112341e162
Stabilize Windows+WSL setup by fixing script line-ending pitfalls, WSL Ollama host detection, and dashboard startup behavior so models are detected reliably in Mission Control. Co-authored-by: Cursor <cursoragent@cursor.com>
335 lines
12 KiB
Bash
335 lines
12 KiB
Bash
#!/bin/bash
|
|
# ============================================================
|
|
# WSL2-Side Setup - Local LLM Stack (Razer Blade 18)
|
|
#
|
|
# Run this INSIDE WSL2 (Ubuntu 24.04) after setup-windows.ps1.
|
|
# This is a one-shot script - safe to re-run (idempotent).
|
|
#
|
|
# What this does:
|
|
# 1. Installs system deps (Node.js 20, Python 3.12, ffmpeg, cmake)
|
|
# 2. Verifies NVIDIA GPU passthrough (CUDA)
|
|
# 3. Clones the repo (if not already cloned)
|
|
# 4. Builds whisper.cpp with CUDA (auto-detects 120 for Blackwell/RTX 50 when nvcc 12.8+)
|
|
# 5. Downloads Whisper large-v3-turbo model
|
|
# 6. Runs setup-tts.sh (Orpheus + Qwen3-TTS)
|
|
# 7. Starts the Mission Control dashboard
|
|
#
|
|
# Usage:
|
|
# bash setup-wsl.sh
|
|
#
|
|
# When run from inside the repo (e.g. script path under __LOCAL_LLMs/), uses
|
|
# that repo and skips clone. Or set EXISTING_REPO_PATH=/path/to/repo to use it.
|
|
#
|
|
# Or one-liner from a fresh WSL2 terminal:
|
|
# curl -fsSL https://raw.githubusercontent.com/saravanakumardb1/learning_ai_common_plat/main/__LOCAL_LLMs/windows_specific/setup-wsl.sh | bash
|
|
# ============================================================
|
|
set -e
|
|
|
|
RED='\033[0;31m'
|
|
GREEN='\033[0;32m'
|
|
YELLOW='\033[1;33m'
|
|
CYAN='\033[0;36m'
|
|
NC='\033[0m'
|
|
|
|
ok() { echo -e " ${GREEN}[OK]${NC} $1"; }
|
|
warn() { echo -e " ${YELLOW}[!!]${NC} $1"; }
|
|
fail() { echo -e " ${RED}[FAIL]${NC} $1"; exit 1; }
|
|
step() { echo -e "\n${CYAN}=== $1 ===${NC}"; }
|
|
|
|
REPO_URL="https://github.com/saravanakumardb1/learning_ai_common_plat.git"
|
|
REPO_DIR="$HOME/code/mygh/learning_ai_common_plat"
|
|
|
|
# Use existing repo if run from repo path or EXISTING_REPO_PATH is set
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
|
|
if [ -n "${EXISTING_REPO_PATH:-}" ]; then
|
|
REPO_DIR="$(cd "$EXISTING_REPO_PATH" && pwd)"
|
|
elif [ -d "$SCRIPT_DIR/../.." ] && [ -d "$SCRIPT_DIR/../../__LOCAL_LLMs" ]; then
|
|
REPO_DIR="$(cd "$SCRIPT_DIR/../.." && pwd)"
|
|
fi
|
|
|
|
LLMS_DIR="$REPO_DIR/__LOCAL_LLMs"
|
|
WHISPER_DIR="$HOME/whisper-cpp"
|
|
WHISPER_MODEL_DIR="$HOME/whisper-models"
|
|
WHISPER_MODEL_FILE="$WHISPER_MODEL_DIR/ggml-large-v3-turbo.bin"
|
|
|
|
echo ""
|
|
echo -e " ${CYAN}=====================================${NC}"
|
|
echo -e " ${CYAN}Local LLM Stack - WSL2 Setup${NC}"
|
|
echo -e " ${CYAN}Ubuntu 24.04 + CUDA${NC}"
|
|
echo -e " ${CYAN}=====================================${NC}"
|
|
echo ""
|
|
|
|
# -- 1. System Dependencies -----------------------------------
|
|
step "Step 1/7: System Dependencies"
|
|
|
|
sudo apt update -qq
|
|
|
|
# Node.js 20 LTS
|
|
if command -v node &>/dev/null && [[ "$(node --version)" == v20* || "$(node --version)" == v22* ]]; then
|
|
ok "Node.js $(node --version) already installed"
|
|
else
|
|
echo " Installing Node.js 20 LTS..."
|
|
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash - 2>/dev/null
|
|
sudo apt install -y nodejs -qq
|
|
ok "Node.js $(node --version) installed"
|
|
fi
|
|
|
|
# Python 3.12
|
|
PYTHON_CMD=""
|
|
for cmd in python3.12 python3; do
|
|
if command -v "$cmd" &>/dev/null; then
|
|
PYTHON_CMD="$cmd"
|
|
break
|
|
fi
|
|
done
|
|
|
|
if [ -z "$PYTHON_CMD" ]; then
|
|
echo " Installing Python 3.12..."
|
|
sudo apt install -y python3.12 python3.12-venv python3-pip -qq
|
|
PYTHON_CMD="python3.12"
|
|
fi
|
|
ok "$($PYTHON_CMD --version) at $(which $PYTHON_CMD)"
|
|
|
|
# Build tools + ffmpeg
|
|
PKGS_TO_INSTALL=""
|
|
for pkg in ffmpeg git curl build-essential cmake; do
|
|
if ! dpkg -s "$pkg" &>/dev/null; then
|
|
PKGS_TO_INSTALL="$PKGS_TO_INSTALL $pkg"
|
|
fi
|
|
done
|
|
|
|
if [ -n "$PKGS_TO_INSTALL" ]; then
|
|
echo " Installing:$PKGS_TO_INSTALL"
|
|
sudo apt install -y $PKGS_TO_INSTALL -qq
|
|
fi
|
|
ok "ffmpeg, git, curl, build-essential, cmake"
|
|
|
|
# -- 2. NVIDIA GPU Passthrough ---------------------------------
|
|
step "Step 2/7: NVIDIA GPU (CUDA passthrough)"
|
|
|
|
if command -v nvidia-smi &>/dev/null; then
|
|
GPU_NAME=$(nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null | head -1)
|
|
GPU_MEM=$(nvidia-smi --query-gpu=memory.total --format=csv,noheader 2>/dev/null | head -1)
|
|
ok "GPU: $GPU_NAME ($GPU_MEM)"
|
|
else
|
|
warn "nvidia-smi not found!"
|
|
echo " Possible fixes:"
|
|
echo " 1. Update NVIDIA drivers on Windows to the latest version"
|
|
echo " 2. Run in PowerShell (Admin): wsl --update"
|
|
echo " 3. Do NOT install nvidia-driver-* inside WSL2"
|
|
echo ""
|
|
read -p " Continue without CUDA? (y/N) " -n 1 -r
|
|
echo
|
|
if [[ ! $REPLY =~ ^[Yy]$ ]]; then exit 1; fi
|
|
fi
|
|
|
|
# -- 3. Verify Ollama (Windows-side) ---------------------------
|
|
step "Step 3/7: Ollama Connectivity"
|
|
|
|
OLLAMA_URL="${OLLAMA_HOST:-http://localhost:11434}"
|
|
OLLAMA_URL="${OLLAMA_URL%/}"
|
|
# WSL2: if localhost fails, try Windows host via default gateway
|
|
if ! curl -s --max-time 3 "$OLLAMA_URL/api/tags" &>/dev/null; then
|
|
if [ -r /proc/version ] && grep -qi microsoft /proc/version 2>/dev/null; then
|
|
GW=$(ip route show default 2>/dev/null | awk '{print $3}' | head -1)
|
|
if [ -n "$GW" ] && curl -s --max-time 3 "http://${GW}:11434/api/tags" &>/dev/null; then
|
|
OLLAMA_URL="http://${GW}:11434"
|
|
fi
|
|
fi
|
|
fi
|
|
export OLLAMA_HOST="$OLLAMA_URL"
|
|
|
|
if curl -s --max-time 3 "$OLLAMA_URL/api/tags" &>/dev/null; then
|
|
MODEL_COUNT=$(curl -s "$OLLAMA_URL/api/tags" | python3 -c "import sys,json; print(len(json.load(sys.stdin).get('models',[])))" 2>/dev/null || echo "?")
|
|
ok "Ollama reachable at $OLLAMA_URL ($MODEL_COUNT models)"
|
|
else
|
|
warn "Ollama not reachable at $OLLAMA_URL"
|
|
echo " Make sure Ollama is running on Windows with OLLAMA_HOST=0.0.0.0:11434"
|
|
echo " Open PowerShell and run: ollama serve"
|
|
echo ""
|
|
read -p " Continue anyway? (y/N) " -n 1 -r
|
|
echo
|
|
if [[ ! $REPLY =~ ^[Yy]$ ]]; then exit 1; fi
|
|
fi
|
|
|
|
# -- 4. Clone Repo ----------------------------------------------
|
|
step "Step 4/7: Clone Repository"
|
|
|
|
DEFAULT_REPO_DIR="$HOME/code/mygh/learning_ai_common_plat"
|
|
# Performance: /mnt/* is slower for node_modules. Keep setup fast by default.
|
|
# Optional: set MIRROR_TO_WSL_LOCAL=1 to copy repo to $DEFAULT_REPO_DIR.
|
|
if [[ "$REPO_DIR" == /mnt/* ]]; then
|
|
warn "Repo is on Windows mount: $REPO_DIR"
|
|
echo " This works, but dashboard/node_modules can be slower on /mnt/*."
|
|
if [ -d "$DEFAULT_REPO_DIR/__LOCAL_LLMs" ]; then
|
|
REPO_DIR="$DEFAULT_REPO_DIR"
|
|
LLMS_DIR="$REPO_DIR/__LOCAL_LLMs"
|
|
ok "Using existing WSL-local repo at $REPO_DIR"
|
|
elif [ "${MIRROR_TO_WSL_LOCAL:-0}" = "1" ] && [ -d "$REPO_DIR/.git" ]; then
|
|
echo " MIRROR_TO_WSL_LOCAL=1 -> copying repo to:"
|
|
echo " $DEFAULT_REPO_DIR"
|
|
mkdir -p "$(dirname "$DEFAULT_REPO_DIR")"
|
|
cp -a "$REPO_DIR" "$DEFAULT_REPO_DIR"
|
|
REPO_DIR="$DEFAULT_REPO_DIR"
|
|
LLMS_DIR="$REPO_DIR/__LOCAL_LLMs"
|
|
ok "Copied repo to $REPO_DIR"
|
|
else
|
|
echo " Continuing on current path for speed."
|
|
echo " Tip: to mirror once into WSL-local FS, re-run with:"
|
|
echo " MIRROR_TO_WSL_LOCAL=1 bash setup-wsl.sh"
|
|
fi
|
|
fi
|
|
|
|
if [ -d "$LLMS_DIR" ]; then
|
|
ok "Repo already cloned at $REPO_DIR"
|
|
# Only pull when using the default clone path.
|
|
if [ "$REPO_DIR" = "$DEFAULT_REPO_DIR" ]; then
|
|
echo " Pulling latest..."
|
|
(cd "$REPO_DIR" && git pull --ff-only origin main 2>/dev/null || true)
|
|
fi
|
|
else
|
|
echo " Cloning into $REPO_DIR..."
