merge: nightly into feature/data-reciper-enchansments
resolve setup.sh conflict by keeping nightly installer flow and preserving local data-designer plugin install via install_python_stack.py
This commit is contained in:
Shine1i
commit
acf7cce4a8
17 changed files with 1961 additions and 135 deletions
118
README.md
118
README.md
|
|
@ -30,28 +30,119 @@
|
|||
|
||||
## Quick Start
|
||||
|
||||
### One-command setup
|
||||
### Prerequisites
|
||||
|
||||
| Requirement | Linux / WSL | Windows |
|
||||
|---|---|---|
|
||||
| **GPU** | NVIDIA GPU with working driver | NVIDIA GPU with working driver |
|
||||
| **Python** | 3.11 – 3.13 | 3.11 – 3.13 |
|
||||
| **Git** | Pre-installed on most distros | Auto-installed by setup script (via `winget`) |
|
||||
| **CMake** | Pre-installed or `sudo apt install cmake` | Auto-installed by setup script (via `winget`) |
|
||||
| **C++ compiler** | `build-essential` (auto-detected) | Visual Studio Build Tools 2022 (auto-installed by setup script) |
|
||||
| **CUDA Toolkit** | Optional — setup auto-detects `nvcc` | Auto-installed by setup script (version matched to driver) |
|
||||
|
||||
> [!NOTE]
|
||||
> On **WSL**, the setup script will also run `sudo apt-get install build-essential cmake curl git libcurl4-openssl-dev` so that GGUF export works in non-interactive subprocesses. You may be prompted for your password during setup.
|
||||
|
||||
---
|
||||
|
||||
### Linux / Windows WSL
|
||||
|
||||
```bash
|
||||
# 1. Clone the repo
|
||||
git clone https://github.com/unslothai/unsloth-studio.git
|
||||
cd unsloth-studio
|
||||
|
||||
# 2. Run setup (installs Node, builds frontend, creates .venv, builds llama.cpp)
|
||||
bash setup.sh
|
||||
|
||||
# 3. Open a new terminal (or source your shell rc), then launch:
|
||||
unsloth-studio -H 0.0.0.0 -p 8000
|
||||
```
|
||||
|
||||
This script will:
|
||||
1. Install **Node.js ≥ 20** via nvm (if needed)
|
||||
2. Build the frontend to `studio/frontend/dist`
|
||||
3. Create a Python virtual environment and install all dependencies (including `unsloth`)
|
||||
4. Register a convenient `unsloth-ui` shell alias
|
||||
<details>
|
||||
<summary><b>What does <code>setup.sh</code> do?</b></summary>
|
||||
|
||||
### Launch the studio
|
||||
1. Installs **Node.js ≥ 20** via nvm (if needed)
|
||||
2. Runs `npm install && npm run build` for the React frontend
|
||||
3. Detects the best **Python 3.11 – 3.13** on your system and creates a `.venv`
|
||||
4. Installs all Python dependencies (unsloth, PyTorch with CUDA, triton kernels, etc.)
|
||||
5. On **WSL**: pre-installs build dependencies via `apt-get`
|
||||
6. Clones and builds **llama.cpp** at `~/.unsloth/llama.cpp` (GPU-accelerated if CUDA is found)
|
||||
7. Registers `unsloth-studio` and `unsloth-ui` shell aliases in your shell rc (bash, zsh, fish, or ksh)
|
||||
|
||||
</details>
|
||||
|
||||
---
|
||||
|
||||
### Windows (Native)
|
||||
|
||||
> [!IMPORTANT]
|
||||
> Requires an **NVIDIA GPU** — CPU-only machines are not supported on Windows.
|
||||
|
||||
```powershell
|
||||
# 1. Clone the repo
|
||||
git clone https://github.com/unslothai/unsloth-studio.git
|
||||
cd unsloth-studio
|
||||
|
||||
# 2. Run setup (Right-click → "Run with PowerShell", or from a terminal):
|
||||
.\setup.bat
|
||||
# Or directly:
|
||||
powershell -ExecutionPolicy Bypass -File setup.ps1
|
||||
```
|
||||
|
||||
After setup completes, **open a new terminal** and run:
|
||||
|
||||
```powershell
|
||||
# PowerShell
|
||||
unsloth-studio -H 0.0.0.0 -p 8000
|
||||
|
||||
# Or cmd.exe
|
||||
unsloth-studio -H 0.0.0.0 -p 8000
|
||||
```
|
||||
|
||||
<details>
|
||||
<summary><b>What does <code>setup.ps1</code> do?</b></summary>
|
||||
|
||||
1. Enables **Windows Long Paths** (required for deep dependency trees — prompts for UAC)
|
||||
2. Auto-installs missing system tools via `winget`: **Git**, **CMake**, **Visual Studio Build Tools 2022**, **CUDA Toolkit** (version-matched to your driver), **Node.js LTS**, **Python 3.12**, **OpenSSL dev**
|
||||
3. Builds the React frontend (`npm install && npm run build`)
|
||||
4. Creates a `.venv` and installs all Python dependencies (including CUDA-enabled PyTorch from the official index)
|
||||
5. Sets `TORCHINDUCTOR_CACHE_DIR=C:\tc` to avoid Windows MAX_PATH issues with Triton
|
||||
6. Clones and builds **llama.cpp** at `%USERPROFILE%\.unsloth\llama.cpp` with CUDA + Visual Studio
|
||||
7. Registers `unsloth-studio` and `unsloth-ui` commands in both PowerShell profile and `cmd.exe` (via batch files on PATH)
|
||||
|
||||
</details>
|
||||
|
||||
---
|
||||
|
||||
### Google Colab
|
||||
|
||||
The setup script auto-detects Colab and installs everything into the existing system Python (no venv):
|
||||
|
||||
```python
|
||||
!bash setup.sh
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Launching the Studio
|
||||
|
||||
After setup on any platform, the command is the same:
|
||||
|
||||
```bash
|
||||
# After setup, open a new terminal (or source ~/.bashrc), then inside your working directory:
|
||||
unsloth-ui -H 0.0.0.0 -p 8000
|
||||
unsloth-studio -H 0.0.0.0 -p 8000
|
||||
```
|
||||
|
||||
On **first launch**, a one-time setup token is printed to the console. Use it in the browser to create your admin account.
|
||||
| Flag | Description |
|
||||
|---|---|
|
||||
| `-H` / `--host` | Bind address (`0.0.0.0` for all interfaces, `127.0.0.1` for local only) |
|
||||
| `-p` / `--port` | Port number (default: `8000`) |
|
||||
|
||||
As this repo is in continuous development, please make sure to run the setup.sh file everytime you pull new changes from the repo.
|
||||
On **first launch**, a one-time setup token is printed to the console. Open the URL shown in your browser and use this token to create your admin account.
|
||||
|
||||
> [!TIP]
|
||||
> This repo is in active development. After pulling new changes, **always re-run the setup script** (`bash setup.sh` or `.\setup.bat`) to pick up dependency and build updates.
|
||||
|
||||
## API Reference
|
||||
|
||||
|
|
@ -105,7 +196,10 @@ new-ui-prototype/
|
|||
│ ├── export.py
|
||||
│ ├── ui.py
|
||||
│ └── studio.py
|
||||
├── setup.sh # One-command bootstrap script
|
||||
├── setup.sh # Bootstrap script (Linux / WSL / Colab)
|
||||
├── setup.ps1 # Bootstrap script (Windows native)
|
||||
├── setup.bat # Wrapper to launch setup.ps1 via double-click
|
||||
├── install_python_stack.py # Cross-platform Python dependency installer
|
||||
└── studio/
|
||||
├── backend/
|
||||
│ ├── main.py # FastAPI app & middleware
|
||||
|
|
|
|||
263
install_python_stack.py
Normal file
263
install_python_stack.py
Normal file
|
|
@ -0,0 +1,263 @@
|
|||
#!/usr/bin/env python3
|
||||
"""Cross-platform Python dependency installer for Unsloth Studio.
|
||||
|
||||
Called by both setup.sh (Linux / WSL) and setup.ps1 (Windows) after the
|
||||
virtual environment is already activated. Expects `pip` and `python` on
|
||||
PATH to point at the venv.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import tempfile
|
||||
import urllib.request
|
||||
from pathlib import Path
|
||||
|
||||
IS_WINDOWS = sys.platform == "win32"
|
||||
|
||||
# ── Paths ──────────────────────────────────────────────────────────────
|
||||
SCRIPT_DIR = Path(__file__).resolve().parent
|
||||
REQ_ROOT = SCRIPT_DIR / "studio" / "backend" / "requirements"
|
||||
SINGLE_ENV = REQ_ROOT / "single-env"
|
||||
CONSTRAINTS = SINGLE_ENV / "constraints.txt"
|
||||
LOCAL_DD_UNSTRUCTURED_PLUGIN = (
|
||||
SCRIPT_DIR / "studio" / "backend" / "plugins" / "data-designer-unstructured-seed"
|
||||
)
|
||||
|
||||
# ── Color support ──────────────────────────────────────────────────────
|
||||
|
||||
def _enable_colors() -> bool:
|
||||
"""Try to enable ANSI color support. Returns True if available."""
|
||||
if not hasattr(sys.stdout, "fileno"):
|
||||
return False
|
||||
try:
|
||||
if not os.isatty(sys.stdout.fileno()):
|
||||
return False
|
||||
except Exception:
|
||||
return False
|
||||
if IS_WINDOWS:
|
||||
try:
|
||||
import ctypes
|
||||
kernel32 = ctypes.windll.kernel32
|
||||
# Enable ENABLE_VIRTUAL_TERMINAL_PROCESSING (0x0004) on stdout
|
||||
handle = kernel32.GetStdHandle(-11) # STD_OUTPUT_HANDLE
|
||||
mode = ctypes.c_ulong()
|
||||
kernel32.GetConsoleMode(handle, ctypes.byref(mode))
|
||||
kernel32.SetConsoleMode(handle, mode.value | 0x0004)
|
||||
return True
|
||||
except Exception:
|
||||
return False
|
||||
return True # Unix terminals support ANSI by default
|
||||
|
||||
_HAS_COLOR = _enable_colors()
|
||||
|
||||
def _green(msg: str) -> str:
|
||||
return f"\033[92m{msg}\033[0m" if _HAS_COLOR else msg
|
||||
|
||||
def _cyan(msg: str) -> str:
|
||||
return f"\033[96m{msg}\033[0m" if _HAS_COLOR else msg
|
||||
|
||||
def _red(msg: str) -> str:
|
||||
return f"\033[91m{msg}\033[0m" if _HAS_COLOR else msg
|
||||
|
||||
|
||||
def run(label: str, cmd: list[str], *, quiet: bool = True) -> None:
|
||||
"""Run a command; on failure print output and exit."""
|
||||
print(_cyan(f" {label}..."))
|
||||
result = subprocess.run(
|
||||
cmd,
|
||||
stdout=subprocess.PIPE if quiet else None,
|
||||
stderr=subprocess.STDOUT if quiet else None,
|
||||
)
|
||||
if result.returncode != 0:
|
||||
print(_red(f"❌ {label} failed (exit code {result.returncode}):"))
|
||||
if result.stdout:
|
||||
print(result.stdout.decode(errors="replace"))
|
||||
sys.exit(result.returncode)
|
||||
|
||||
|
||||
# Packages to skip on Windows (require special build steps)
|
||||
WINDOWS_SKIP_PACKAGES = {"open_spiel"}
|
||||
|
||||
|
||||
def _filter_requirements(req: Path, skip: set[str]) -> Path:
|
||||
"""Return a temp copy of a requirements file with certain packages removed."""
|
||||
lines = req.read_text(encoding="utf-8").splitlines(keepends=True)
|
||||
filtered = [
|
||||
line for line in lines
|
||||
if not any(line.strip().lower().startswith(pkg) for pkg in skip)
|
||||
]
|
||||
tmp = tempfile.NamedTemporaryFile(
|
||||
mode="w", suffix=".txt", delete=False, encoding="utf-8",
|
||||
)
|
||||
tmp.writelines(filtered)
|
||||
tmp.close()
|
||||
return Path(tmp.name)
|
||||
|
||||
|
||||
def pip_install(
|
||||
label: str,
|
||||
*args: str,
|
||||
req: Path | None = None,
|
||||
constrain: bool = True,
|
||||
) -> None:
|
||||
"""Build and run a pip install command."""
|
||||
cmd = [sys.executable, "-m", "pip", "install"]
|
||||
cmd.extend(args)
|
||||
if constrain and CONSTRAINTS.is_file():
|
||||
cmd.extend(["-c", str(CONSTRAINTS)])
|
||||
actual_req = req
|
||||
if req is not None and IS_WINDOWS and WINDOWS_SKIP_PACKAGES:
|
||||
actual_req = _filter_requirements(req, WINDOWS_SKIP_PACKAGES)
|
||||
if actual_req is not None:
|
||||
cmd.extend(["-r", str(actual_req)])
|
||||
try:
|
||||
run(label, cmd)
|
||||
finally:
|
||||
# Clean up temp file if we created one
|
||||
if actual_req is not None and actual_req != req:
|
||||
actual_req.unlink(missing_ok=True)
|
||||
|
||||
|
||||
|
||||
def download_file(url: str, dest: Path) -> None:
|
||||
"""Download a file using urllib (no curl dependency)."""
|
||||
urllib.request.urlretrieve(url, dest)
|
||||
|
||||
|
||||
def patch_package_file(package_name: str, relative_path: str, url: str) -> None:
|
||||
"""Download a file from url and overwrite a file inside an installed package."""
|
||||
result = subprocess.run(
|
||||
[sys.executable, "-m", "pip", "show", package_name],
|
||||
capture_output=True, text=True,
|
||||
)
|
||||
if result.returncode != 0:
|
||||
print(_red(f" ⚠️ Could not find package {package_name}, skipping patch"))
|
||||
return
|
||||
|
||||
location = None
|
||||
for line in result.stdout.splitlines():
|
||||
if line.lower().startswith("location:"):
|
||||
location = line.split(":", 1)[1].strip()
|
||||
break
|
||||
|
||||
if not location:
|
||||
print(_red(f" ⚠️ Could not determine location of {package_name}"))
|
||||
return
|
||||
|
||||
dest = Path(location) / relative_path
|
||||
print(_cyan(f" Patching {dest.name} in {package_name}..."))
