ollama >/dev/null; then log "Ollama already installed"; echo "ollama_installed=true" >> "$INSTALL_STATE_FILE"; return; fi
ARCH=$(uname -m); OS=$(uname | tr '[:upper:]' '[:lower:]')
ARCH=${ARCH/x86_64/amd64}; ARCH=${ARCH/aarch64/arm64}
TARBALL="ollama-$OS-$ARCH.tgz"
TMP=$(mktemp -d)
run curl -fsSL -o "$TMP/$TARBALL" "https://ollama.com/download/$TARBALL"
run tar -xzf "$TMP/$TARBALL" -C "$TMP"
run install -Dm755 "$TMP/ollama" /usr/local/bin/ollama
rm -rf "$TMP"
echo "ollama_installed=true" >> "$INSTALL_STATE_FILE"
}
select_models() {
local ram=$(get_total_ram)
local disk=$(get_free_disk)
local models=("mistral" "llama3")
if [[ "$CPU_MODE" == false ]]; then
models+=("deepseek-coder")
((ram >= 32 && disk >= 30)) && models+=("codellama")
fi
echo "${models[@]}"
}
prefetch_model() {
local model="$1"
run ollama pull "$model"
echo "model_${model}_pulled=true" >> "$INSTALL_STATE_FILE"
}
# =========================
# 💻 VSCode Integration
# =========================
configure_vscode() {
require_cmd jq
command -v code >/dev/null || return
log "⚙️ Configuring VSCode extensions"
for ext in \
continue.continue ms-python.python ms-toolsai.jupyter \
streetsidesoftware.code-spell-checker ms-python.vscode-pylance \
golang.go rust-lang.rust-analyzer dbaeumer.vscode-eslint \
esbenp.prettier-vscode timonwong.shellcheck redhat.vscode-yaml \
yzhang.markdown-all-in-one ms-azuretools.vscode-docker \
be5invis.toml ms-vscode.cpptools redhat.java \
ecmel.vscode-html-css aeschli.vscode-css-formatter \
sumneko.lua editorconfig.editorconfig; do
run code --install-extension "$ext" || true
done
local cfg="$HOME/.config/Code/User/settings.json"
mkdir -p "$(dirname "$cfg")"
jq -s 'reduce .[] as $item ({}; . * $item)' \
<(cat "$cfg" 2>/dev/null || echo '{}') \
<(cat <<EOF
{
"continue.serverUrl": "http://localhost:$LOCALAI_PORT",
"continue.backend": "ollama",
"cSpell.enabled": true,
"cSpell.language": "en,technical",
"python.languageServer": "Pylance",
"python.formatting.provider": "black",
"python.linting.enabled": true,
"python.linting.pylintEnabled": true,
"python.linting.flake8Enabled": true,
"go.useLanguageServer": true,
"go.formatTool": "gofmt",
"go.lintTool": "golint",
"rust-analyzer.cargo.allFeatures": true,
"rust-analyzer.procMacro.enable": true,
"shellcheck.enable": true,
"yaml.validate": true,
"yaml.schemas": {
"https://json.schemastore.org/github-workflow.json": "/.github/workflows/*"
},
"html.format.enable": true,
"css.validate": true
}
EOF
) > "$cfg"
echo "vscode_configured=true" >> "$INSTALL_STATE_FILE"
}
# =========================
# 📊 Status Summary
# =========================
status_summary() {
echo -e "\n📊 Stack Summary"
echo "---------------------------"
printf "CPU: %s\n" "$(lscpu | grep 'Model name' | cut -d ':' -f2 | xargs)"
printf "RAM: %s GB\n" "$(get_total_ram)"
printf "Free Disk: %s GB\n" "$(get_free_disk)"
printf "NVIDIA GPU: %s\n" "$(has_nvidia_gpu && echo Yes || echo No)"
printf "AMD GPU: %s\n" "$(has_amd_gpu && echo Yes || echo No)"
printf "Models Dir: %s\n" "$MODELS_DIR"
printf "Log Dir: %s\n" "$LOG_DIR"
printf "VSCode: %s\n" "$(command -v code &>/dev/null && echo Yes || echo No)"
printf "Ollama Installed:%s\n" "$(command -v ollama &>/dev/null && echo Yes || echo No)"
printf "Install State: %s\n" "$INSTALL_STATE_FILE"
echo "OpenWebUI Port: $OPENWEBUI_PORT"
echo "OpenDiffusion Port: $OPENDIFFUSION_PORT"
echo "LocalAI Port: $LOCALAI_PORT"
echo "---------------------------"
}
# =========================
# 🧼 CLI Entrypoints
# =========================
[[ "$UNINSTALL" == true ]] && {
echo "🧼 Uninstalling..."
run docker rm -f openwebui open-diffusion localai || true
rm -rf "$STACK_ROOT" "$LOG_DIR"
echo "✅ Removed"
exit 0
}
[[ "$STATUS" == true ]] && { status_summary; exit 0; }
# =========================
# 🚀 Main Install Logic
# =========================
install_ollama
[[ "$UPDATE" == true ]] && run docker pull ghcr.io/open-webui/open-webui:latest
[[ "$CPU_MODE" == true ]] && echo "⚠️ CPU-only mode: expect slower inference."
echo "🧠 Prefetching models..."
for m in $(select_models); do prefetch_model "$m"; done
OPENWEBUI_PORT=$(find_free_port)
OPENDIFFUSION_PORT=$(find_free_port)
LOCALAI_PORT=$(find_free_port)
generate_localai_config
echo "🔁 Stopping old containers..."
for c in openwebui open-diffusion localai; do
run docker rm -f "$c" || true
done
echo "📦 Launching containers..."
run docker run -d --name openwebui --restart=always \
-p "$OPENWEBUI_PORT:3000" \
-v "$OPENWEBUI_DIR:/app/data" \
-v "$MODELS_DIR:/models" \
-v "$LOG_DIR/openwebui:/var/log/localai/openwebui" \
ghcr.io/open-webui/open-webui:latest
run docker run -d --name open-diffusion --restart=always \
-p "$OPENDIFFUSION_PORT:7860" \
-v "$OPENDIFFUSION_DIR:/data" \
-v "$MODELS_DIR:/models" \
-v "$LOG_DIR/opendiffusion:/var/log/localai/opendiffusion" \
ghcr.io/opensd/open-diffusion:latest
run docker run -d --name localai --restart=always \
-p "$LOCALAI_PORT:8080" \
-v "$OLLAMA_DIR:/data" \
-v "$MODELS_DIR:/models" \
-v "$LOG_DIR/localai:/var/log/localai/localai" \
-v "$LOCALAI_CONFIG_FILE:/etc/localai/config.yaml" \
quay.io/go-skynet/local-ai:latest
configure_vscode
status_summary
echo -e "\n🎉 Stack ready! Access points:"
echo "- 🌐 OpenWebUI: http://localhost:$OPENWEBUI_PORT"
echo "- 🎨 Diffusion: http://localhost:$OPENDIFFUSION_PORT"
echo "- 🤖 LocalAI API: http://localhost:$LOCALAI_PORT"
echo "- 🧠 VSCode: Ready with Continue + LSP"
echo "- 📦 Models: $MODELS_DIR"
echo "- 🪵 Logs: $LOG_DIR"
echo ""
echo "🔧 To stop: docker stop openwebui open-diffusion localai"
echo "🧼 To uninstall: run this script with --uninstall"
echo "⬆️ To update: run this script with --update"
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment