非rootなdocker環境にllama.cpp+open-webui環境を構築する方法
小型モデルにお勧め．WEB外部検索と日本語グラフ生成付き

sudo apt install uidmap dbus-user-session

which newuidmap
which newgidmap

curl -fsSL https://get.docker.com/rootless | sh

ps aux | grep rootless

pkill rootlesskit
pkill dockerd

#! /bin/bash

# --- LLM Config ---

NAME="llama.cpp-server"

LM_MODEL_DIR="$HOME/AI/llama_models"

## for CPU
#LM_HOME_DIR="$HOME/AI/llama.cpp/build/bin"
#LM_NGL=0 # CPU

## for GPU
LM_HOME_DIR="$HOME/AI/llama.cpp/build-cuda/bin"
LM_NGL=99 # GPU


#LM_CTX=2048 # コンテキスト
#LM_CTX=3072 # コンテキスト
LM_CTX=4096 # コンテキスト
LM_N=1024 # 打ち止め
LM_TEMP=0.1 # 正確さ
OTHER_OPT=''

PIDFILE="$HOME/.${NAME}.pid"
LOGFILE="$HOME/${NAME}.log"

# モデル指定用
ARGS2="$2"

start() {
    if [ -f "$PIDFILE" ] && kill -0 "$(cat "$PIDFILE")" 2>/dev/null; then
	echo "$NAME already running (PID $(cat $PIDFILE))"
	return 0
    fi

    echo "Starting $NAME..."
    # 1. モデルファイルを自動スキャン (.ggufのみ)
    local FILES=($(ls "$LM_MODEL_DIR"/*.gguf 2>/dev/null | xargs -n 1 basename))

    if [ ${#FILES[@]} -eq 0 ]; then
	echo "Error: No .gguf files found in $LM_MODEL_DIR"
	return 1
    fi

    # 2. モデルがない場合メニュー表示
    if [ -z "$ARGS2" ]; then
	echo "--- Found Models in $LM_MODEL_DIR ---"
	for i in "${!FILES[@]}"; do
	    echo "$((i+1))) ${FILES[$i]}"
	done
	read -p "Select number: " MODEL_CH
    else	
	MODEL_CH="$ARGS2"
    fi

    # 3. 選択したモデルの設定
    local SELECTED_FILE="${FILES[$((MODEL_CH-1))]}"
    if [ -z "$SELECTED_FILE" ]; then echo "Invalid selection"; return 1; fi

    LM_MODEL="$SELECTED_FILE"

    "$LM_HOME_DIR/llama-server" -m "$LM_MODEL_DIR/$LM_MODEL" --host 0.0.0.0 --port 8080 -ngl $LM_NGL -c $LM_CTX $OTHER_OPT  >> "$LOGFILE" 2>&1 &
    echo $! > "$PIDFILE"
    echo "$NAME started (PID $(cat $PIDFILE)) MODEL: $LM_MODEL"
}


stop() {
    if [ ! -f "$PIDFILE" ]; then
	echo "$NAME not running (no PID file)"
	return 1
    fi

    PID=$(cat "$PIDFILE")

    if kill -0 "$PID" 2>/dev/null; then
	echo "Stopping $NAME (PID $PID)..."
	kill "$PID"
	sleep 2
    fi

    if kill -0 "$PID" 2>/dev/null; then
	echo "Force killing $NAME..."
	kill -9 "$PID"
    fi

    rm -f "$PIDFILE"
    echo "$NAME stopped"
}

status() {
    if [ -f "$PIDFILE" ] && kill -0 "$(cat "$PIDFILE")" 2>/dev/null; then
	echo "$NAME running (PID $(cat $PIDFILE))"
	return 0
    else
	echo "$NAME not running"
	return 1
    fi
}


case "$1" in
    start) start ;;
    stop) stop ;;
    restart) stop; start ;;
    status) status ;;
    *)
	echo "Usage: $0 {start|stop|restart|status} [model_no]"
	exit 1
	;;
esac