|
||||
download_file(url, dest)
|
||||
|
||||
|
||||
# ── Main install sequence ─────────────────────────────────────────────
|
||||
|
||||
def install_python_stack() -> int:
|
||||
print(_cyan("── Installing Python stack ──"))
|
||||
|
||||
# 1. Upgrade pip
|
||||
run("Upgrading pip", [sys.executable, "-m", "pip", "install", "--upgrade", "pip"])
|
||||
|
||||
# 2. Core packages: unsloth-zoo + unsloth
|
||||
pip_install(
|
||||
"Installing unsloth-zoo + unsloth",
|
||||
"--no-cache-dir",
|
||||
req=REQ_ROOT / "base.txt",
|
||||
)
|
||||
|
||||
# 3. Extra dependencies
|
||||
pip_install(
|
||||
"Installing additional unsloth dependencies",
|
||||
"--no-cache-dir",
|
||||
req=REQ_ROOT / "extras.txt",
|
||||
)
|
||||
|
||||
# 4. Overrides (torchao, transformers) — force-reinstall
|
||||
pip_install(
|
||||
"Installing torchao + transformers overrides",
|
||||
"--force-reinstall", "--no-cache-dir",
|
||||
req=REQ_ROOT / "overrides.txt",
|
||||
)
|
||||
|
||||
# 5. Triton kernels (no-deps, from source)
|
||||
pip_install(
|
||||
"Installing triton kernels",
|
||||
"--no-deps", "--no-cache-dir",
|
||||
req=REQ_ROOT / "triton-kernels.txt",
|
||||
constrain=False,
|
||||
)
|
||||
|
||||
# 6. Patch: override llama_cpp.py with fix from unsloth-zoo feature/llama-cpp-windows-support branch
|
||||
patch_package_file(
|
||||
"unsloth-zoo",
|
||||
os.path.join("unsloth_zoo", "llama_cpp.py"),
|
||||
"https://raw.githubusercontent.com/unslothai/unsloth-zoo/refs/heads/main/unsloth_zoo/llama_cpp.py",
|
||||
)
|
||||
|
||||
# 7a. Patch: override vision.py with fix from unsloth PR #4091
|
||||
patch_package_file(
|
||||
"unsloth",
|
||||
os.path.join("unsloth", "models", "vision.py"),
|
||||
"https://raw.githubusercontent.com/unslothai/unsloth/80e0108a684c882965a02a8ed851e3473c1145ab/unsloth/models/vision.py",
|
||||
)
|
||||
|
||||
# 7b. Patch : override save.py with fix from feature/llama-cpp-windows-support
|
||||
patch_package_file(
|
||||
"unsloth",
|
||||
os.path.join("unsloth", "save.py"),
|
||||
"https://raw.githubusercontent.com/unslothai/unsloth/refs/heads/main/unsloth/save.py",
|
||||
)
|
||||
|
||||
# 8. Studio dependencies
|
||||
pip_install(
|
||||
"Installing studio dependencies",
|
||||
"--no-cache-dir",
|
||||
req=REQ_ROOT / "studio.txt",
|
||||
)
|
||||
|
||||
# 9. Data-designer dependencies
|
||||
pip_install(
|
||||
"Installing data-designer dependencies",
|
||||
"--no-cache-dir",
|
||||
req=SINGLE_ENV / "data-designer-deps.txt",
|
||||
)
|
||||
|
||||
# 10. Data-designer packages (no-deps to avoid conflicts)
|
||||
pip_install(
|
||||
"Installing data-designer",
|
||||
"--no-cache-dir", "--no-deps",
|
||||
req=SINGLE_ENV / "data-designer.txt",
|
||||
)
|
||||
|
||||
# 11. Local Data Designer seed plugin
|
||||
if not LOCAL_DD_UNSTRUCTURED_PLUGIN.is_dir():
|
||||
print(
|
||||
_red(
|
||||
f"❌ Missing local plugin directory: {LOCAL_DD_UNSTRUCTURED_PLUGIN}",
|
||||
),
|
||||
)
|
||||
return 1
|
||||
pip_install(
|
||||
"Installing local data-designer unstructured plugin",
|
||||
"--no-cache-dir", "--no-deps",
|
||||
"-e", str(LOCAL_DD_UNSTRUCTURED_PLUGIN),
|
||||
constrain=False,
|
||||
)
|
||||
|
||||
# 12. Patch metadata for single-env compatibility
|
||||
run(
|
||||
"Patching single-env metadata",
|
||||
[sys.executable, str(SINGLE_ENV / "patch_metadata.py")],
|
||||
)
|
||||
|
||||
# 13. Final check
|
||||
run("Running pip check", [sys.executable, "-m", "pip", "check"], quiet=False)
|
||||
|
||||
print(_green("✅ Python dependencies installed"))
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(install_python_stack())
|
||||
2
setup.bat
Normal file
2
setup.bat
Normal file
|
|
@ -0,0 +1,2 @@
|
|||
@echo off
|
||||
powershell -ExecutionPolicy Bypass -File "%~dp0setup.ps1" %*
|
||||
964
setup.ps1
Normal file
964
setup.ps1
Normal file
|
|
@ -0,0 +1,964 @@
|
|||
#Requires -Version 5.1
|
||||
<#
|
||||
.SYNOPSIS
|
||||
Full environment setup for Unsloth Studio on Windows (bundled version).
|
||||
.DESCRIPTION
|
||||
Always installs Node.js if needed. When running from pip install:
|
||||
skips frontend build (already bundled). When running from git repo:
|
||||
full setup including frontend build.
|
||||
Requires an NVIDIA GPU -- CPU-only machines are not supported.
|
||||
.NOTES
|
||||
Usage: powershell -ExecutionPolicy Bypass -File setup.ps1
|
||||
#>
|
||||
|
||||
$ErrorActionPreference = "Stop"
|
||||
$ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path
|
||||
$PackageDir = Split-Path -Parent $ScriptDir
|
||||
|
||||
# Detect if running from pip install (no studio/frontend/ dir in repo)
|
||||
$FrontendDir = Join-Path $ScriptDir "studio\frontend"
|
||||
$IsPipInstall = -not (Test-Path $FrontendDir)
|
||||
|
||||
# ─────────────────────────────────────────────
|
||||
# Helper functions
|
||||
# ─────────────────────────────────────────────
|
||||
|
||||
# Reload ALL environment variables from registry.
|
||||
# Picks up changes made by installers (winget, msi, etc.) including
|
||||
# Path, CUDA_PATH, CUDA_PATH_V*, and any other vars they set.
|
||||
function Refresh-Environment {
|
||||
foreach ($level in @('Machine', 'User')) {
|
||||
$vars = [System.Environment]::GetEnvironmentVariables($level)
|
||||
foreach ($key in $vars.Keys) {
|
||||
if ($key -eq 'Path') { continue }
|
||||
Set-Item -Path "Env:$key" -Value $vars[$key] -ErrorAction SilentlyContinue
|
||||
}
|
||||
}
|
||||
$machinePath = [System.Environment]::GetEnvironmentVariable('Path', 'Machine')
|
||||
$userPath = [System.Environment]::GetEnvironmentVariable('Path', 'User')
|
||||
$env:Path = "$machinePath;$userPath"
|
||||
}
|
||||
|
||||
# Find nvcc on PATH, CUDA_PATH, or standard toolkit dirs.
|
||||
# Returns the path to nvcc.exe, or $null if not found.
|
||||
function Find-Nvcc {
|
||||
param([string]$MaxVersion = "")
|
||||
|
||||
# If MaxVersion is set, we need to find a toolkit <= that version.
|
||||
# CUDA toolkits install side-by-side under C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vX.Y\
|
||||
|
||||
$toolkitBase = 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA'
|
||||
|
||||
if ($MaxVersion -and (Test-Path $toolkitBase)) {
|
||||
$drMajor = [int]$MaxVersion.Split('.')[0]
|
||||
$drMinor = [int]$MaxVersion.Split('.')[1]
|
||||
|
||||
# Get all installed CUDA dirs, sorted descending (highest first)
|
||||
$cudaDirs = Get-ChildItem -Directory $toolkitBase | Where-Object {
|
||||
$_.Name -match '^v(\d+)\.(\d+)'
|
||||
} | Sort-Object { [version]($_.Name -replace '^v','') } -Descending
|
||||
|
||||
foreach ($dir in $cudaDirs) {
|
||||
if ($dir.Name -match '^v(\d+)\.(\d+)') {
|
||||
$tkMajor = [int]$Matches[1]; $tkMinor = [int]$Matches[2]
|
||||
$compatible = ($tkMajor -lt $drMajor) -or ($tkMajor -eq $drMajor -and $tkMinor -le $drMinor)
|
||||
if ($compatible) {
|
||||
$nvcc = Join-Path $dir.FullName 'bin\nvcc.exe'
|
||||
if (Test-Path $nvcc) {
|
||||
return $nvcc
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
# No compatible side-by-side version found
|
||||
return $null
|
||||
}
|
||||
|
||||
# Fallback: no version constraint — pick latest or whatever is available
|
||||
|
||||
# 1. Check nvcc on PATH
|
||||
$cmd = Get-Command nvcc -ErrorAction SilentlyContinue
|
||||
if ($cmd) { return $cmd.Source }
|
||||
|
||||
# 2. Check CUDA_PATH env var
|
||||
$cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'Process')
|
||||
if (-not $cudaRoot) { $cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'Machine') }
|
||||
if (-not $cudaRoot) { $cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'User') }
|
||||
if ($cudaRoot -and (Test-Path (Join-Path $cudaRoot 'bin\nvcc.exe'))) {
|
||||
return (Join-Path $cudaRoot 'bin\nvcc.exe')
|
||||
}
|
||||
|
||||
# 3. Scan standard toolkit directory
|
||||
if (Test-Path $toolkitBase) {
|
||||
$latest = Get-ChildItem -Directory $toolkitBase | Sort-Object Name | Select-Object -Last 1
|
||||
if ($latest -and (Test-Path (Join-Path $latest.FullName 'bin\nvcc.exe'))) {
|
||||
return (Join-Path $latest.FullName 'bin\nvcc.exe')
|
||||
}
|
||||
}
|
||||
|
||||
return $null
|
||||
}
|
||||
|
||||
# Detect CUDA Compute Capability via nvidia-smi.
|
||||
# Returns e.g. "80" for A100 (8.0), "89" for RTX 4090 (8.9), etc.
|
||||
# Returns $null if detection fails.
|
||||
function Get-CudaComputeCapability {
|
||||
$nvSmi = Get-Command nvidia-smi -ErrorAction SilentlyContinue
|
||||
if (-not $nvSmi) { return $null }
|
||||
|
||||
try {
|
||||
$raw = & nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>$null
|
||||
if ($LASTEXITCODE -ne 0 -or -not $raw) { return $null }
|
||||
|
||||
# nvidia-smi may return multiple GPUs; take the first one
|
||||
$cap = ($raw -split "`n")[0].Trim()
|
||||
if ($cap -match '^(\d+)\.(\d+)$') {
|
||||
$major = $Matches[1]
|
||||
$minor = $Matches[2]
|
||||
return "$major$minor"
|
||||
}
|
||||
} catch { }
|
||||
|
||||
return $null
|
||||
}
|
||||
|
||||
# Detect driver's max CUDA version from nvidia-smi and return the highest
|
||||
# compatible PyTorch CUDA index tag (e.g. "cu128").
|
||||
# PyTorch on Windows ships CPU-only by default from PyPI; CUDA wheels live at
|
||||
# https://download.pytorch.org/whl/<tag>. The tag must not exceed the driver's
|
||||
# capability: e.g. driver "CUDA Version: 12.9" → cu128 (not cu130).
|
||||
function Get-PytorchCudaTag {
|
||||
$nvSmi = Get-Command nvidia-smi -ErrorAction SilentlyContinue
|
||||
if (-not $nvSmi) { return "cu124" }
|
||||
|
||||
try {
|
||||
# 2>&1 | Out-String merges stderr into stdout then converts to a single
|
||||
# string. Plain 2>$null doesn't fully suppress stderr in PS 5.1 —
|
||||
# ErrorRecord objects leak into $output and break the -match.
|
||||
$output = & nvidia-smi 2>&1 | Out-String
|
||||
if ($output -match 'CUDA Version:\s+(\d+)\.(\d+)') {
|
||||
$major = [int]$Matches[1]
|
||||
$minor = [int]$Matches[2]
|
||||
# PyTorch 2.10 offers: cu124, cu126, cu128, cu130
|
||||
if ($major -ge 13) { return "cu130" }
|
||||
if ($major -eq 12 -and $minor -ge 8) { return "cu128" }
|
||||
if ($major -eq 12 -and $minor -ge 6) { return "cu126" }
|
||||
return "cu124"
|
||||
}
|
||||
} catch { }
|
||||
|
||||
return "cu124"
|
||||
}
|
||||
|
||||
# Find Visual Studio Build Tools for cmake -G flag.
|
||||
# Strategy: (1) vswhere, (2) scan filesystem (handles broken vswhere registration).
|
||||
# Returns @{ Generator = "Visual Studio 17 2022"; InstallPath = "C:\..."; Source = "..." } or $null.
|
||||
function Find-VsBuildTools {
|
||||
$map = @{ '2022' = '17'; '2019' = '16'; '2017' = '15' }
|
||||
|
||||
# --- Try vswhere first (works when VS is properly registered) ---
|
||||
$vsw = "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe"
|
||||
if (Test-Path $vsw) {
|
||||
$info = & $vsw -latest -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property catalog_productLineVersion 2>$null
|
||||
$path = & $vsw -latest -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath 2>$null
|
||||
if ($info -and $path) {
|
||||
$y = $info.Trim()
|
||||
$n = $map[$y]
|
||||
if ($n) {
|
||||
return @{ Generator = "Visual Studio $n $y"; InstallPath = $path.Trim(); Source = 'vswhere' }
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
# --- Scan filesystem (handles broken vswhere registration after winget cycles) ---
|
||||
$roots = @($env:ProgramFiles, ${env:ProgramFiles(x86)})
|
||||
$editions = @('BuildTools', 'Community', 'Professional', 'Enterprise')
|
||||
$years = @('2022', '2019', '2017')
|
||||
|
||||
foreach ($y in $years) {
|
||||
foreach ($r in $roots) {
|
||||
foreach ($ed in $editions) {
|
||||
$candidate = Join-Path $r "Microsoft Visual Studio\$y\$ed"
|
||||
if (Test-Path $candidate) {
|
||||
$vcDir = Join-Path $candidate "VC\Tools\MSVC"
|
||||
if (Test-Path $vcDir) {
|
||||
$cl = Get-ChildItem -Path $vcDir -Filter "cl.exe" -Recurse -ErrorAction SilentlyContinue | Select-Object -First 1
|
||||
if ($cl) {
|
||||
$n = $map[$y]
|
||||
if ($n) {
|
||||
return @{ Generator = "Visual Studio $n $y"; InstallPath = $candidate; Source = "filesystem ($ed)"; ClExe = $cl.FullName }
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return $null
|
||||
}
|
||||
|
||||
# ─────────────────────────────────────────────
|
||||
# Banner
|
||||
# ─────────────────────────────────────────────
|
||||
Write-Host "+==============================================+" -ForegroundColor Green
|
||||
Write-Host "| Unsloth Studio Setup (Windows) |" -ForegroundColor Green
|
||||
Write-Host "+==============================================+" -ForegroundColor Green
