From 3521de7040573a8253306dae3ca1d5abcb283383 Mon Sep 17 00:00:00 2001
From: Roland Tannous <rolandtannous@gravityq.ai>
Date: Sat, 28 Feb 2026 07:45:40 +0000
Subject: [PATCH] Build llama.cpp in-tree, auto-detect driver CUDA version for
 compatible toolkit

---
 setup.ps1                                  | 7 +++----
 studio/backend/core/inference/llama_cpp.py | 6 +++---
 test_llama_cpp.ps1                         | 1 +
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/setup.ps1 b/setup.ps1
index c213dcbf50..1269467368 100644
--- a/setup.ps1
+++ b/setup.ps1
@@ -556,12 +556,13 @@ $ErrorActionPreference = $prevEAP
 # ==========================================================================
 #  PHASE 4: Build llama.cpp with CUDA for GGUF inference + export
 # ==========================================================================
-# Builds at ~/.unsloth/llama.cpp/ (persistent across pip upgrades).
+# Builds in-tree at $REPO/llama.cpp/ (same as setup.sh on Linux).
+# This directory is already in .gitignore.
 # We build:
 #   - llama-server:   for GGUF model inference
 #   - llama-quantize: for GGUF export quantization
 # Prerequisites (git, cmake, VS Build Tools, CUDA Toolkit) already installed in Phase 1.
-$LlamaCppDir = Join-Path $env:USERPROFILE ".unsloth\llama.cpp"
+$LlamaCppDir = Join-Path $PSScriptRoot "llama.cpp"
 $BuildDir = Join-Path $LlamaCppDir "build"
 $LlamaServerBin = Join-Path $BuildDir "bin\Release\llama-server.exe"
 
@@ -588,8 +589,6 @@ if (Test-Path $LlamaServerBin) {
     $FailedStep = ""
 
     # -- Step A: Clone or pull llama.cpp --
-    $UnslothDir = Join-Path $env:USERPROFILE ".unsloth"
-    if (-not (Test-Path $UnslothDir)) { New-Item -ItemType Directory -Path $UnslothDir -Force | Out-Null }
 
     if (Test-Path (Join-Path $LlamaCppDir ".git")) {
         Write-Host "   llama.cpp repo already cloned, pulling latest..." -ForegroundColor Gray
diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index cf4494df30..7ae9bd142a 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -100,13 +100,13 @@ class LlamaCppBackend:
         if build_path.is_file():
             return str(build_path)
 
-        # 3. Windows MSVC build (setup.ps1 builds here — Release config)
+        # 3. Windows MSVC build (Release config, in-tree — matches setup.ps1)
         if sys.platform == "win32":
-            # In-tree
+            # In-tree (primary — setup.ps1 now builds here)
             win_path = project_root / "llama.cpp" / "build" / "bin" / "Release" / binary_name
             if win_path.is_file():
                 return str(win_path)
-            # ~/.unsloth (setup.ps1 default location)
+            # Legacy: ~/.unsloth (older setup.ps1 versions built here)
             home_path = Path.home() / ".unsloth" / "llama.cpp" / "build" / "bin" / "Release" / binary_name
             if home_path.is_file():
                 return str(home_path)
diff --git a/test_llama_cpp.ps1 b/test_llama_cpp.ps1
index 6e16057640..b98177eac6 100644
--- a/test_llama_cpp.ps1
+++ b/test_llama_cpp.ps1
@@ -32,6 +32,7 @@ if ($BinaryPath) {
 $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