| @echo off |
| |
| |
| |
| setlocal enabledelayedexpansion |
| cd /d "%~dp0\.." |
|
|
| echo. |
| echo ____ _ ____ _ _ |
| echo ^| _ \ __ _^(_^)___ _ _^/ ___^| ^|__ __ _^(_^)_ __ |
| echo ^| ^| ^| / _` ^| / __^| ^| ^| ^| ^| ^| '_ \ / _` ^| ^| '_ \ |
| echo ^| ^|_^| ^(_^| ^| \__ \ ^|_^| ^| ^|___^| ^| ^| ^| ^(_^| ^| ^| ^| ^| ^| |
| echo ^|____/ \__,_^|_^|___/\__, ^|\____^|_^| ^|_^|\__,_^|_^|_^| ^|_^| |
| echo ^|___/ Old Hardware Training Pipeline |
| echo. |
| echo Choose how to run DaisyChain on this machine: |
| echo [1] Docker (recommended on Windows - most reliable) |
| echo [2] Python (native; multi-node gloo is unstable on Windows - use WSL/Linux) |
| echo [3] Just install Python deps |
| echo [Q] Quit |
| echo. |
| set /p choice=" Your choice: " |
|
|
| if /i "%choice%"=="1" goto docker |
| if /i "%choice%"=="2" goto python |
| if /i "%choice%"=="3" goto deps |
| goto end |
|
|
| :docker |
| where docker >nul 2>nul |
| if errorlevel 1 ( |
| echo [!] Docker not found. Install Docker Desktop from https://docker.com and re-run. |
| goto end |
| ) |
| echo Building and starting the demo cluster ^(3 nodes + dashboard^)... |
| docker compose -f docker/docker-compose.yml up --build -d |
| echo. |
| echo [OK] Cluster starting. Opening the dashboard at http://localhost:8080 |
| timeout /t 4 >nul |
| start "" http://localhost:8080 |
| echo To stop: docker compose -f docker/docker-compose.yml down |
| goto end |
|
|
| :deps |
| where python >nul 2>nul |
| if errorlevel 1 ( echo [!] Python not found. Install Python 3.9+ first. & goto end ) |
| echo Installing dependencies... |
| python -m pip install --upgrade pip |
| python -m pip install torch numpy psutil |
| python -m pip install -e . |
| echo [OK] Installed. You can now run: daisychain-train |
| goto end |
|
|
| :python |
| call :deps |
| echo. |
| echo ---- Configure this node ---- |
| set /p master=" Coordinator IP (MASTER_ADDR, e.g. Tailscale 100.x): " |
| set /p world=" Total number of machines (WORLD_SIZE): " |
| set /p rank=" This machine's RANK (0 = coordinator): " |
| set /p iface=" Network interface (GLOO_SOCKET_IFNAME, e.g. tailscale0): " |
| set MASTER_ADDR=%master% |
| set MASTER_PORT=29560 |
| set WORLD_SIZE=%world% |
| set RANK=%rank% |
| set GLOO_SOCKET_IFNAME=%iface% |
| set USE_LIBUV=0 |
| echo. |
| echo [!] Note: native multi-node training over gloo is unstable on Windows. |
| echo If it hangs, use the Docker option or run the nodes on Linux/WSL. |
| echo Launching node RANK=%rank% ... |
| python -m daisychain.train |
| goto end |
|
|
| :end |
| echo. |
| pause |
| endlocal |
|
|