|
||||
|
||||
# ==========================================================================
|
||||
# PHASE 1: System-level prerequisites (winget installs, env vars)
|
||||
# All heavy system tool installs happen here BEFORE touching Python.
|
||||
# ==========================================================================
|
||||
|
||||
# ============================================
|
||||
# 1a. GPU requirement check
|
||||
# ============================================
|
||||
$HasNvidiaSmi = $null -ne (Get-Command nvidia-smi -ErrorAction SilentlyContinue)
|
||||
if (-not $HasNvidiaSmi) {
|
||||
Write-Host ""
|
||||
Write-Host "[ERROR] Unsloth Studio requires an NVIDIA GPU." -ForegroundColor Red
|
||||
Write-Host " CPU-only machines are not supported." -ForegroundColor Red
|
||||
Write-Host ""
|
||||
Write-Host " If you have an NVIDIA GPU, ensure the driver is installed:" -ForegroundColor Yellow
|
||||
Write-Host " https://www.nvidia.com/Download/index.aspx" -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
Write-Host "[OK] NVIDIA GPU detected" -ForegroundColor Green
|
||||
|
||||
# ============================================
|
||||
# 1a.5. Windows Long Paths (required for deep node_modules / Python paths)
|
||||
# ============================================
|
||||
$LongPathsEnabled = $false
|
||||
try {
|
||||
$regVal = Get-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -ErrorAction SilentlyContinue
|
||||
if ($regVal -and $regVal.LongPathsEnabled -eq 1) {
|
||||
$LongPathsEnabled = $true
|
||||
}
|
||||
} catch {}
|
||||
|
||||
if ($LongPathsEnabled) {
|
||||
Write-Host "[OK] Windows Long Paths enabled" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host "Windows Long Paths not enabled (required for Triton compilation and deep dependency paths)." -ForegroundColor Yellow
|
||||
Write-Host " Requesting admin access to fix..." -ForegroundColor Yellow
|
||||
try {
|
||||
# Spawn an elevated process to set the registry key (triggers UAC prompt)
|
||||
$proc = Start-Process -FilePath "reg.exe" `
|
||||
-ArgumentList 'add "HKLM\SYSTEM\CurrentControlSet\Control\FileSystem" /v LongPathsEnabled /t REG_DWORD /d 1 /f' `
|
||||
-Verb RunAs -Wait -PassThru -ErrorAction Stop
|
||||
if ($proc.ExitCode -eq 0) {
|
||||
$LongPathsEnabled = $true
|
||||
Write-Host "[OK] Windows Long Paths enabled (via UAC)" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host "[WARN] Failed to enable Long Paths (exit code: $($proc.ExitCode))" -ForegroundColor Yellow
|
||||
}
|
||||
} catch {
|
||||
Write-Host "[WARN] Could not enable Long Paths (UAC was declined or not available)" -ForegroundColor Yellow
|
||||
Write-Host " Run this manually in an Admin terminal:" -ForegroundColor Yellow
|
||||
Write-Host ' reg add "HKLM\SYSTEM\CurrentControlSet\Control\FileSystem" /v LongPathsEnabled /t REG_DWORD /d 1 /f' -ForegroundColor Cyan
|
||||
}
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# 1b. Git (required by pip for git+https:// deps and by npm)
|
||||
# ============================================
|
||||
$HasGit = $null -ne (Get-Command git -ErrorAction SilentlyContinue)
|
||||
if (-not $HasGit) {
|
||||
Write-Host "Git not found -- installing via winget..." -ForegroundColor Yellow
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
try {
|
||||
winget install Git.Git --source winget --accept-package-agreements --accept-source-agreements 2>&1 | Out-Null
|
||||
Refresh-Environment
|
||||
$HasGit = $null -ne (Get-Command git -ErrorAction SilentlyContinue)
|
||||
} catch { }
|
||||
}
|
||||
if (-not $HasGit) {
|
||||
Write-Host "[ERROR] Git is required but could not be installed automatically." -ForegroundColor Red
|
||||
Write-Host " Install Git from https://git-scm.com/download/win and re-run." -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
Write-Host "[OK] Git installed: $(git --version)" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host "[OK] Git found: $(git --version)" -ForegroundColor Green
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# 1c. CMake (required for llama.cpp build)
|
||||
# ============================================
|
||||
$HasCmake = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
|
||||
if (-not $HasCmake) {
|
||||
Write-Host "CMake not found -- installing via winget..." -ForegroundColor Yellow
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
try {
|
||||
winget install Kitware.CMake --source winget --accept-package-agreements --accept-source-agreements 2>&1 | Out-Null
|
||||
Refresh-Environment
|
||||
$HasCmake = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
|
||||
} catch { }
|
||||
}
|
||||
if ($HasCmake) {
|
||||
Write-Host "[OK] CMake installed" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host "[ERROR] CMake is required but could not be installed." -ForegroundColor Red
|
||||
Write-Host " Install CMake from https://cmake.org/download/ and re-run." -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
} else {
|
||||
Write-Host "[OK] CMake found: $(cmake --version | Select-Object -First 1)" -ForegroundColor Green
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# 1d. Visual Studio Build Tools (C++ compiler for llama.cpp)
|
||||
# ============================================
|
||||
$CmakeGenerator = $null
|
||||
$VsInstallPath = $null
|
||||
$vsResult = Find-VsBuildTools
|
||||
|
||||
if (-not $vsResult) {
|
||||
Write-Host "Visual Studio Build Tools not found -- installing via winget..." -ForegroundColor Yellow
|
||||
Write-Host " (This is a one-time install, may take several minutes)" -ForegroundColor Gray
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
$prevEAPTemp = $ErrorActionPreference
|
||||
$ErrorActionPreference = "Continue"
|
||||
winget install Microsoft.VisualStudio.2022.BuildTools --source winget --accept-package-agreements --accept-source-agreements --override "--add Microsoft.VisualStudio.Workload.VCTools --includeRecommended --passive --wait"
|
||||
$ErrorActionPreference = $prevEAPTemp
|
||||
# Re-scan after install (don't trust vswhere catalog)
|
||||
$vsResult = Find-VsBuildTools
|
||||
}
|
||||
}
|
||||
|
||||
if ($vsResult) {
|
||||
$CmakeGenerator = $vsResult.Generator
|
||||
$VsInstallPath = $vsResult.InstallPath
|
||||
Write-Host "[OK] $CmakeGenerator detected via $($vsResult.Source)" -ForegroundColor Green
|
||||
if ($vsResult.ClExe) { Write-Host " cl.exe: $($vsResult.ClExe)" -ForegroundColor Gray }
|
||||
} else {
|
||||
Write-Host "[ERROR] Visual Studio Build Tools could not be found or installed." -ForegroundColor Red
|
||||
Write-Host " Manual install:" -ForegroundColor Red
|
||||
Write-Host ' 1. winget install Microsoft.VisualStudio.2022.BuildTools --source winget' -ForegroundColor Yellow
|
||||
Write-Host ' 2. Open Visual Studio Installer -> Modify -> check "Desktop development with C++"' -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# 1e. CUDA Toolkit (nvcc for llama.cpp build + env vars)
|
||||
# ============================================
|
||||
# IMPORTANT: The CUDA Toolkit version must be <= the max CUDA version the
|
||||
# NVIDIA driver supports. nvidia-smi reports this as "CUDA Version: X.Y".
|
||||
# If we install a toolkit newer than the driver supports, llama-server will
|
||||
# fail at runtime with "ggml_cuda_init: failed to initialize CUDA: (null)".
|
||||
|
||||
# -- Detect max CUDA version the driver supports --
|
||||
$DriverMaxCuda = $null
|
||||
try {
|
||||
$smiOut = nvidia-smi 2>&1 | Out-String
|
||||
if ($smiOut -match "CUDA Version:\s+([\d]+)\.([\d]+)") {
|
||||
$DriverMaxCuda = "$($Matches[1]).$($Matches[2])"
|
||||
Write-Host " Driver supports up to CUDA $DriverMaxCuda" -ForegroundColor Gray
|
||||
}
|
||||
} catch {}
|
||||
|
||||
# -- Find a toolkit that's compatible with the driver --
|
||||
$IncompatibleToolkit = $null
|
||||
if ($DriverMaxCuda) {
|
||||
$NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
|
||||
if ($NvccPath) {
|
||||
Write-Host " [OK] Found compatible CUDA Toolkit (nvcc: $NvccPath)" -ForegroundColor Green
|
||||
} else {
|
||||
# Check if there's an incompatible (too new) toolkit installed
|
||||
$AnyNvcc = Find-Nvcc
|
||||
if ($AnyNvcc) {
|
||||
$NvccOut = & $AnyNvcc --version 2>&1 | Out-String
|
||||
if ($NvccOut -match "release\s+([\d]+\.[\d]+)") {
|
||||
$IncompatibleToolkit = $Matches[1]
|
||||
}
|
||||
}
|
||||
}
|
||||
} else {
|
||||
$NvccPath = Find-Nvcc
|
||||
}
|
||||
|
||||
# -- If incompatible toolkit is blocking, tell user to uninstall it --
|
||||
if (-not $NvccPath -and $IncompatibleToolkit) {
|
||||
Write-Host "" -ForegroundColor Red
|
||||
Write-Host "========================================================================" -ForegroundColor Red
|
||||
Write-Host "[ERROR] CUDA Toolkit $IncompatibleToolkit is installed but INCOMPATIBLE" -ForegroundColor Red
|
||||
Write-Host " with your NVIDIA driver (which supports up to CUDA $DriverMaxCuda)." -ForegroundColor Red
|
||||
Write-Host "" -ForegroundColor Red
|
||||
Write-Host " This will cause 'failed to initialize CUDA' errors at runtime." -ForegroundColor Red
|
||||
Write-Host "" -ForegroundColor Red
|
||||
Write-Host " To fix:" -ForegroundColor Yellow
|
||||
Write-Host " 1. Open Control Panel -> Programs -> Uninstall a program" -ForegroundColor Yellow
|
||||
Write-Host " 2. Uninstall 'NVIDIA CUDA Toolkit $IncompatibleToolkit'" -ForegroundColor Yellow
|
||||
Write-Host " 3. Re-run setup.bat (it will install CUDA $DriverMaxCuda automatically)" -ForegroundColor Yellow
|
||||
Write-Host "" -ForegroundColor Yellow
|
||||
Write-Host " Alternatively, update your NVIDIA driver to one that supports CUDA $IncompatibleToolkit." -ForegroundColor Gray
|
||||
Write-Host "========================================================================" -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
|
||||
# -- No toolkit at all: install via winget --
|
||||
if (-not $NvccPath) {
|
||||
Write-Host "CUDA toolkit (nvcc) not found -- installing via winget..." -ForegroundColor Yellow
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
if ($DriverMaxCuda) {
|
||||
# Try descending compatible versions
|
||||
$drMajor = [int]$DriverMaxCuda.Split('.')[0]
|
||||
$drMinor = [int]$DriverMaxCuda.Split('.')[1]
|
||||
for ($m = $drMinor; $m -ge 0; $m--) {
|
||||
$ver = "$drMajor.$m"
|
||||
Write-Host " Trying CUDA Toolkit $ver via winget..." -ForegroundColor Cyan
|
||||
$prevEAPCuda = $ErrorActionPreference
|
||||
$ErrorActionPreference = "Continue"
|
||||
winget install --id=Nvidia.CUDA --version=$ver -e --source winget --accept-package-agreements --accept-source-agreements 2>&1 | Out-Null
|
||||
$ErrorActionPreference = $prevEAPCuda
|
||||
Refresh-Environment
|
||||
$NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
|
||||
if ($NvccPath) {
|
||||
Write-Host " [OK] CUDA Toolkit $ver installed (nvcc: $NvccPath)" -ForegroundColor Green
|
||||
break
|
||||
}
|
||||
}
|
||||
} else {
|
||||
Write-Host " Installing CUDA Toolkit (latest) via winget..." -ForegroundColor Cyan
|
||||
winget install --id=Nvidia.CUDA -e --source winget --accept-package-agreements --accept-source-agreements
|
||||
Refresh-Environment
|
||||
$NvccPath = Find-Nvcc
|
||||
if ($NvccPath) {
|
||||
Write-Host " [OK] CUDA Toolkit installed (nvcc: $NvccPath)" -ForegroundColor Green
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (-not $NvccPath) {
|
||||
Write-Host "[ERROR] CUDA Toolkit (nvcc) is required but could not be found or installed." -ForegroundColor Red
|
||||
if ($DriverMaxCuda) {
|
||||
Write-Host " Install CUDA Toolkit $DriverMaxCuda from https://developer.nvidia.com/cuda-toolkit-archive" -ForegroundColor Yellow
|
||||
} else {
|
||||
Write-Host " Install CUDA Toolkit from https://developer.nvidia.com/cuda-downloads" -ForegroundColor Yellow
|
||||
}
|
||||
exit 1
|
||||
}
|
||||
|
||||
# -- Set CUDA env vars so cmake AND MSBuild can find the toolkit --
|
||||
$CudaToolkitRoot = Split-Path (Split-Path $NvccPath -Parent) -Parent
|
||||
# CUDA_PATH: used by cmake's find_package(CUDAToolkit)
|
||||
[Environment]::SetEnvironmentVariable('CUDA_PATH', $CudaToolkitRoot, 'Process')
|
||||
# CudaToolkitDir: the MSBuild property that CUDA .targets checks directly
|
||||
# Trailing backslash required -- the .targets file appends subpaths to it
|
||||
[Environment]::SetEnvironmentVariable('CudaToolkitDir', "$CudaToolkitRoot\", 'Process')
|
||||
# Always persist CUDA_PATH to User registry so the compatible toolkit is used
|
||||
# in future sessions (overwrites any existing value pointing to a newer, incompatible version)
|
||||
[Environment]::SetEnvironmentVariable('CUDA_PATH', $CudaToolkitRoot, 'User')
|
||||
Write-Host " Persisted CUDA_PATH=$CudaToolkitRoot to user environment" -ForegroundColor Gray
|
||||
# Ensure nvcc's bin dir is on PATH for this process
|
||||
$nvccBinDir = Split-Path $NvccPath -Parent
|
||||
if ($env:PATH -notlike "*$nvccBinDir*") {
|
||||
[Environment]::SetEnvironmentVariable('PATH', "$nvccBinDir;$env:PATH", 'Process')
|
||||
}
|
||||
# Persist nvcc bin dir to User PATH so it works in new terminals
|
||||
$userPath = [Environment]::GetEnvironmentVariable('Path', 'User')
|
||||
if (-not $userPath -or $userPath -notlike "*$nvccBinDir*") {
|
||||
if ($userPath) {
|
||||
[Environment]::SetEnvironmentVariable('Path', "$nvccBinDir;$userPath", 'User')
|
||||
} else {
|
||||
[Environment]::SetEnvironmentVariable('Path', "$nvccBinDir", 'User')
|
||||
}
|
||||
Write-Host " Persisted CUDA bin dir to user PATH" -ForegroundColor Gray
|
||||
}
|
||||
|
||||
Write-Host "[OK] CUDA Toolkit: $NvccPath" -ForegroundColor Green
|
||||
Write-Host " CUDA_PATH = $CudaToolkitRoot" -ForegroundColor Gray
|
||||
Write-Host " CudaToolkitDir = $CudaToolkitRoot\" -ForegroundColor Gray
|
||||
|
||||
# Detect compute capability (used later for llama.cpp cmake)
|
||||
$CudaArch = Get-CudaComputeCapability
|
||||
if ($CudaArch) {
|
||||
Write-Host " Compute Capability = $($CudaArch.Insert($CudaArch.Length-1, '.')) (sm_$CudaArch)" -ForegroundColor Gray
|
||||
} else {
|
||||
Write-Host " [WARN] Could not detect compute capability -- cmake will use defaults" -ForegroundColor Yellow
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# 1f. Node.js / npm (always -- needed regardless of install method)