|
|
mkdir -p "$(dirname "$REPO_DIR")"
|
|
git clone "$REPO_URL" "$REPO_DIR"
|
|
ok "Cloned"
|
|
fi
|
|
|
|
# Verify __LOCAL_LLMs exists
|
|
if [ ! -d "$LLMS_DIR" ]; then
|
|
fail "__LOCAL_LLMs directory not found at $LLMS_DIR"
|
|
fi
|
|
ok "__LOCAL_LLMs directory: $LLMS_DIR"
|
|
|
|
# -- 5. Build Whisper.cpp (CUDA) ---------------------------------
|
|
step "Step 5/7: Whisper.cpp (CUDA build)"
|
|
|
|
if command -v whisper-cli &>/dev/null; then
|
|
ok "whisper-cli already installed: $(which whisper-cli)"
|
|
else
|
|
if [ ! -d "$WHISPER_DIR" ]; then
|
|
echo " Cloning whisper.cpp..."
|
|
git clone https://github.com/ggerganov/whisper.cpp.git "$WHISPER_DIR"
|
|
fi
|
|
|
|
echo " Building with CUDA support (this may take a few minutes)..."
|
|
cd "$WHISPER_DIR"
|
|
|
|
# Check if CUDA toolkit headers are available for build
|
|
if command -v nvidia-smi &>/dev/null; then
|
|
# Need CUDA toolkit for compilation
|
|
if ! dpkg -s nvidia-cuda-toolkit &>/dev/null 2>&1; then
|
|
echo " Installing CUDA toolkit for compilation..."
|
|
sudo apt install -y nvidia-cuda-toolkit -qq 2>/dev/null || true
|
|
fi
|
|
# Use latest GPU arch supported by this nvcc: Blackwell (RTX 50) = 120 (CUDA 12.8+), else Ada = 90.
|
|
if [ -n "${CMAKE_CUDA_ARCHITECTURES:-}" ]; then
|
|
CUDA_ARCH="$CMAKE_CUDA_ARCHITECTURES"
|
|
echo " Using CUDA architectures: $CUDA_ARCH (from CMAKE_CUDA_ARCHITECTURES)"
|
|
else
|
|
NVCC_VER=""
|
|
if command -v nvcc &>/dev/null; then
|
|
NVCC_VER=$(nvcc --version 2>/dev/null | sed -n 's/.*release \([0-9]*\)\.\([0-9]*\).*/\1.\2/p' | head -1)
|
|
fi
|
|
# Blackwell (sm_120) requires CUDA 12.8+
|
|
if [ -n "$NVCC_VER" ] && [ "$(printf '%s\n' "12.8" "$NVCC_VER" | sort -V | tail -1)" = "$NVCC_VER" ]; then
|
|
CUDA_ARCH="120"
|
|
echo " Using CUDA architecture 120 (Blackwell / RTX 50) - nvcc $NVCC_VER"
|
|
else
|
|
CUDA_ARCH="90"
|
|
echo " Using CUDA architecture 90 (Ada) - nvcc ${NVCC_VER:-unknown} (use CUDA 12.8+ for Blackwell 120)"
|
|
fi
|
|
fi
|
|
cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES="${CUDA_ARCH}" 2>/dev/null || cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES="${CUDA_ARCH}"
|
|
else
|
|
warn "No CUDA - building CPU-only whisper.cpp"
|
|
cmake -B build
|
|
fi
|
|
|
|
cmake --build build --config Release -j$(nproc)
|
|
sudo cp build/bin/whisper-cli /usr/local/bin/
|
|
ok "whisper-cli installed to /usr/local/bin/"
|
|
fi
|
|
|
|
# Download Whisper model
|
|
if [ -f "$WHISPER_MODEL_FILE" ]; then
|
|
SIZE=$(stat -c%s "$WHISPER_MODEL_FILE" 2>/dev/null || echo 0)
|
|
if [ "$SIZE" -gt 100000000 ]; then
|
|
ok "Whisper model already downloaded ($(echo "scale=0; $SIZE/1048576" | bc) MB)"
|
|
else
|
|
warn "Whisper model looks incomplete ($SIZE bytes). Re-downloading..."