jupyter notebook --ip 0.0.0.0 --port 8888

jupyter server list

docker run -d \
  -p 3000:8080 \
  -v ~/ai/open-webui-data:/app/backend/data \
  --name open-webui \
  --restart unless-stopped \
  ghcr.io/open-webui/open-webui:main

mkdir -p ~/ai/open-webui-data

docker volume ls

docker run --rm \
-v open-webui:/source \
-v ~/ai/open-webui-data:/dest \
alpine sh -c "cp -a /source/* /dest/"

docker rm -f open-webui

jupyter server list

#!/usr/bin/env bash

DATA="$HOME/ai/open-webui-data"

echo "Open WebUI Cleaner"
echo "Data directory: $DATA"
echo

if [ ! -d "$DATA" ]; then
echo "Directory not found"
exit 1
fi

echo "Current usage:"
du -sh "$DATA"
echo

echo "Detail:"
du -sh "$DATA"/* 2>/dev/null
echo

BEFORE=$(du -sb "$DATA" | cut -f1)

read -p "Clean generated data? (y/N): " ans
[ "$ans" != "y" ] && exit 0

echo
echo "Cleaning..."

rm -rf "$DATA/uploads/"*
rm -rf "$DATA/cache/"*
rm -rf "$DATA/vector_db/"*
rm -rf "$DATA/documents/"*
rm -rf "$DATA/logs/"*

echo
AFTER=$(du -sb "$DATA" | cut -f1)

echo "After cleaning:"
du -sh "$DATA"
echo

echo "Detail:"
du -sh "$DATA"/* 2>/dev/null
echo

DIFF=$((BEFORE-AFTER))

echo "Freed space:"
numfmt --to=iec $DIFF

start-stop-llama.sh start 1

jupyter notebook --ip 0.0.0.0 --port 8888

start-stop-docker.sh start

ai start|stop|restart|status|clean|gpu|force-kill n

#!/usr/bin/env bash

LLAMA="$HOME/bin/Start-stop-llama-server.sh"
JUPYTER="$HOME/bin/Start-stop-jupyter.sh"
DOCKER="$HOME/bin/Start-stop-docker.sh"

MODEL_DIR="$HOME/AI/llama_models"

MODEL=${2:-1}

start() {
    echo "Starting AI stack"

    $LLAMA start "$MODEL" || exit 1
	sleep 2

    $JUPYTER start || exit 1
	sleep 2

    $DOCKER start || exit 1

    echo "AI stack ready"
    }

stop() {
    echo "Stopping AI stack"

    $DOCKER stop
	$JUPYTER stop
	    $LLAMA stop

    echo "Stopped"
    }

restart() {
    stop
	sleep 2
	    start "$MODEL"
	    }

status() {
    echo "=== LLAMA ==="
	$LLAMA status
	    echo

    echo "=== JUPYTER ==="
	$JUPYTER status
	    echo

    echo "=== DOCKER ==="
	$DOCKER status
	    echo
	}

logs() {
    echo "=== llama-server log ==="
	tail -n 20 ~/.llama.log 2>/dev/null
	    echo

    echo "=== jupyter log ==="
	tail -n 20 ~/.jupyter.log 2>/dev/null
	    echo

    echo "=== docker containers ==="
	docker ps
	}

models() {
    echo "Available models:"
	ls -1 "$MODEL_DIR"
	}

gpu() {
    if command -v nvidia-smi >/dev/null; then
	nvidia-smi
    else
	echo "GPU info unavailable"
    fi
}
clean() {
    ~/bin/open-webui-cleaner.sh
}

case "$1" in
start) start "$@" ;;
stop) stop ;;
restart) restart "$@" ;;
status) status ;;
logs) logs ;;
models) models ;;
gpu) gpu ;;
clean) celan ;;
*)
echo "Usage:"
echo "  ai start [model]"
echo "  ai stop"
echo "  ai restart [model]"
echo "  ai status"
echo "  ai logs"
echo "  ai models"
echo "  ai gpu"
echo "  ai clean"
exit 1
;;
esac

ai start 1

ai stop

ai status