|
||||
# ============================================
|
||||
# setup.sh installs Node LTS (v22) via nvm. We enforce the same range here:
|
||||
# Node >= 20, npm >= 11.
|
||||
$NeedNode = $true
|
||||
try {
|
||||
$NodeVersion = (node -v 2>$null)
|
||||
$NpmVersion = (npm -v 2>$null)
|
||||
if ($NodeVersion -and $NpmVersion) {
|
||||
$NodeMajor = [int]($NodeVersion -replace 'v','').Split('.')[0]
|
||||
$NpmMajor = [int]$NpmVersion.Split('.')[0]
|
||||
|
||||
if ($NodeMajor -ge 20 -and $NpmMajor -ge 11) {
|
||||
Write-Host "[OK] Node $NodeVersion and npm $NpmVersion already meet requirements." -ForegroundColor Green
|
||||
$NeedNode = $false
|
||||
} else {
|
||||
Write-Host "[WARN] Node $NodeVersion / npm $NpmVersion too old." -ForegroundColor Yellow
|
||||
}
|
||||
}
|
||||
} catch {
|
||||
Write-Host "[WARN] Node/npm not found." -ForegroundColor Yellow
|
||||
}
|
||||
|
||||
if ($NeedNode) {
|
||||
Write-Host "Installing Node.js LTS via winget..." -ForegroundColor Cyan
|
||||
try {
|
||||
winget install OpenJS.NodeJS.LTS --source winget --accept-package-agreements --accept-source-agreements
|
||||
Refresh-Environment
|
||||
} catch {
|
||||
Write-Host "[ERROR] Could not install Node.js automatically." -ForegroundColor Red
|
||||
Write-Host "Please install Node.js >= 20 from https://nodejs.org/" -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
}
|
||||
|
||||
Write-Host "[OK] Node $(node -v) | npm $(npm -v)" -ForegroundColor Green
|
||||
|
||||
# ============================================
|
||||
# 1g. Python (>= 3.11 and < 3.14, matching setup.sh)
|
||||
# ============================================
|
||||
$HasPython = $null -ne (Get-Command python -ErrorAction SilentlyContinue)
|
||||
$PythonOk = $false
|
||||
|
||||
if ($HasPython) {
|
||||
$PyVer = python --version 2>&1
|
||||
if ($PyVer -match "(\d+)\.(\d+)") {
|
||||
$PyMajor = [int]$Matches[1]; $PyMinor = [int]$Matches[2]
|
||||
if ($PyMajor -eq 3 -and $PyMinor -ge 11 -and $PyMinor -lt 14) {
|
||||
Write-Host "[OK] Python $PyVer" -ForegroundColor Green
|
||||
$PythonOk = $true
|
||||
} else {
|
||||
Write-Host "[ERROR] Python $PyVer is outside supported range (need >= 3.11 and < 3.14)." -ForegroundColor Red
|
||||
Write-Host " Install Python 3.12 from https://python.org/downloads/" -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
}
|
||||
} else {
|
||||
# No Python at all -- install 3.12
|
||||
Write-Host "Python not found -- installing Python 3.12 via winget..." -ForegroundColor Yellow
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
winget install -e --id Python.Python.3.12 --source winget --accept-package-agreements --accept-source-agreements
|
||||
Refresh-Environment
|
||||
}
|
||||
$HasPython = $null -ne (Get-Command python -ErrorAction SilentlyContinue)
|
||||
if (-not $HasPython) {
|
||||
Write-Host "[ERROR] Python could not be installed automatically." -ForegroundColor Red
|
||||
Write-Host " Install Python 3.12 from https://python.org/downloads/" -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
Write-Host "[OK] Python $(python --version)" -ForegroundColor Green
|
||||
$PythonOk = $true
|
||||
}
|
||||
|
||||
Write-Host ""
|
||||
Write-Host "--- System prerequisites ready ---" -ForegroundColor Green
|
||||
Write-Host ""
|
||||
|
||||
# ==========================================================================
|
||||
# PHASE 2: Frontend build (skip if pip-installed -- already bundled)
|
||||
# ==========================================================================
|
||||
if ($IsPipInstall) {
|
||||
Write-Host "[OK] Running from pip install - frontend already bundled, skipping build" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host ""
|
||||
Write-Host "Building frontend..." -ForegroundColor Cyan
|
||||
# npm writes warnings to stderr; lower ErrorActionPreference so PS doesn't
|
||||
# treat them as terminating errors (same pattern as the pip section below).
|
||||
$prevEAP_npm = $ErrorActionPreference
|
||||
$ErrorActionPreference = "Continue"
|
||||
Push-Location $FrontendDir
|
||||
# Remove stale node_modules and package-lock.json to avoid version conflicts
|
||||
if (Test-Path "node_modules") { Remove-Item -Recurse -Force "node_modules" }
|
||||
if (Test-Path "package-lock.json") { Remove-Item -Force "package-lock.json" }
|
||||
npm install 2>&1 | Out-Null
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
Pop-Location
|
||||
$ErrorActionPreference = $prevEAP_npm
|
||||
Write-Host "[ERROR] npm install failed (exit code $LASTEXITCODE)" -ForegroundColor Red
|
||||
Write-Host " Try running 'npm install' manually in studio/frontend/ to see errors" -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
npm run build 2>&1 | Out-Null
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
Pop-Location
|
||||
$ErrorActionPreference = $prevEAP_npm
|
||||
Write-Host "[ERROR] npm run build failed (exit code $LASTEXITCODE)" -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
Pop-Location
|
||||
$ErrorActionPreference = $prevEAP_npm
|
||||
Write-Host "[OK] Frontend built to studio/frontend/dist" -ForegroundColor Green
|
||||
}
|
||||
|
||||
# ==========================================================================
|
||||
# PHASE 3: Python environment + dependencies
|
||||
# ==========================================================================
|
||||
Write-Host ""
|
||||
Write-Host "Setting up Python environment..." -ForegroundColor Cyan
|
||||
|
||||
# Find Python
|
||||
$PythonCmd = $null
|
||||
foreach ($candidate in @("python3.12", "python3.11", "python3.10", "python3.9", "python3", "python")) {
|
||||
try {
|
||||
$ver = & $candidate --version 2>&1
|
||||
if ($ver -match 'Python 3\.(\d+)') {
|
||||
$minor = [int]$Matches[1]
|
||||
if ($minor -le 12) {
|
||||
$PythonCmd = $candidate
|
||||
break
|
||||
}
|
||||
}
|
||||
} catch { }
|
||||
}
|
||||
|
||||
if (-not $PythonCmd) {
|
||||
Write-Host "[ERROR] No Python <= 3.12 found." -ForegroundColor Red
|
||||
exit 1
|
||||
}
|
||||
|
||||
Write-Host "[OK] Using $PythonCmd ($(& $PythonCmd --version 2>&1))" -ForegroundColor Green
|
||||
|
||||
# Always create a .venv for isolation -- even for pip installs.
|
||||
# Created in the current working directory (where user ran the command).
|
||||
$VenvDir = Join-Path (Get-Location) ".venv"
|
||||
if (-not (Test-Path $VenvDir)) {
|
||||
Write-Host " Creating virtual environment at $VenvDir..." -ForegroundColor Cyan
|
||||
& $PythonCmd -m venv $VenvDir
|
||||
} else {
|
||||
Write-Host " Reusing existing virtual environment at $VenvDir" -ForegroundColor Green
|
||||
}
|
||||
|
||||
# pip and python write to stderr even on success (progress bars, warnings).
|
||||
# With $ErrorActionPreference = "Stop" (set at top of script), PS 5.1
|
||||
# converts stderr lines into terminating ErrorRecords, breaking output.
|
||||
# Lower to "Continue" for the pip/python section.
|
||||
$prevEAP = $ErrorActionPreference
|
||||
$ErrorActionPreference = "Continue"
|
||||
|
||||
$ActivateScript = Join-Path $VenvDir "Scripts\Activate.ps1"
|
||||
. $ActivateScript
|
||||
pip install --upgrade pip 2>&1 | Out-Null
|
||||
|
||||
# if (-not $IsPipInstall) {
|
||||
# # Running from repo: copy requirements and do editable install
|
||||
# $RepoRoot = (Resolve-Path (Join-Path $ScriptDir "..\..")).Path
|
||||
# $ReqsSrc = Join-Path $RepoRoot "backend\requirements"
|
||||
# $ReqsDst = Join-Path $PackageDir "requirements"
|
||||
# if (-not (Test-Path $ReqsDst)) { New-Item -ItemType Directory -Path $ReqsDst | Out-Null }
|
||||
# Copy-Item (Join-Path $ReqsSrc "*.txt") $ReqsDst -Force
|
||||
|
||||
# Write-Host " Installing CLI entry point..." -ForegroundColor Cyan
|
||||
# pip install -e $RepoRoot 2>&1 | Out-Null
|
||||
# } else {
|
||||
# # Running from pip install: the package is in system Python but not in
|
||||
# # the fresh .venv. Install it so run_install() can find its modules
|
||||
# # and bundled requirements files.
|
||||
# Write-Host " Installing package into venv..." -ForegroundColor Cyan
|
||||
# pip install unsloth-roland-test 2>&1 | Out-Null
|
||||
# }
|
||||
|
||||
# Pre-install PyTorch with CUDA support.
|
||||
# On Windows, the default PyPI torch wheel is CPU-only.
|
||||
# We need PyTorch's CUDA index to get GPU-enabled wheels.
|
||||
# PyTorch bundles its own CUDA runtime, so this works regardless
|
||||
# of whether the CUDA Toolkit is installed yet.
|
||||
# The CUDA tag is chosen based on the driver's max supported CUDA version.
|
||||
|
||||
# Windows MAX_PATH (260 chars) causes Triton kernel compilation to fail because
|
||||
# the auto-generated filenames are extremely long. Use a short cache directory.
|
||||
$TorchCacheDir = "C:\tc"
|
||||
if (-not (Test-Path $TorchCacheDir)) { New-Item -ItemType Directory -Path $TorchCacheDir -Force | Out-Null }
|
||||
$env:TORCHINDUCTOR_CACHE_DIR = $TorchCacheDir
|
||||
[Environment]::SetEnvironmentVariable('TORCHINDUCTOR_CACHE_DIR', $TorchCacheDir, 'User')
|
||||
Write-Host "[OK] TORCHINDUCTOR_CACHE_DIR set to $TorchCacheDir (avoids MAX_PATH issues)" -ForegroundColor Green
|
||||
|
||||
$CuTag = Get-PytorchCudaTag
|
||||
Write-Host " Installing PyTorch with CUDA support ($CuTag)..." -ForegroundColor Cyan
|
||||
pip install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/$CuTag" 2>&1 | Out-Null
|
||||
|
||||
# Ordered heavy dependency installation — shared cross-platform script
|
||||
Write-Host " Running ordered dependency installation..." -ForegroundColor Cyan
|
||||
python "$PSScriptRoot\install_python_stack.py"
|
||||
# Restore ErrorActionPreference after pip/python work
|
||||
$ErrorActionPreference = $prevEAP
|
||||
|
||||
# ==========================================================================
|
||||
# PHASE 3.5: Install OpenSSL dev (for HTTPS support in llama-server)
|
||||
# ==========================================================================
|
||||
# llama-server needs OpenSSL to download models from HuggingFace via -hf.
|
||||
# ShiningLight.OpenSSL.Dev includes headers + libs that cmake can find.
|
||||
$OpenSslAvailable = $false
|
||||
|
||||
# Check if OpenSSL dev is already installed (look for include dir)
|
||||
$OpenSslRoots = @(
|
||||
'C:\Program Files\OpenSSL-Win64',
|
||||
'C:\Program Files\OpenSSL',
|
||||
'C:\OpenSSL-Win64'
|
||||
)
|
||||
$OpenSslRoot = $null
|
||||
foreach ($root in $OpenSslRoots) {
|
||||
if (Test-Path (Join-Path $root 'include\openssl\ssl.h')) {
|
||||
$OpenSslRoot = $root
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
if ($OpenSslRoot) {
|
||||
$OpenSslAvailable = $true
|
||||
Write-Host "[OK] OpenSSL dev found at $OpenSslRoot" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host ""
|
||||
Write-Host "Installing OpenSSL dev (for HTTPS in llama-server)..." -ForegroundColor Cyan
|
||||
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
|
||||
if ($HasWinget) {
|
||||
winget install -e --id ShiningLight.OpenSSL.Dev --accept-package-agreements --accept-source-agreements
|
||||
# Re-check after install
|
||||
foreach ($root in $OpenSslRoots) {
|
||||
if (Test-Path (Join-Path $root 'include\openssl\ssl.h')) {
|
||||
$OpenSslRoot = $root
|
||||
$OpenSslAvailable = $true
|
||||
Write-Host "[OK] OpenSSL dev installed at $OpenSslRoot" -ForegroundColor Green
|
||||
break
|
||||
}
|
||||
}
|
||||
}
|
||||
if (-not $OpenSslAvailable) {
|
||||
Write-Host "[WARN] OpenSSL dev not available -- llama-server will be built without HTTPS" -ForegroundColor Yellow
|
||||
}
|
||||
}
|
||||
|
||||
# ==========================================================================
|
||||
# PHASE 4: Build llama.cpp with CUDA for GGUF inference + export
|
||||
# ==========================================================================
|
||||
# Builds at ~/.unsloth/llama.cpp — a single shared location under the user's
|
||||
# home directory. This is used by both the inference server and the GGUF
|
||||
# export pipeline (unsloth-zoo).
|
||||
# We build:
|
||||
# - llama-server: for GGUF model inference (with HTTPS if OpenSSL available)
|
||||
# - llama-quantize: for GGUF export quantization
|
||||
# Prerequisites (git, cmake, VS Build Tools, CUDA Toolkit) already installed in Phase 1.
|
||||
$UnslothHome = Join-Path $env:USERPROFILE ".unsloth"
|
||||
if (-not (Test-Path $UnslothHome)) { New-Item -ItemType Directory -Force $UnslothHome | Out-Null }
|
||||
$LlamaCppDir = Join-Path $UnslothHome "llama.cpp"
|
||||
$BuildDir = Join-Path $LlamaCppDir "build"
|
||||
$LlamaServerBin = Join-Path $BuildDir "bin\Release\llama-server.exe"
|
||||
|
||||
if (Test-Path $LlamaServerBin) {
|
||||
Write-Host ""
|
||||
Write-Host "[OK] llama-server already exists at $LlamaServerBin" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host ""
|
||||
Write-Host "Building llama.cpp with CUDA support..." -ForegroundColor Cyan
|
||||
Write-Host " This typically takes 5-10 minutes on first build." -ForegroundColor Gray
|
||||
Write-Host ""
|
||||
|
||||
# Start total build timer
|
||||
$totalSw = [System.Diagnostics.Stopwatch]::StartNew()