|
|
rm -f "$WHISPER_MODEL_FILE"
|
|
fi
|
|
fi
|
|
|
|
if [ ! -f "$WHISPER_MODEL_FILE" ]; then
|
|
mkdir -p "$WHISPER_MODEL_DIR"
|
|
echo " Downloading ggml-large-v3-turbo.bin (~1.5 GB)..."
|
|
curl -L --progress-bar -o "$WHISPER_MODEL_FILE" \
|
|
"https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo.bin"
|
|
ok "Whisper model downloaded"
|
|
fi
|
|
|
|
# -- 6. TTS Setup ------------------------------------------------
|
|
step "Step 6/7: TTS Setup (Orpheus + Qwen3-TTS)"
|
|
|
|
cd "$LLMS_DIR"
|
|
|
|
# Ensure shell scripts have LF line endings (Windows clone may have CRLF)
|
|
for f in setup-tts.sh start-dashboard.sh download-tts-models.sh; do
|
|
[ -f "$f" ] && sed -i 's/\r$//' "$f" 2>/dev/null || true
|
|
done
|
|
|
|
# Use huggingface.co directly (personal machine, no corporate proxy)
|
|
echo " Running setup-tts.sh (this may take several minutes on first run)..."
|
|
HF_MIRROR=https://huggingface.co bash setup-tts.sh
|
|
|
|
ok "TTS setup complete"
|
|
|
|
# -- 7. Start Dashboard ------------------------------------------
|
|
step "Step 7/7: Mission Control Dashboard"
|
|
|
|
cd "$LLMS_DIR"
|
|
|
|
# Install npm deps if needed
|
|
if [ ! -d "$LLMS_DIR/dashboard/node_modules" ]; then
|
|
echo " Installing dashboard dependencies..."
|
|
(cd "$LLMS_DIR/dashboard" && npm install --silent)
|
|
ok "Dashboard dependencies installed"
|
|
fi
|
|
|
|
# Check if already running
|
|
if curl -s --max-time 2 http://localhost:3000 &>/dev/null; then
|
|
ok "Dashboard already running at http://localhost:3000"
|
|
else
|
|
echo " Starting dashboard..."
|
|
bash start-dashboard.sh
|
|
fi
|
|
|
|
# -- Summary -----------------------------------------------------
|
|
echo ""
|
|
echo -e " ${GREEN}=====================================${NC}"
|
|
echo -e " ${GREEN}WSL2 Setup Complete!${NC}"
|
|
echo -e " ${GREEN}=====================================${NC}"
|
|
echo ""
|
|
echo " Components installed:"
|
|
echo " Node.js: $(node --version)"
|
|
echo " Python: $($PYTHON_CMD --version 2>&1)"
|
|
if command -v nvidia-smi &>/dev/null; then
|
|
echo " GPU: $(nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null | head -1)"
|
|
fi
|
|
echo " Whisper: $(which whisper-cli 2>/dev/null || echo 'not found')"
|
|
echo " Repo: $LLMS_DIR"
|
|
echo ""
|
|
echo " Dashboard: http://localhost:3000"
|
|
echo " Ollama API: http://localhost:11434"
|
|
echo ""
|
|
echo " Useful commands:"
|
|
echo " bash start-dashboard.sh # start dashboard"
|
|
echo " bash start-dashboard.sh status # check status"
|
|
echo " bash start-dashboard.sh stop # stop dashboard"
|
|
echo ""
|
|
echo " Test TTS:"
|
|
echo " .venv-qwen-tts/bin/python test_orpheus_tts.py"
|
|
echo " .venv-qwen-tts/bin/python test_qwen_tts.py"
|
|
echo ""
|