|
||||
|
||||
# Native commands (git, cmake) write to stderr even on success.
|
||||
# With $ErrorActionPreference = "Stop" (set at top of script), PS 5.1
|
||||
# converts stderr lines into terminating ErrorRecords, breaking output.
|
||||
# Lower to "Continue" for the build section.
|
||||
$prevEAP = $ErrorActionPreference
|
||||
$ErrorActionPreference = "Continue"
|
||||
|
||||
$BuildOk = $true
|
||||
$FailedStep = ""
|
||||
|
||||
# -- Step A: Clone or pull llama.cpp --
|
||||
|
||||
if (Test-Path (Join-Path $LlamaCppDir ".git")) {
|
||||
Write-Host " llama.cpp repo already cloned, pulling latest..." -ForegroundColor Gray
|
||||
git -C $LlamaCppDir pull
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
Write-Host " [WARN] git pull failed -- using existing source" -ForegroundColor Yellow
|
||||
}
|
||||
} else {
|
||||
Write-Host " Cloning llama.cpp..." -ForegroundColor Gray
|
||||
if (Test-Path $LlamaCppDir) { Remove-Item -Recurse -Force $LlamaCppDir }
|
||||
git clone --depth 1 https://github.com/ggml-org/llama.cpp.git $LlamaCppDir
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
$BuildOk = $false
|
||||
$FailedStep = "git clone"
|
||||
}
|
||||
}
|
||||
|
||||
# -- Step B: cmake configure (CUDA + Unsloth flags) --
|
||||
if ($BuildOk) {
|
||||
Write-Host ""
|
||||
Write-Host "--- cmake configure ---" -ForegroundColor Cyan
|
||||
|
||||
$CmakeArgs = @(
|
||||
'-S', $LlamaCppDir,
|
||||
'-B', $BuildDir,
|
||||
'-G', $CmakeGenerator,
|
||||
'-Wno-dev'
|
||||
)
|
||||
# Tell cmake exactly where VS is (bypasses registry lookup)
|
||||
if ($VsInstallPath) {
|
||||
$CmakeArgs += "-DCMAKE_GENERATOR_INSTANCE=$VsInstallPath"
|
||||
}
|
||||
# Common flags
|
||||
$CmakeArgs += '-DBUILD_SHARED_LIBS=OFF'
|
||||
# HTTPS support via OpenSSL
|
||||
if ($OpenSslAvailable -and $OpenSslRoot) {
|
||||
$CmakeArgs += "-DOPENSSL_ROOT_DIR=$OpenSslRoot"
|
||||
$CmakeArgs += '-DLLAMA_OPENSSL=ON'
|
||||
} else {
|
||||
$CmakeArgs += '-DLLAMA_CURL=OFF'
|
||||
}
|
||||
$CmakeArgs += '-DCMAKE_EXE_LINKER_FLAGS=/NODEFAULTLIB:LIBCMT'
|
||||
# CUDA flags (Unsloth-aligned)
|
||||
$CmakeArgs += '-DGGML_CUDA=ON'
|
||||
$CmakeArgs += "-DCUDAToolkit_ROOT=$CudaToolkitRoot"
|
||||
$CmakeArgs += "-DCMAKE_CUDA_COMPILER=$NvccPath"
|
||||
$CmakeArgs += '-DGGML_CUDA_FA_ALL_QUANTS=ON'
|
||||
$CmakeArgs += '-DGGML_CUDA_F16=OFF'
|
||||
$CmakeArgs += '-DGGML_CUDA_GRAPHS=OFF'
|
||||
$CmakeArgs += '-DGGML_CUDA_FORCE_CUBLAS=OFF'
|
||||
$CmakeArgs += '-DGGML_CUDA_PEER_MAX_BATCH_SIZE=8192'
|
||||
if ($CudaArch) {
|
||||
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$CudaArch"
|
||||
}
|
||||
|
||||
Write-Host " cmake args:" -ForegroundColor Gray
|
||||
foreach ($arg in $CmakeArgs) {
|
||||
Write-Host " $arg" -ForegroundColor Gray
|
||||
}
|
||||
Write-Host ""
|
||||
|
||||
cmake @CmakeArgs
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
$BuildOk = $false
|
||||
$FailedStep = "cmake configure"
|
||||
}
|
||||
}
|
||||
|
||||
# -- Step C: Build llama-server --
|
||||
$NumCpu = [Environment]::ProcessorCount
|
||||
if ($NumCpu -lt 1) { $NumCpu = 4 }
|
||||
|
||||
if ($BuildOk) {
|
||||
Write-Host ""
|
||||
Write-Host "--- cmake build (llama-server) ---" -ForegroundColor Cyan
|
||||
Write-Host " Parallel jobs: $NumCpu" -ForegroundColor Gray
|
||||
Write-Host ""
|
||||
|
||||
cmake --build $BuildDir --config Release --target llama-server -j $NumCpu
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
$BuildOk = $false
|
||||
$FailedStep = "cmake build (llama-server)"
|
||||
}
|
||||
}
|
||||
|
||||
# -- Step D: Build llama-quantize (optional, best-effort) --
|
||||
if ($BuildOk) {
|
||||
Write-Host ""
|
||||
Write-Host "--- cmake build (llama-quantize) ---" -ForegroundColor Cyan
|
||||
cmake --build $BuildDir --config Release --target llama-quantize -j $NumCpu
|
||||
if ($LASTEXITCODE -ne 0) {
|
||||
Write-Host " [WARN] llama-quantize build failed (GGUF export may be unavailable)" -ForegroundColor Yellow
|
||||
}
|
||||
}
|
||||
|
||||
# Restore ErrorActionPreference
|
||||
$ErrorActionPreference = $prevEAP
|
||||
|
||||
# Stop timer
|
||||
$totalSw.Stop()
|
||||
$totalMin = [math]::Floor($totalSw.Elapsed.TotalMinutes)
|
||||
$totalSec = [math]::Round($totalSw.Elapsed.TotalSeconds % 60, 1)
|
||||
|
||||
# -- Summary --
|
||||
Write-Host ""
|
||||
if ($BuildOk -and (Test-Path $LlamaServerBin)) {
|
||||
Write-Host "[OK] llama-server built at $LlamaServerBin" -ForegroundColor Green
|
||||
$QuantizeBin = Join-Path $BuildDir "bin\Release\llama-quantize.exe"
|
||||
if (Test-Path $QuantizeBin) {
|
||||
Write-Host "[OK] llama-quantize available for GGUF export" -ForegroundColor Green
|
||||
}
|
||||
Write-Host " Build time: ${totalMin}m ${totalSec}s" -ForegroundColor Cyan
|
||||
} else {
|
||||
# Check alternate paths (some cmake generators don't use Release subdir)
|
||||
$altBin = Join-Path $BuildDir "bin\llama-server.exe"
|
||||
if ($BuildOk -and (Test-Path $altBin)) {
|
||||
Write-Host "[OK] llama-server built at $altBin" -ForegroundColor Green
|
||||
Write-Host " Build time: ${totalMin}m ${totalSec}s" -ForegroundColor Cyan
|
||||
} else {
|
||||
Write-Host "[FAILED] llama.cpp build failed at step: $FailedStep (${totalMin}m ${totalSec}s)" -ForegroundColor Red
|
||||
Write-Host " To retry: delete $LlamaCppDir and re-run setup." -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
# ============================================
|
||||
# Add shell aliases (PowerShell profile + cmd batch files)
|
||||
# ============================================
|
||||
Write-Host ""
|
||||
$RepoDir = $PSScriptRoot
|
||||
$VenvPython = Join-Path $RepoDir ".venv\Scripts\python.exe"
|
||||
$CliScript = Join-Path $RepoDir "cli.py"
|
||||
$FrontendDist = Join-Path $RepoDir "studio\frontend\dist"
|
||||
$AliasAdded = $false
|
||||
|
||||
# --- PowerShell profile: add functions ---
|
||||
$ProfileDir = Split-Path $PROFILE -Parent
|
||||
if (-not (Test-Path $ProfileDir)) { New-Item -ItemType Directory -Path $ProfileDir -Force | Out-Null }
|
||||
if (-not (Test-Path $PROFILE)) { New-Item -ItemType File -Path $PROFILE -Force | Out-Null }
|
||||
|
||||
if (-not (Select-String -Path $PROFILE -Pattern "unsloth-studio" -Quiet -ErrorAction SilentlyContinue)) {
|
||||
$block = @"
|
||||
|
||||
# Unsloth Studio launcher
|
||||
function unsloth-studio { & "$VenvPython" "$CliScript" studio -f "$FrontendDist" @args }
|
||||
function unsloth-ui { & "$VenvPython" "$CliScript" studio -f "$FrontendDist" @args }
|
||||
"@
|
||||
Add-Content -Path $PROFILE -Value $block
|
||||
Write-Host "[OK] Aliases 'unsloth-studio' and 'unsloth-ui' added to $PROFILE" -ForegroundColor Green
|
||||
$AliasAdded = $true
|
||||
} else {
|
||||
Write-Host "[OK] Aliases 'unsloth-studio' and 'unsloth-ui' already exist in $PROFILE" -ForegroundColor Green
|
||||
}
|
||||
|
||||
# --- cmd.exe: create batch files and ensure they're on PATH ---
|
||||
$BatDir = Join-Path $RepoDir ".venv\Scripts"
|
||||
foreach ($name in @("unsloth-studio", "unsloth-ui")) {
|
||||
$batPath = Join-Path $BatDir "$name.bat"
|
||||
if (-not (Test-Path $batPath)) {
|
||||
Set-Content -Path $batPath -Value "@echo off`r`n`"$VenvPython`" `"$CliScript`" studio -f `"$FrontendDist`" %*"
|
||||
}
|
||||
}
|
||||
# Persist .venv\Scripts to User PATH so commands work in new cmd.exe terminals without activation
|
||||
$userPath = [Environment]::GetEnvironmentVariable('Path', 'User')
|
||||
if (-not $userPath -or $userPath -notlike "*$BatDir*") {
|
||||
if ($userPath) {
|
||||
[Environment]::SetEnvironmentVariable('Path', "$BatDir;$userPath", 'User')
|
||||
} else {
|
||||
[Environment]::SetEnvironmentVariable('Path', "$BatDir", 'User')
|
||||
}
|
||||
Write-Host " Persisted $BatDir to User PATH" -ForegroundColor Gray
|
||||
}
|
||||
Write-Host "[OK] Batch launchers created (works from any new cmd.exe or PowerShell)" -ForegroundColor Green
|
||||
|
||||
# ============================================
|
||||
# Done
|
||||
# ============================================
|
||||
Write-Host ""
|
||||
Write-Host "+===============================================+" -ForegroundColor Green
|
||||
Write-Host "| Setup Complete! |" -ForegroundColor Green
|
||||
Write-Host "| |" -ForegroundColor Green
|
||||
Write-Host "| IMPORTANT: Open a NEW terminal, then run: |" -ForegroundColor Yellow
|
||||
Write-Host "| |" -ForegroundColor Green
|
||||
Write-Host "| unsloth-studio -H 0.0.0.0 -p 8000 |" -ForegroundColor Green
|
||||
Write-Host "| |" -ForegroundColor Green
|
||||
Write-Host "+===============================================+" -ForegroundColor Green
|
||||
37
setup.sh
37
setup.sh
|
|
@ -168,35 +168,9 @@ SINGLE_ENV_CONSTRAINTS="$REQ_ROOT/single-env/constraints.txt"
|
|||
SINGLE_ENV_DATA_DESIGNER="$REQ_ROOT/single-env/data-designer.txt"
|
||||
SINGLE_ENV_DATA_DESIGNER_DEPS="$REQ_ROOT/single-env/data-designer-deps.txt"
|
||||
SINGLE_ENV_PATCH="$REQ_ROOT/single-env/patch_metadata.py"
|
||||
LOCAL_DD_UNSTRUCTURED_PLUGIN="$SCRIPT_DIR/studio/backend/plugins/data-designer-unstructured-seed"
|
||||
|
||||
install_python_stack() {
|
||||
run_quiet "pip upgrade" pip install --upgrade pip
|
||||
echo " Installing unsloth-zoo + unsloth..."
|
||||
run_quiet "pip install unsloth" pip install --no-cache-dir -c "$SINGLE_ENV_CONSTRAINTS" -r "$REQ_ROOT/base.txt"
|
||||
echo " Installing additional unsloth dependencies..."
|
||||
run_quiet "pip install extras" pip install --no-cache-dir -c "$SINGLE_ENV_CONSTRAINTS" -r "$REQ_ROOT/extras.txt"
|
||||
run_quiet "pip install torchao+transformers" pip install --force-reinstall --no-cache-dir -c "$SINGLE_ENV_CONSTRAINTS" -r "$REQ_ROOT/overrides.txt"
|
||||
run_quiet "pip install triton_kernels" pip install --no-deps --no-cache-dir -r "$REQ_ROOT/triton-kernels.txt"
|
||||
# Patch: override llama_cpp.py with fix from unsloth-zoo branch
|
||||
LLAMA_CPP_DST="$(pip show unsloth-zoo | grep -i '^Location:' | awk '{print $2}')/unsloth_zoo/llama_cpp.py"
|
||||
curl -sSL "https://raw.githubusercontent.com/unslothai/unsloth-zoo/refs/heads/main/unsloth_zoo/llama_cpp.py" \
|
||||
-o "$LLAMA_CPP_DST"
|
||||
# Patch: override vision.py with fix from unsloth PR: https://github.com/unslothai/unsloth/pull/4091 until next pypi release
|
||||
VISION_DST="$(pip show unsloth | grep -i '^Location:' | awk '{print $2}')/unsloth/models/vision.py"
|
||||
curl -sSL "https://raw.githubusercontent.com/unslothai/unsloth/80e0108a684c882965a02a8ed851e3473c1145ab/unsloth/models/vision.py" \
|
||||
-o "$VISION_DST"
|
||||
echo " Installing studio dependencies..."
|
||||
run_quiet "pip install studio" pip install --no-cache-dir -c "$SINGLE_ENV_CONSTRAINTS" -r "$REQ_ROOT/studio.txt"
|
||||
echo " Installing data-designer dependencies..."
|
||||
run_quiet "pip install data-designer deps" pip install --no-cache-dir -c "$SINGLE_ENV_CONSTRAINTS" -r "$SINGLE_ENV_DATA_DESIGNER_DEPS"
|
||||
echo " Installing data-designer..."
|
||||
run_quiet "pip install data-designer" pip install --no-cache-dir --no-deps -c "$SINGLE_ENV_CONSTRAINTS" -r "$SINGLE_ENV_DATA_DESIGNER"
|
||||
echo " Installing local data-designer unstructured plugin..."
|
||||
run_quiet "pip install data-designer-unstructured-seed" pip install --no-cache-dir --no-deps -e "$LOCAL_DD_UNSTRUCTURED_PLUGIN"
|
||||
run_quiet "patch single-env metadata" python "$SINGLE_ENV_PATCH"
|
||||
run_quiet "pip check" pip check
|
||||
echo "✅ Python dependencies installed"
|
||||
python "$SCRIPT_DIR/install_python_stack.py"
|
||||
}
|
||||
|
||||
if [ "$IS_COLAB" = true ]; then
|
||||
|
|
@ -223,11 +197,14 @@ else
|
|||
fi
|
||||
|
||||
# ── 8. Build llama.cpp binaries for GGUF inference + export ──
|
||||
# Builds in-tree at $REPO/llama.cpp/. This directory is shared with
|
||||
# unsloth-zoo's GGUF export pipeline. We build:
|
||||
# Builds at ~/.unsloth/llama.cpp — a single shared location under the user's
|
||||
# home directory. This is used by both the inference server and the GGUF
|
||||
# export pipeline (unsloth-zoo).
|
||||
# - llama-server: for GGUF model inference
|
||||
# - llama-quantize: for GGUF export quantization (symlinked to root for check_llama_cpp())
|
||||
LLAMA_CPP_DIR="$SCRIPT_DIR/llama.cpp"
|
||||
UNSLOTH_HOME="$HOME/.unsloth"
|
||||
mkdir -p "$UNSLOTH_HOME"
|
||||
LLAMA_CPP_DIR="$UNSLOTH_HOME/llama.cpp"
|
||||
LLAMA_SERVER_BIN="$LLAMA_CPP_DIR/build/bin/llama-server"
|
||||
rm -rf "$LLAMA_CPP_DIR"
|
||||
{
|
||||
|
|
|
|||
|
|
@ -418,9 +418,8 @@ class ExportBackend:
|
|||
pre_existing_ggufs = set(glob.glob(os.path.join(cwd, "*.gguf")))
|
||||
|
||||
# Pass absolute path — no os.chdir needed.
|
||||
# unsloth saves intermediate HF model files into model_save_path,
|
||||
# while check_llama_cpp("llama.cpp") resolves against cwd (repo root)
|
||||
# where setup.sh already built llama.cpp with quantizer.
|
||||
# unsloth saves intermediate HF model files into model_save_path.
|
||||
# unsloth-zoo's check_llama_cpp() uses ~/.unsloth/llama.cpp by default.
|
||||
model_save_path = os.path.join(abs_save_dir, "model")
|
||||
self.current_model.save_pretrained_gguf(
|
||||
model_save_path,
|
||||
|
|
|
|||
|
|
@ -41,6 +41,8 @@ class LlamaCppBackend:
|
|||
self._is_vision: bool = False
|
||||
self._healthy = False
|
||||
self._lock = threading.Lock()
|
||||
self._stdout_lines: list[str] = []
|
||||
self._stdout_thread: Optional[threading.Thread] = None
|
||||
|
||||
atexit.register(self._cleanup)
|
||||
|
||||
|
|
@ -74,33 +76,83 @@ class LlamaCppBackend:
|
|||
Locate the llama-server binary.
|
||||
|
||||
Search order:
|
||||
1. LLAMA_SERVER_PATH environment variable
|
||||
2. ./llama.cpp/build/bin/llama-server (built by setup.sh in-tree)
|
||||
3. llama-server on PATH (system install)
|
||||
4. ./bin/llama-server (legacy: extracted binary)
|
||||
1. LLAMA_SERVER_PATH environment variable (direct path to binary)
|
||||
1b. UNSLOTH_LLAMA_CPP_PATH env var (custom llama.cpp install dir)
|
||||
2. ~/.unsloth/llama.cpp/llama-server (make build, root dir)
|
||||
3. ~/.unsloth/llama.cpp/build/bin/llama-server (cmake build, Linux)
|
||||
4. ~/.unsloth/llama.cpp/build/bin/Release/llama-server.exe (cmake build, Windows)
|
||||
5. ./llama.cpp/llama-server (legacy: make build, root dir)
|
||||
6. ./llama.cpp/build/bin/llama-server (legacy: cmake in-tree build)
|
||||
7. llama-server on PATH (system install)
|
||||
8. ./bin/llama-server (legacy: extracted binary)
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
|
||||
# 1. Env var
|
||||
binary_name = "llama-server.exe" if sys.platform == "win32" else "llama-server"
|
||||
|
||||
# 1. Env var — direct path to binary
|
||||
env_path = os.environ.get("LLAMA_SERVER_PATH")
|
||||
if env_path and Path(env_path).is_file():
|
||||
return env_path
|
||||
|
||||
# Project root: llama_cpp.py → inference/ → core/ → backend/ → studio/ → root
|
||||
project_root = Path(__file__).resolve().parents[4]
|
||||
# 1b. UNSLOTH_LLAMA_CPP_PATH — custom llama.cpp install directory
|
||||
custom_llama_cpp = os.environ.get("UNSLOTH_LLAMA_CPP_PATH")
|
||||
if custom_llama_cpp:
|
||||
custom_dir = Path(custom_llama_cpp)
|
||||
# Root dir (make builds)
|
||||
root_bin = custom_dir / binary_name
|
||||
if root_bin.is_file():
|
||||
return str(root_bin)
|
||||
# build/bin/ (cmake builds on Linux)
|
||||
cmake_bin = custom_dir / "build" / "bin" / binary_name
|
||||
if cmake_bin.is_file():
|
||||
return str(cmake_bin)
|
||||
# build/bin/Release/ (cmake builds on Windows)
|
||||
if sys.platform == "win32":
|
||||
win_bin = custom_dir / "build" / "bin" / "Release" / binary_name
|
||||
if win_bin.is_file():
|
||||
return str(win_bin)
|
||||
|
||||
# 2. In-tree llama.cpp build (setup.sh builds here)
|
||||
build_path = project_root / "llama.cpp" / "build" / "bin" / "llama-server"
|
||||
# 2–4. ~/.unsloth/llama.cpp (primary — setup.sh / setup.ps1 build here)
|
||||
unsloth_home = Path.home() / ".unsloth" / "llama.cpp"
|
||||
# Root dir (make builds copy binaries here)
|
||||
home_root = unsloth_home / binary_name
|
||||
if home_root.is_file():
|
||||
return str(home_root)
|
||||
# build/bin/ (cmake builds on Linux)
|
||||
home_linux = unsloth_home / "build" / "bin" / binary_name
|
||||
if home_linux.is_file():
|
||||
return str(home_linux)
|
||||
|
||||
# 3. Windows MSVC build has Release subdir
|
||||
if sys.platform == "win32":
|
||||
home_win = unsloth_home / "build" / "bin" / "Release" / binary_name
|
||||
if home_win.is_file():
|
||||
return str(home_win)
|
||||
|
||||
# 5–6. Legacy: in-tree build (older setup.sh / setup.ps1 versions)
|
||||
project_root = Path(__file__).resolve().parents[4]
|
||||
# Root dir (make builds)
|
||||
root_path = project_root / "llama.cpp" / binary_name
|
||||
if root_path.is_file():
|
||||
return str(root_path)
|
||||
# build/bin/ (cmake builds)
|
||||
build_path = project_root / "llama.cpp" / "build" / "bin" / binary_name
|
||||
if build_path.is_file():
|
||||
return str(build_path)
|
||||
if sys.platform == "win32":
|
||||
win_path = project_root / "llama.cpp" / "build" / "bin" / "Release" / binary_name
|
||||
if win_path.is_file():
|
||||
return str(win_path)
|
||||
|
||||
# 3. System PATH
|
||||
# 7. System PATH
|
||||
system_path = shutil.which("llama-server")
|
||||
if system_path:
|
||||
return system_path
|
||||
|
||||
# 4. Legacy: extracted to bin/
|
||||
bin_path = project_root / "bin" / "llama-server"
|
||||
# 8. Legacy: extracted to bin/
|
||||
bin_path = project_root / "bin" / binary_name
|
||||
if bin_path.is_file():
|
||||
return str(bin_path)
|
||||
|
||||
|
|
@ -115,6 +167,26 @@ class LlamaCppBackend:
|
|||
s.bind(("", 0))
|
||||
return s.getsockname()[1]
|
||||
|
||||
# ── Stdout drain (prevents pipe deadlock on Windows) ─────────
|
||||
|
||||
def _drain_stdout(self):
|
||||
"""
|
||||
Read lines from the subprocess stdout in a background thread.
|
||||
|
||||
This prevents a pipe-buffer deadlock on Windows where the default
|
||||
pipe buffer is only ~4 KB. Without draining, llama-server blocks
|
||||
on writes and never becomes healthy.
|
||||
"""
|
||||
try:
|
||||
for line in self._process.stdout:
|
||||
line = line.rstrip()
|
||||
if line:
|
||||
self._stdout_lines.append(line)
|
||||
logger.info(f"[llama-server] {line}")
|
||||
except (ValueError, OSError):
|
||||
# Pipe closed — process is terminating
|
||||
pass
|
||||
|
||||
# ── Lifecycle ─────────────────────────────────────────────────
|
||||
|
||||
def load_model(
|
||||
|
|
@ -122,6 +194,8 @@ class LlamaCppBackend:
|
|||
*,
|
||||
# Local mode: pass a path to a .gguf file
|
||||
gguf_path: Optional[str] = None,
|
||||
# Vision projection (mmproj) for local vision models
|
||||
mmproj_path: Optional[str] = None,
|
||||
# HF mode: let llama-server download via -hf "repo:quant"
|
||||
hf_repo: Optional[str] = None,
|
||||
hf_variant: Optional[str] = None,
|
||||
|
|
@ -160,16 +234,58 @@ class LlamaCppBackend:
|
|||
|
||||
# Build command based on mode
|
||||
if hf_repo:
|
||||
hf_spec = f"{hf_repo}:{hf_variant}" if hf_variant else hf_repo
|
||||
# Download the GGUF file ourselves using huggingface_hub
|
||||
# (llama-server's -hf flag requires HTTPS/curl which may not
|
||||
# be available, e.g. Windows builds with -DLLAMA_CURL=OFF)
|
||||
try:
|
||||
from huggingface_hub import hf_hub_download
|
||||
except ImportError:
|
||||
raise RuntimeError(
|
||||
"huggingface_hub is required for HF model loading. "
|
||||
"Install it with: pip install huggingface_hub"
|
||||
)
|
||||
|
||||
# Determine the filename from the variant (e.g., "Q4_K_M" -> find matching file)
|
||||
gguf_filename = None
|
||||
if hf_variant:
|
||||
# Try common naming patterns
|
||||
try:
|
||||
from huggingface_hub import list_repo_files
|
||||
files = list_repo_files(hf_repo, token=hf_token)
|
||||
variant_lower = hf_variant.lower()
|
||||
for f in files:
|
||||
if f.endswith(".gguf") and variant_lower in f.lower():
|
||||
gguf_filename = f
|
||||
break
|
||||
except Exception as e:
|
||||
logger.warning(f"Could not list repo files: {e}")
|
||||
|
||||
if not gguf_filename:
|
||||
# Fallback: construct common filename pattern
|
||||
# e.g., "unsloth/gemma-3-4b-it-GGUF" + "Q4_K_M" -> try model name
|
||||
repo_name = hf_repo.split("/")[-1].replace("-GGUF", "")
|
||||
gguf_filename = f"{repo_name}-{hf_variant}.gguf"
|
||||
|
||||
logger.info(f"Downloading GGUF: {hf_repo}/{gguf_filename}")
|
||||
try:
|
||||
local_path = hf_hub_download(
|
||||
repo_id=hf_repo,
|
||||
filename=gguf_filename,
|
||||
token=hf_token,
|
||||
)
|
||||
except Exception as e:
|
||||
raise RuntimeError(
|
||||
f"Failed to download GGUF file '{gguf_filename}' from {hf_repo}: {e}"
|
||||
)
|
||||
|
||||
logger.info(f"GGUF downloaded to: {local_path}")
|
||||
cmd = [
|
||||
binary,
|
||||
"-hf", hf_spec,
|
||||
"-m", local_path,
|
||||
"--port", str(self._port),
|
||||
"-c", str(n_ctx),
|
||||
"-ngl", str(n_gpu_layers),
|
||||
]
|
||||
if hf_token:
|
||||
cmd.extend(["--hf-token", hf_token])
|
||||
elif gguf_path:
|
||||
if not Path(gguf_path).is_file():
|
||||
raise FileNotFoundError(f"GGUF file not found: {gguf_path}")
|
||||
|
|
@ -186,16 +302,43 @@ class LlamaCppBackend:
|
|||
if n_threads is not None:
|
||||
cmd.extend(["--threads", str(n_threads)])
|
||||
|
||||
# Append mmproj for local vision models
|
||||
if mmproj_path:
|
||||
if not Path(mmproj_path).is_file():
|
||||
logger.warning(f"mmproj file not found: {mmproj_path}")
|
||||
else:
|
||||
cmd.extend(["--mmproj", mmproj_path])
|
||||
logger.info(f"Using mmproj for vision: {mmproj_path}")
|
||||
|
||||
logger.info(f"Starting llama-server: {' '.join(cmd)}")
|
||||
|
||||
# Set LD_LIBRARY_PATH so llama-server can find its shared libs
|
||||
# (libmtmd.so, libllama.so, etc.) which live next to the binary
|
||||
# Set library paths so llama-server can find its shared libs and CUDA DLLs
|
||||
import os
|
||||
import sys
|
||||
env = os.environ.copy()
|
||||
binary_dir = str(Path(binary).parent)
|
||||
existing_ld = env.get("LD_LIBRARY_PATH", "")
|
||||
env["LD_LIBRARY_PATH"] = f"{binary_dir}:{existing_ld}" if existing_ld else binary_dir
|
||||
|
||||
if sys.platform == "win32":
|
||||
# On Windows, CUDA DLLs (cublas64_12.dll, cudart64_12.dll, etc.)
|
||||
# must be on PATH. Add CUDA_PATH\bin if available.
|
||||
path_dirs = [binary_dir]
|
||||
cuda_path = os.environ.get("CUDA_PATH", "")
|
||||
if cuda_path:
|
||||
cuda_bin = os.path.join(cuda_path, "bin")
|
||||
if os.path.isdir(cuda_bin):
|
||||
path_dirs.append(cuda_bin)
|
||||
# Some CUDA installs put DLLs in bin\x64
|
||||
cuda_bin_x64 = os.path.join(cuda_path, "bin", "x64")
|
||||
if os.path.isdir(cuda_bin_x64):
|
||||
path_dirs.append(cuda_bin_x64)
|
||||
existing_path = env.get("PATH", "")
|
||||
env["PATH"] = ";".join(path_dirs) + ";" + existing_path
|
||||
else:
|
||||
# Linux: set LD_LIBRARY_PATH for shared libs next to the binary
|
||||
existing_ld = env.get("LD_LIBRARY_PATH", "")
|
||||
env["LD_LIBRARY_PATH"] = f"{binary_dir}:{existing_ld}" if existing_ld else binary_dir
|
||||
|
||||
self._stdout_lines = []
|
||||
self._process = subprocess.Popen(
|
||||
cmd,
|
||||
stdout=subprocess.PIPE,
|
||||
|
|
@ -204,15 +347,20 @@ class LlamaCppBackend:
|
|||
env=env,
|
||||
)
|
||||
|
||||
# Start background thread to drain stdout and prevent pipe deadlock
|
||||
self._stdout_thread = threading.Thread(
|
||||
target=self._drain_stdout, daemon=True, name="llama-stdout"
|
||||
)
|
||||
self._stdout_thread.start()
|
||||
|
||||
self._gguf_path = gguf_path
|
||||
self._hf_repo = hf_repo
|
||||
self._hf_variant = hf_variant
|
||||
self._is_vision = is_vision
|
||||
self._model_identifier = model_identifier
|
||||
|
||||
# HF mode: llama-server downloads before becoming healthy — need longer timeout
|
||||
timeout = 600.0 if hf_repo else 120.0
|
||||
if not self._wait_for_health(timeout=timeout):
|
||||
# Wait for llama-server to become healthy
|
||||
if not self._wait_for_health(timeout=120.0):
|
||||
self._kill_process()
|
||||
raise RuntimeError(
|
||||
"llama-server failed to start. "
|
||||
|
|
@ -256,6 +404,9 @@ class LlamaCppBackend:
|
|||
logger.warning(f"Error killing llama-server process: {e}")
|
||||
finally:
|
||||
self._process = None
|
||||
if self._stdout_thread is not None:
|
||||
self._stdout_thread.join(timeout=2)
|
||||
self._stdout_thread = None
|
||||
|
||||
def _cleanup(self):
|
||||
"""atexit handler to ensure llama-server is terminated."""
|
||||
|
|
@ -273,8 +424,10 @@ class LlamaCppBackend:
|
|||
while time.monotonic() < deadline:
|
||||
# Check if process crashed
|
||||
if self._process.poll() is not None:
|
||||
# Read remaining output for error info
|
||||
output = self._process.stdout.read() if self._process.stdout else ""
|
||||
# Give the drain thread a moment to collect final output
|
||||
if self._stdout_thread is not None:
|
||||
self._stdout_thread.join(timeout=2)
|
||||
output = "\n".join(self._stdout_lines[-50:])
|
||||
logger.error(
|
||||
f"llama-server exited with code {self._process.returncode}. "
|
||||
f"Output: {output[:2000]}"
|
||||
|
|
|
|||
|
|
@ -125,6 +125,7 @@ async def load_model(
|
|||
# Local mode: llama-server loads via -m <path>
|
||||
success = llama_backend.load_model(
|
||||
gguf_path=config.gguf_file,
|
||||
mmproj_path=config.gguf_mmproj_file,
|
||||
model_identifier=config.identifier,
|
||||
is_vision=config.is_vision,
|
||||
n_ctx=request.max_seq_length,
|
||||
|
|
|
|||
|
|
@ -423,6 +423,11 @@ def safe_num_proc(desired: Optional[int] = None) -> int:
|
|||
"""
|
||||
Return a safe ``num_proc`` for ``dataset.map()`` calls.
|
||||
|
||||
On Windows, always returns 1 because Python uses ``spawn`` instead of
|
||||
``fork`` for multiprocessing — the overhead of re-importing torch,
|
||||
transformers, unsloth etc. per worker is typically slower than
|
||||
single-process for normal dataset sizes.
|
||||
|
||||
On multi-GPU machines the NVIDIA driver spawns extra background threads,
|
||||
making ``os.fork()`` prone to deadlocks when many workers are created.
|
||||
This helper caps ``num_proc`` to 4 on such machines.
|
||||
|
|
@ -438,6 +443,12 @@ def safe_num_proc(desired: Optional[int] = None) -> int:
|
|||
A safe integer ≥ 1.
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
|
||||
# Windows uses 'spawn' for multiprocessing — the overhead of re-importing
|
||||
# torch/transformers/unsloth per worker is typically slower than single-process.
|
||||
if sys.platform == "win32":
|
||||
return 1
|
||||
|
||||
if desired is None or not isinstance(desired, int):
|
||||
desired = max(1, os.cpu_count() // 3)
|
||||
|
|
|
|||
|
|
@ -422,6 +422,32 @@ def is_vision_model(model_name: str, hf_token: Optional[str] = None) -> bool:
|
|||
pass
|
||||
|
||||
|
||||
def _is_mmproj(filename: str) -> bool:
|
||||
"""Check if a GGUF filename is a vision projection (mmproj) file."""
|
||||
return "mmproj" in filename.lower()
|
||||
|
||||
|
||||
def detect_mmproj_file(path: str) -> Optional[str]:
|
||||
"""
|
||||
Find the mmproj (vision projection) GGUF file in a directory.
|
||||
|
||||
Args:
|
||||
path: Directory to search — or a .gguf file (uses its parent dir).
|
||||
|
||||
Returns:
|
||||
Full path to the mmproj .gguf file, or None if not found.
|
||||
"""
|
||||
p = Path(path)
|
||||
search_dir = p.parent if p.is_file() else p
|
||||
if not search_dir.is_dir():
|
||||
return None
|
||||
|
||||
for f in search_dir.glob("*.gguf"):
|
||||
if _is_mmproj(f.name):
|
||||
return str(f.resolve())
|
||||
return None
|
||||
|
||||
|
||||
def detect_gguf_model(path: str) -> Optional[str]:
|
||||
"""
|
||||
Check if the given local path is or contains a GGUF model file.
|
||||
|
|
@ -430,6 +456,9 @@ def detect_gguf_model(path: str) -> Optional[str]:
|
|||
1. path is a direct .gguf file path
|
||||
2. path is a directory containing .gguf files
|
||||
|
||||
Skips mmproj (vision projection) files — those must be passed via
|
||||
``--mmproj``, not ``-m``. Use :func:`detect_mmproj_file` instead.
|
||||
|
||||
Returns the full path to the .gguf file if found, None otherwise.
|
||||
For HuggingFace repo detection, use detect_gguf_model_remote() instead.
|
||||
"""
|
||||
|
|
@ -437,11 +466,16 @@ def detect_gguf_model(path: str) -> Optional[str]:
|
|||
|
||||
# Case 1: direct .gguf file
|
||||
if p.suffix == ".gguf" and p.is_file():
|
||||
if _is_mmproj(p.name):
|
||||
return None
|
||||
return str(p.resolve())
|
||||
|
||||
# Case 2: directory containing .gguf files
|
||||
# Case 2: directory containing .gguf files (skip mmproj)
|
||||
if p.is_dir():
|
||||
gguf_files = sorted(p.glob("*.gguf"), key=lambda f: f.stat().st_size, reverse=True)
|
||||
gguf_files = sorted(
|
||||
(f for f in p.glob("*.gguf") if not _is_mmproj(f.name)),
|
||||
key=lambda f: f.stat().st_size, reverse=True,
|
||||
)
|
||||
if gguf_files:
|
||||
return str(gguf_files[0].resolve())
|
||||
|
||||
|
|
@ -677,7 +711,8 @@ def scan_exported_models(exports_dir: str = "./exports") -> List[Tuple[str, str,
|
|||
continue
|
||||
|
||||
# Check for flat GGUF export (e.g. exports/gemma-3-4b-it-finetune-gguf/)
|
||||
gguf_files = list(run_dir.glob("*.gguf"))
|
||||
# Filter out mmproj (vision projection) files — they aren't loadable as main models
|
||||
gguf_files = [f for f in run_dir.glob("*.gguf") if not _is_mmproj(f.name)]
|
||||
if gguf_files:
|
||||
base_model = None
|
||||
export_meta = run_dir / "export_metadata.json"
|
||||
|
|
@ -907,6 +942,7 @@ class ModelConfig:
|
|||
is_lora: bool # Is this a lora adapter?
|
||||
is_gguf: bool = False # Is this a GGUF model?
|
||||
gguf_file: Optional[str] = None # Full path to the .gguf file (local mode)
|
||||
gguf_mmproj_file: Optional[str] = None # Full path to the mmproj .gguf file (vision projection)
|
||||
gguf_hf_repo: Optional[str] = None # HF repo ID for -hf mode (e.g. "unsloth/gemma-3-4b-it-GGUF")
|
||||
gguf_variant: Optional[str] = None # Quantization variant (e.g. "Q4_K_M")
|
||||
base_model: Optional[str] = None # Base model (for LoRAs)
|
||||
|
|
@ -1005,16 +1041,44 @@ class ModelConfig:
|
|||
if gguf_file:
|
||||
display_name = Path(gguf_file).stem
|
||||
logger.info(f"Detected local GGUF model: {gguf_file}")
|
||||
|
||||
# Detect vision: check if base model is vision, then look for mmproj
|
||||
mmproj_file = None
|
||||
gguf_is_vision = False
|
||||
gguf_dir = Path(gguf_file).parent
|
||||
|
||||
# Determine if this is a vision model from export metadata
|
||||
base_is_vision = False
|
||||
meta_path = gguf_dir / "export_metadata.json"
|
||||
if meta_path.exists():
|
||||
try:
|
||||
meta = json.loads(meta_path.read_text())
|
||||
base = meta.get("base_model")
|
||||
if base and is_vision_model(base, hf_token=hf_token):
|
||||
base_is_vision = True
|
||||
logger.info(f"GGUF base model '{base}' is a vision model")
|
||||
except Exception as e:
|
||||
logger.debug(f"Could not read export metadata: {e}")
|
||||
|
||||
# If vision (or mmproj happens to exist), find the mmproj file
|
||||
mmproj_file = detect_mmproj_file(gguf_file)
|
||||
if mmproj_file:
|
||||
gguf_is_vision = True
|
||||
logger.info(f"Detected mmproj for vision: {mmproj_file}")
|
||||
elif base_is_vision:
|
||||
logger.warning(f"Base model is vision but no mmproj file found in {gguf_dir}")
|
||||
|
||||
return cls(
|
||||
identifier=identifier,
|
||||
display_name=display_name,
|
||||
path=path,
|
||||
is_local=True,
|
||||
is_cached=True,
|
||||
is_vision=False,
|
||||
is_vision=gguf_is_vision,
|
||||
is_lora=False,
|
||||
is_gguf=True,
|
||||
gguf_file=gguf_file,
|
||||
gguf_mmproj_file=mmproj_file,
|
||||
)
|
||||
else:
|
||||
# Check if the HF repo contains GGUF files
|
||||
|
|
|
|||
|
|
@ -103,9 +103,6 @@ function ModelRow({
|
|||
{vramStatus === "tight" && (
|
||||
<span className="text-[9px] font-medium text-amber-400">TIGHT</span>
|
||||
)}
|
||||
{vramStatus === "fits" && (
|
||||
<span className="text-[9px] font-medium text-emerald-500/90">FIT</span>
|
||||
)}
|
||||
{meta ? (
|
||||
<span className="text-[10px] text-muted-foreground">{meta}</span>
|
||||
) : null}
|
||||
|
|
@ -253,9 +250,6 @@ function GgufVariantExpander({
|
|||
{fitStatus === "tight" && (
|
||||
<span className="text-[9px] font-medium text-amber-400">TIGHT</span>
|
||||
)}
|
||||
{fitStatus === "fits" && (
|
||||
<span className="text-[9px] font-medium text-emerald-500/90">FIT</span>
|
||||
)}
|
||||
<span className="text-[10px] text-muted-foreground">
|
||||
{formatBytes(v.size_bytes)}
|
||||
</span>
|
||||
|
|
|
|||
|
|
@ -24,7 +24,7 @@ import {
|
|||
import { HugeiconsIcon } from "@hugeicons/react";
|
||||
import { useTrainingRuntimeStore } from "@/features/training";
|
||||
import { Link, useRouterState } from "@tanstack/react-router";
|
||||
import { AnimatePresence, motion } from "motion/react";
|
||||
import { motion } from "motion/react";
|
||||
import { useState } from "react";
|
||||
import { TOUR_OPEN_EVENT } from "@/features/tour";
|
||||
|
||||
|
|
@ -58,9 +58,9 @@ export function Navbar() {
|
|||
|
||||
return (
|
||||
<header className="relative top-0 z-40 h-16 w-full">
|
||||
<div className="mx-auto flex h-full max-w-7xl items-center justify-between px-4 sm:px-6">
|
||||
<div className="mx-auto grid h-full max-w-7xl grid-cols-[1fr_auto_1fr] items-center px-4 sm:px-6">
|
||||
{/* Left: logo */}
|
||||
<Link to="/studio" className="flex items-center select-none">
|
||||
<Link to="/studio" className="flex items-center justify-self-start select-none">
|
||||
<img
|
||||
src="/blacklogo.png"
|
||||
alt="Unsloth"
|
||||
|
|
@ -117,23 +117,21 @@ export function Navbar() {
|
|||
/>
|
||||
)}
|
||||
<span className="relative z-10 flex items-center gap-1.5">
|
||||
<AnimatePresence mode="popLayout">
|
||||
{active && item.icon && (
|
||||
<motion.span
|
||||
key={item.href}
|
||||
initial={{ width: 0, opacity: 0 }}
|
||||
animate={{ width: "auto", opacity: 1 }}
|
||||
exit={{ width: 0, opacity: 0 }}
|
||||
transition={{ duration: 0.2, ease: [0.165, 0.84, 0.44, 1] }}
|
||||
className="overflow-hidden"
|
||||
>
|
||||
<HugeiconsIcon
|
||||
icon={item.icon}
|
||||
className="size-3.5 -mt-px"
|
||||
/>
|
||||
</motion.span>
|
||||
)}
|
||||
</AnimatePresence>
|
||||
<span className="inline-flex size-3.5 items-center justify-center overflow-hidden">
|
||||
<motion.span
|
||||
initial={false}
|
||||
animate={{
|
||||
opacity: active ? 1 : 0,
|
||||
scale: active ? 1 : 0.9,
|
||||
}}
|
||||
transition={{ duration: 0.2, ease: [0.165, 0.84, 0.44, 1] }}
|
||||
>
|
||||
<HugeiconsIcon
|
||||
icon={item.icon}
|
||||
className="size-3.5 -mt-px"
|
||||
/>
|
||||
</motion.span>
|
||||
</span>
|
||||
{item.label}
|
||||
</span>
|
||||
</Link>
|
||||
|
|
@ -142,7 +140,7 @@ export function Navbar() {
|
|||
</nav>
|
||||
|
||||
{/* Right: docs/tour desktop */}
|
||||
<div className="hidden items-center gap-2 md:flex">
|
||||
<div className="hidden items-center justify-self-end gap-2 md:flex">
|
||||
<AnimatedThemeToggler
|
||||
className="flex h-9 w-9 items-center justify-center rounded-md text-muted-foreground transition-colors hover:bg-accent hover:text-foreground [&_svg]:size-4"
|
||||
title="Toggle theme"
|
||||
|
|
@ -182,17 +180,20 @@ export function Navbar() {
|
|||
</HoverCardContent>
|
||||
</HoverCard>
|
||||
|
||||
{tourId ? (
|
||||
<button
|
||||
type="button"
|
||||
onClick={openTour}
|
||||
className="flex h-9 items-center gap-1.5 rounded-md px-3 text-muted-foreground transition-colors hover:bg-accent hover:text-foreground"
|
||||
title="Tour"
|
||||
>
|
||||
<HugeiconsIcon icon={CursorInfo02Icon} className="size-4" />
|
||||
<span className="text-sm font-medium">Tour</span>
|
||||
</button>
|
||||
) : null}
|
||||
<button
|
||||
type="button"
|
||||
onClick={tourId ? openTour : undefined}
|
||||
className={cn(
|
||||
"flex h-9 items-center gap-1.5 rounded-md px-3 text-muted-foreground transition-colors hover:bg-accent hover:text-foreground",
|
||||
!tourId && "invisible pointer-events-none",
|
||||
)}
|
||||
title="Tour"
|
||||
aria-hidden={!tourId}
|
||||
tabIndex={tourId ? 0 : -1}
|
||||
>
|
||||
<HugeiconsIcon icon={CursorInfo02Icon} className="size-4" />
|
||||
<span className="text-sm font-medium">Tour</span>
|
||||
</button>
|
||||
</div>
|
||||
|
||||
{/* Right: mobile */}
|
||||
|
|
|
|||
|
|
@ -33,11 +33,17 @@ import {
|
|||
import { MODEL_TYPE_TO_HF_TASK } from "@/config/training";
|
||||
import {
|
||||
useDebouncedValue,
|
||||
useGpuInfo,
|
||||
useHfModelSearch,
|
||||
useHfTokenValidation,
|
||||
useInfiniteScroll,
|
||||
} from "@/hooks";
|
||||
import { formatCompact } from "@/lib/utils";
|
||||
import {
|
||||
type TrainingMethod as VramTrainingMethod,
|
||||
type VramFitStatus,
|
||||
buildModelVramMap,
|
||||
} from "@/lib/vram";
|
||||
import { useTrainingConfigStore } from "@/features/training";
|
||||
import type { TrainingMethod } from "@/types/training";
|
||||
import {
|
||||
|
|
@ -50,6 +56,7 @@ import { useEffect, useMemo, useRef, useState } from "react";
|
|||
import { useShallow } from "zustand/react/shallow";
|
||||
|
||||
export function ModelSelectionStep() {
|
||||
const gpu = useGpuInfo();
|
||||
const {
|
||||
modelType,
|
||||
selectedModel,
|
||||
|
|
@ -93,6 +100,24 @@ export function ModelSelectionStep() {
|
|||
|
||||
const resultIds = useMemo(() => hfResults.map((r) => r.id), [hfResults]);
|
||||
|
||||
// Match Studio behavior: only show exception signals (OOM/TIGHT) in training flows.
|
||||
const vramMap = useMemo(() => {
|
||||
const fitMap = buildModelVramMap(
|
||||
hfResults,
|
||||
trainingMethod as VramTrainingMethod,
|
||||
gpu,
|
||||
);
|
||||
const map = new Map<string, { status: VramFitStatus | null; detail: string | null }>();
|
||||
for (const r of hfResults) {
|
||||
const fit = fitMap.get(r.id);
|
||||
map.set(r.id, {
|
||||
status: fit?.status ?? null,
|
||||
detail: r.totalParams ? formatCompact(r.totalParams) : null,
|
||||
});
|
||||
}
|
||||
return map;
|
||||
}, [hfResults, gpu, trainingMethod]);
|
||||
|
||||
const comboboxAnchorRef = useRef<HTMLDivElement>(null);
|
||||
const { scrollRef, sentinelRef } = useInfiniteScroll(
|
||||
fetchMore,
|
||||
|
|
@ -218,19 +243,21 @@ export function ModelSelectionStep() {
|
|||
>
|
||||
<ComboboxList className="p-1 !max-h-none !overflow-visible">
|
||||
{(id: string) => {
|
||||
const r = hfResults.find((r) => r.id === id);
|
||||
const sizeLabel = r?.totalParams
|
||||
? formatCompact(r.totalParams)
|
||||
: null;
|
||||
const entry = vramMap.get(id);
|
||||
const sizeLabel = entry?.detail ?? null;
|
||||
const fitStatus = entry?.status ?? null;
|
||||
const exceeds = fitStatus === "exceeds";
|
||||
return (
|
||||
<ComboboxItem
|
||||
key={id}
|
||||
value={id}
|
||||
className="justify-between"
|
||||
className={`justify-between ${exceeds ? "opacity-50" : ""}`}
|
||||
>
|
||||
<Tooltip>
|
||||
<TooltipTrigger asChild={true}>
|
||||
<span className="min-w-0 flex-1 truncate">
|
||||
<span
|
||||
className={`min-w-0 flex-1 truncate ${exceeds ? "line-through decoration-muted-foreground/50" : ""}`}
|
||||
>
|
||||
{id}
|
||||
</span>
|
||||
</TooltipTrigger>
|
||||
|
|
@ -241,11 +268,23 @@ export function ModelSelectionStep() {
|
|||
{id}
|
||||
</TooltipContent>
|
||||
</Tooltip>
|
||||
{sizeLabel ? (
|
||||
<span className="text-xs text-muted-foreground shrink-0">
|
||||
{sizeLabel}
|
||||
</span>
|
||||
) : null}
|
||||
<span className="flex items-center gap-1.5 shrink-0">
|
||||
{fitStatus === "exceeds" && (
|
||||
<span className="text-[9px] font-medium text-red-400">
|
||||
OOM
|
||||
</span>
|
||||
)}
|
||||
{fitStatus === "tight" && (
|
||||
<span className="text-[9px] font-medium text-amber-400">
|
||||
TIGHT
|
||||
</span>
|
||||
)}
|
||||
{sizeLabel ? (
|
||||
<span className="text-xs text-muted-foreground">
|
||||
{sizeLabel}
|
||||
</span>
|
||||
) : null}
|
||||
</span>
|
||||
</ComboboxItem>
|
||||
);
|
||||
}}
|
||||
|
|
|
|||
|
|
@ -37,8 +37,7 @@ import { formatCompact } from "@/lib/utils";
|
|||
import {
|
||||
type TrainingMethod as VramTrainingMethod,
|
||||
type VramFitStatus,
|
||||
checkVramFit,
|
||||
estimateLoadingVram,
|
||||
buildModelVramMap,
|
||||
} from "@/lib/vram";
|
||||
import {
|
||||
listLocalModels,
|
||||
|
|
@ -218,22 +217,23 @@ export function ModelSection() {
|
|||
// Keyed by model id so the render callback is a simple O(1) lookup.
|
||||
// Re-computes when the training method changes (QLoRA=4-bit vs LoRA/Full=fp16).
|
||||
const vramMap = useMemo(() => {
|
||||
const method = trainingMethod as VramTrainingMethod;
|
||||
const fitMap = buildModelVramMap(
|
||||
hfResults,
|
||||
trainingMethod as VramTrainingMethod,
|
||||
gpu,
|
||||
);
|
||||
const map = new Map<
|
||||
string,
|
||||
{ est: number; status: VramFitStatus | null; detail: string | null }
|
||||
>();
|
||||
for (const r of hfResults) {
|
||||
const detail = r.totalParams ? formatCompact(r.totalParams) : null;
|
||||
if (r.totalParams) {
|
||||
const est = estimateLoadingVram(r.totalParams, method);
|
||||
const status = gpu.available
|
||||
? checkVramFit(est, gpu.memoryTotalGb)
|
||||
: null;
|
||||
map.set(r.id, { est, status, detail });
|
||||
} else {
|
||||
map.set(r.id, { est: 0, status: null, detail });
|
||||
}
|
||||
const fit = fitMap.get(r.id);
|
||||
map.set(r.id, {
|
||||
est: fit?.est ?? 0,
|
||||
status: fit?.status ?? null,
|
||||
detail,
|
||||
});
|
||||
}
|
||||
return map;
|
||||
}, [hfResults, gpu, trainingMethod]);
|
||||
|
|
|
|||
|
|
@ -268,6 +268,7 @@
|
|||
}
|
||||
html {
|
||||
@apply font-sans;
|
||||
scrollbar-gutter: stable;
|
||||
}
|
||||
h1,
|
||||
h2,
|
||||
|
|
|
|||
|
|
@ -93,3 +93,32 @@ export function checkVramFit(
|
|||
if (ratio <= 1.0) return "tight";
|
||||
return "exceeds";
|
||||
}
|
||||
|
||||
export interface ModelVramMapInput {
|
||||
id: string;
|
||||
totalParams?: number;
|
||||
}
|
||||
|
||||
export interface ModelVramMapEntry {
|
||||
est: number;
|
||||
status: VramFitStatus | null;
|
||||
}
|
||||
|
||||
export function buildModelVramMap(
|
||||
models: ModelVramMapInput[],
|
||||
method: TrainingMethod,
|
||||
gpu: { available: boolean; memoryTotalGb: number },
|
||||
): Map<string, ModelVramMapEntry> {
|
||||
const map = new Map<string, ModelVramMapEntry>();
|
||||
for (const model of models) {
|
||||
if (!model.totalParams) {
|
||||
map.set(model.id, { est: 0, status: null });
|
||||
continue;
|
||||
}
|
||||
|
||||
const est = estimateLoadingVram(model.totalParams, method);
|
||||
const status = gpu.available ? checkVramFit(est, gpu.memoryTotalGb) : null;
|
||||
map.set(model.id, { est, status });
|
||||
}
|
||||
return map;
|
||||
}
|
||||
|
|
|
|||
234
test_llama_cpp.ps1
Normal file
234
test_llama_cpp.ps1
Normal file
|
|
@ -0,0 +1,234 @@
|
|||
<#
|
||||
.SYNOPSIS
|
||||
Test script for llama.cpp compilation and binary validation on Windows.
|
||||
Verifies that llama-server was built with CUDA support and can start.
|
||||
|
||||
.USAGE
|
||||
.\test_llama_cpp.ps1
|
||||
.\test_llama_cpp.ps1 -BinaryPath "C:\path\to\llama-server.exe"
|
||||
#>
|
||||
param(
|
||||
[string]$BinaryPath = ""
|
||||
)
|
||||
|
||||
$ErrorActionPreference = "Continue"
|
||||
|
||||
Write-Host ""
|
||||
Write-Host "============================================" -ForegroundColor Cyan
|
||||
Write-Host " llama.cpp Windows Build Test" -ForegroundColor Cyan
|
||||
Write-Host "============================================" -ForegroundColor Cyan
|
||||
Write-Host ""
|
||||
|
||||
# -- Step 1: Locate the binary ------------------------------------------
|
||||
Write-Host "1. Locating llama-server binary..." -ForegroundColor Yellow
|
||||
|
||||
$SearchPaths = @()
|
||||
|
||||
if ($BinaryPath) {
|
||||
$SearchPaths += $BinaryPath
|
||||
}
|
||||
|
||||
# Add all known locations
|
||||
$RepoRoot = $PSScriptRoot
|
||||
$SearchPaths += Join-Path $RepoRoot "llama.cpp\build\bin\Release\llama-server.exe"
|
||||
$SearchPaths += Join-Path $RepoRoot "llama.cpp\build\bin\llama-server.exe"
|
||||
# Legacy: older setup.ps1 built under ~/.unsloth
|
||||
$SearchPaths += Join-Path $env:USERPROFILE ".unsloth\llama.cpp\build\bin\Release\llama-server.exe"
|
||||
|
||||
# Check LLAMA_SERVER_PATH env var
|
||||
$envPath = $env:LLAMA_SERVER_PATH
|
||||
if ($envPath) {
|
||||
$SearchPaths = @($envPath) + $SearchPaths
|
||||
}
|
||||
|
||||
# Also check system PATH
|
||||
$systemPath = (Get-Command llama-server -ErrorAction SilentlyContinue)
|
||||
if ($systemPath) {
|
||||
$SearchPaths += $systemPath.Source
|
||||
}
|
||||
|
||||
$FoundBinary = $null
|
||||
foreach ($p in $SearchPaths) {
|
||||
if (Test-Path $p) {
|
||||
$FoundBinary = $p
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
if (-not $FoundBinary) {
|
||||
Write-Host " [FAIL] llama-server.exe not found!" -ForegroundColor Red
|
||||
Write-Host ""
|
||||
Write-Host " Searched locations:" -ForegroundColor Gray
|
||||
foreach ($p in $SearchPaths) {
|
||||
Write-Host " - $p" -ForegroundColor Gray
|
||||
}
|
||||
Write-Host ""
|
||||
Write-Host " To fix: Run setup.bat to build llama.cpp, or set:" -ForegroundColor Yellow
|
||||
Write-Host ' $env:LLAMA_SERVER_PATH = "C:\path\to\llama-server.exe"' -ForegroundColor Yellow
|
||||
exit 1
|
||||
}
|
||||
|
||||
Write-Host " [OK] Found: $FoundBinary" -ForegroundColor Green
|
||||
|
||||
# -- Step 2: Check file info --------------------------------------------
|
||||
Write-Host ""
|
||||
Write-Host "2. Binary info..." -ForegroundColor Yellow
|
||||
|
||||
$fileInfo = Get-Item $FoundBinary
|
||||
$sizeMB = [math]::Round($fileInfo.Length / 1MB, 1)
|
||||
Write-Host " Size: $sizeMB MB" -ForegroundColor Gray
|
||||
Write-Host " Modified: $($fileInfo.LastWriteTime)" -ForegroundColor Gray
|
||||
|
||||
# -- Step 3: Check for CUDA symbols ------------------------------------
|
||||
Write-Host ""
|
||||
Write-Host "3. Checking for CUDA support..." -ForegroundColor Yellow
|
||||
|
||||
# Run with --help or -v and capture output to check for CUDA indicators
|
||||
$helpOutput = & $FoundBinary --version 2>&1 | Out-String
|
||||
if (-not $helpOutput) {
|
||||
$helpOutput = ""
|
||||
}
|
||||
|
||||
# Check binary dependencies for CUDA DLLs using dumpbin if available
|
||||
$dumpbin = (Get-Command dumpbin -ErrorAction SilentlyContinue)
|
||||
$hasCudaDlls = $false
|
||||
|
||||
if ($dumpbin) {
|
||||
$deps = & dumpbin /dependents $FoundBinary 2>&1 | Out-String
|
||||
if ($deps -match "cudart|cublas|cublasLt|nvcuda") {
|
||||
$hasCudaDlls = $true
|
||||
Write-Host " [OK] CUDA DLLs found in dependencies (dumpbin)" -ForegroundColor Green
|
||||
# Extract CUDA DLL names
|
||||
$cudaDlls = ($deps -split "`n") | Where-Object { $_ -match "cuda|cublas|nvcuda" } | ForEach-Object { $_.Trim() }
|
||||
foreach ($dll in $cudaDlls) {
|
||||
if ($dll) { Write-Host " - $dll" -ForegroundColor Gray }
|
||||
}
|
||||
} else {
|
||||
Write-Host " [WARN] No CUDA DLLs found in dependencies!" -ForegroundColor Red
|
||||
Write-Host " This binary was likely compiled WITHOUT -DGGML_CUDA=ON" -ForegroundColor Red
|
||||
}
|
||||
} else {
|
||||
# Fallback: check file size (CUDA builds are typically > 50MB)
|
||||
if ($sizeMB -gt 40) {
|
||||
Write-Host " [LIKELY OK] Binary is $sizeMB MB (CUDA builds are typically > 50MB)" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host " [WARN] Binary is only $sizeMB MB (CPU-only builds are typically < 30MB)" -ForegroundColor Yellow
|
||||
Write-Host " dumpbin not available for detailed check. Install VS Build Tools." -ForegroundColor Gray
|
||||
}
|
||||
}
|
||||
|
||||
# -- Step 4: Quick startup test -----------------------------------------
|
||||
Write-Host ""
|
||||
Write-Host "4. Running startup test (will start and immediately stop)..." -ForegroundColor Yellow
|
||||
|
||||
# Start llama-server on a random port with no model -- just check it initializes
|
||||
$testPort = Get-Random -Minimum 49152 -Maximum 65535
|
||||
$proc = $null
|
||||
|
||||
try {
|
||||
$proc = Start-Process -FilePath $FoundBinary `
|
||||
-ArgumentList "--port", $testPort, "--host", "127.0.0.1" `
|
||||
-PassThru -NoNewWindow -RedirectStandardError "$env:TEMP\llama_test_stderr.txt" `
|
||||
-RedirectStandardOutput "$env:TEMP\llama_test_stdout.txt"
|
||||
|
||||
# Give it 3 seconds to start
|
||||
Start-Sleep -Seconds 3
|
||||
|
||||
# Check if it crashed
|
||||
if ($proc.HasExited) {
|
||||
$exitCode = $proc.ExitCode
|
||||
$stderr = ""
|
||||
if (Test-Path "$env:TEMP\llama_test_stderr.txt") {
|
||||
$stderr = Get-Content "$env:TEMP\llama_test_stderr.txt" -Raw
|
||||
}
|
||||
$stdout = ""
|
||||
if (Test-Path "$env:TEMP\llama_test_stdout.txt") {
|
||||
$stdout = Get-Content "$env:TEMP\llama_test_stdout.txt" -Raw
|
||||
}
|
||||
|
||||
$allOutput = "$stdout`n$stderr"
|
||||
|
||||
if ($allOutput -match "failed to initialize CUDA") {
|
||||
Write-Host " [FAIL] CUDA initialization failed!" -ForegroundColor Red
|
||||
Write-Host " The binary was compiled without CUDA support or CUDA drivers are missing." -ForegroundColor Red
|
||||
Write-Host ""
|
||||
Write-Host " Rebuild with: cmake -DGGML_CUDA=ON ..." -ForegroundColor Yellow
|
||||
} elseif ($allOutput -match "HTTPS is not supported") {
|
||||
# This is expected when LLAMA_CURL=OFF -- not a real failure
|
||||
Write-Host " [OK] Binary started (HTTPS warning is expected -- we use local files)" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host " [WARN] Process exited with code $exitCode" -ForegroundColor Yellow
|
||||
}
|
||||
|
||||
if ($allOutput.Trim()) {
|
||||
Write-Host ""
|
||||
Write-Host " --- Output ---" -ForegroundColor Gray
|
||||
$allOutput.Trim().Split("`n") | ForEach-Object { Write-Host " $_" -ForegroundColor Gray }
|
||||
}
|
||||
} else {
|
||||
Write-Host " [OK] llama-server started successfully on port $testPort" -ForegroundColor Green
|
||||
|
||||
# Check CUDA detection from startup output
|
||||
Start-Sleep -Seconds 1
|
||||
$stderr = ""
|
||||
if (Test-Path "$env:TEMP\llama_test_stderr.txt") {
|
||||
$stderr = Get-Content "$env:TEMP\llama_test_stderr.txt" -Raw
|
||||
}
|
||||
|
||||
if ($stderr -match "CUDA") {
|
||||
if ($stderr -match "failed to initialize CUDA") {
|
||||
Write-Host " [FAIL] CUDA init failed at runtime!" -ForegroundColor Red
|
||||
} else {
|
||||
Write-Host " [OK] CUDA detected at runtime" -ForegroundColor Green
|
||||
}
|
||||
}
|
||||
|
||||
# Kill it
|
||||
Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue
|
||||
Write-Host " Stopped test server." -ForegroundColor Gray
|
||||
}
|
||||
} catch {
|
||||
Write-Host " [ERROR] Could not start llama-server: $_" -ForegroundColor Red
|
||||
} finally {
|
||||
if ($proc -and -not $proc.HasExited) {
|
||||
Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue
|
||||
}
|
||||
Remove-Item "$env:TEMP\llama_test_stderr.txt" -ErrorAction SilentlyContinue
|
||||
Remove-Item "$env:TEMP\llama_test_stdout.txt" -ErrorAction SilentlyContinue
|
||||
}
|
||||
|
||||
# -- Step 5: Check llama-quantize ---------------------------------------
|
||||
Write-Host ""
|
||||
Write-Host "5. Checking llama-quantize..." -ForegroundColor Yellow
|
||||
|
||||
$quantizePath = Join-Path (Split-Path $FoundBinary) "llama-quantize.exe"
|
||||
if (Test-Path $quantizePath) {
|
||||
$qSize = [math]::Round((Get-Item $quantizePath).Length / 1MB, 1)
|
||||
Write-Host " [OK] Found: $quantizePath ($qSize MB)" -ForegroundColor Green
|
||||
} else {
|
||||
Write-Host " [WARN] llama-quantize.exe not found alongside llama-server" -ForegroundColor Yellow
|
||||
Write-Host " GGUF export/quantization won't work without it" -ForegroundColor Yellow
|
||||
}
|
||||
|
||||
# -- Summary ------------------------------------------------------------
|
||||
Write-Host ""
|
||||
Write-Host "============================================" -ForegroundColor Cyan
|
||||
Write-Host " Summary" -ForegroundColor Cyan
|
||||
Write-Host "============================================" -ForegroundColor Cyan
|
||||
Write-Host " Binary: $FoundBinary" -ForegroundColor Gray
|
||||
Write-Host " Size: $sizeMB MB" -ForegroundColor Gray
|
||||
if ($hasCudaDlls) {
|
||||
Write-Host " CUDA: YES (confirmed via DLL deps)" -ForegroundColor Green
|
||||
} elseif ($sizeMB -gt 40) {
|
||||
Write-Host " CUDA: LIKELY (large binary size)" -ForegroundColor Yellow
|
||||
} else {
|
||||
Write-Host " CUDA: NO (rebuild with -DGGML_CUDA=ON)" -ForegroundColor Red
|
||||
}
|
||||
Write-Host ""
|
||||
|
||||
if (-not $hasCudaDlls -and $sizeMB -le 40) {
|
||||
Write-Host "To rebuild with CUDA:" -ForegroundColor Yellow
|
||||
Write-Host ' 1. Delete the build dir: Remove-Item -Recurse -Force "$env:USERPROFILE\.unsloth\llama.cpp\build"' -ForegroundColor Gray
|
||||
Write-Host ' 2. Re-run: .\setup.bat' -ForegroundColor Gray
|
||||
Write-Host ""
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue