chore: comment out llama.cpp build in setup.sh

Allow users to set UNSLOTH_STUDIO_HOME to relocate all studio data
(venvs, assets, outputs, exports, auth, cache, tensorboard runs).
Defaults to ~/.unsloth/studio when unset (no behaviour change).

cli/commands/studio.py

@@ -12,7 +12,18 @@ import typer
 
 studio_app = typer.Typer(help = "Unsloth Studio commands.")
 
-STUDIO_HOME = Path.home() / ".unsloth" / "studio"
+
+def _studio_home() -> Path:
+    """Studio root: env var > config file > default."""
+    custom = os.environ.get("UNSLOTH_STUDIO_HOME")
+    if custom:
+        return Path(custom).expanduser().resolve()
+    conf = Path.home() / ".unsloth" / "studio_home"
+    if conf.is_file():
+        saved = conf.read_text().strip()
+        if saved:
+            return Path(saved).expanduser().resolve()
+    return Path.home() / ".unsloth" / "studio"
 
 # __file__ is cli/commands/studio.py — two parents up is the package root
 # (either site-packages or the repo root for editable installs).
@@ -22,9 +33,9 @@ _PACKAGE_ROOT = Path(__file__).resolve().parent.parent.parent
 def _studio_venv_python() -> Optional[Path]:
     """Return the studio venv Python binary, or None if not set up."""
     if platform.system() == "Windows":
-        p = STUDIO_HOME / ".venv" / "Scripts" / "python.exe"
+        p = _studio_home() / ".venv" / "Scripts" / "python.exe"
     else:
-        p = STUDIO_HOME / ".venv" / "bin" / "python"
+        p = _studio_home() / ".venv" / "bin" / "python"
     return p if p.is_file() else None
 
 
@@ -44,7 +55,7 @@ def _find_run_py() -> Optional[Path]:
         "lib/python*/site-packages/studio/backend/run.py",
         "Lib/site-packages/studio/backend/run.py",
     ):
-        for match in (STUDIO_HOME / ".venv").glob(pattern):
+        for match in (_studio_home() / ".venv").glob(pattern):
             return match
     return None
 
@@ -64,7 +75,7 @@ def _find_setup_script() -> Optional[Path]:
         f"lib/python*/site-packages/studio/{name}",
         f"Lib/site-packages/studio/{name}",
     ):
-        for match in (STUDIO_HOME / ".venv").glob(pattern):
+        for match in (_studio_home() / ".venv").glob(pattern):
             return match
     return None
 
@@ -85,7 +96,7 @@ def studio_default(
         return
 
     # Always use the studio venv if it exists and we're not already in it
-    studio_venv_dir = STUDIO_HOME / ".venv"
+    studio_venv_dir = _studio_home() / ".venv"
     in_studio_venv = sys.prefix.startswith(str(studio_venv_dir))
 
     if not in_studio_venv:

studio/backend/core/export/worker.py

@@ -44,9 +44,8 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        venv_t5 = os.path.join(
-            os.path.expanduser("~"), ".unsloth", "studio", ".venv_t5"
-        )
+        from utils.paths.storage_roots import venv_t5_root
+        venv_t5 = str(venv_t5_root())
         if os.path.isdir(venv_t5):
             sys.path.insert(0, venv_t5)
             logger.info("Activated transformers 5.x from %s", venv_t5)

studio/backend/core/inference/worker.py

@@ -46,9 +46,8 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        venv_t5 = os.path.join(
-            os.path.expanduser("~"), ".unsloth", "studio", ".venv_t5"
-        )
+        from utils.paths.storage_roots import venv_t5_root
+        venv_t5 = str(venv_t5_root())
         if os.path.isdir(venv_t5):
             sys.path.insert(0, venv_t5)
             logger.info("Activated transformers 5.x from %s", venv_t5)

studio/backend/core/training/worker.py

@@ -40,9 +40,8 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        venv_t5 = os.path.join(
-            os.path.expanduser("~"), ".unsloth", "studio", ".venv_t5"
-        )
+        from utils.paths.storage_roots import venv_t5_root
+        venv_t5 = str(venv_t5_root())
         if os.path.isdir(venv_t5):
             sys.path.insert(0, venv_t5)
             logger.info("Activated transformers 5.x from %s", venv_t5)

studio/backend/utils/models/model_config.py

@@ -427,7 +427,8 @@ _VLM_MODEL_TYPES = {
 }
 
 # Pre-computed .venv_t5 path and backend dir for subprocess version switching.
-_VENV_T5_DIR = str(Path.home() / ".unsloth" / "studio" / ".venv_t5")
+from utils.paths.storage_roots import venv_t5_root
+_VENV_T5_DIR = str(venv_t5_root())
 _BACKEND_DIR = str(Path(__file__).resolve().parent.parent.parent)
 
 # Inline script executed in a subprocess with transformers 5.x activated.

studio/backend/utils/paths/__init__.py

@@ -8,6 +8,7 @@ Path utilities for model and dataset handling
 from .path_utils import normalize_path, is_local_path, is_model_cached, get_cache_path
 from .storage_roots import (
     studio_root,
+    venv_t5_root,
     assets_root,
     datasets_root,
     dataset_uploads_root,
@@ -36,6 +37,7 @@ __all__ = [
     "is_model_cached",
     "get_cache_path",
     "studio_root",
+    "venv_t5_root",
     "assets_root",
     "datasets_root",
     "dataset_uploads_root",

studio/backend/utils/paths/storage_roots.py

@@ -9,12 +9,26 @@ import tempfile
 
 
 def studio_root() -> Path:
+    """Studio root: env var > config file > default."""
+    custom = os.environ.get("UNSLOTH_STUDIO_HOME")
+    if custom:
+        return Path(custom).expanduser().resolve()
+    conf = Path.home() / ".unsloth" / "studio_home"
+    if conf.is_file():
+        saved = conf.read_text().strip()
+        if saved:
+            return Path(saved).expanduser().resolve()
     return Path.home() / ".unsloth" / "studio"
 
 
+def venv_t5_root() -> Path:
+    """Pre-installed transformers 5.x directory, respects UNSLOTH_STUDIO_HOME."""
+    return studio_root() / ".venv_t5"
+
+
 def cache_root() -> Path:
     """Central cache directory for all studio downloads (models, datasets, etc.)."""
-    return Path.home() / ".unsloth" / "studio" / "cache"
+    return studio_root() / "cache"
 
 
 def assets_root() -> Path:

studio/backend/utils/transformers_version.py

@@ -61,7 +61,8 @@ TRANSFORMERS_5_VERSION = "5.3.0"
 TRANSFORMERS_DEFAULT_VERSION = "4.57.1"
 
 # Pre-installed directory for transformers 5.x — created by setup.sh / setup.ps1
-_VENV_T5_DIR = str(Path.home() / ".unsloth" / "studio" / ".venv_t5")
+from utils.paths.storage_roots import venv_t5_root
+_VENV_T5_DIR = str(venv_t5_root())
 
 
 def _resolve_base_model(model_name: str) -> str:

studio/setup.ps1

@@ -826,9 +826,14 @@ if (-not $PythonCmd) {
 
 Write-Host "[OK] Using $PythonCmd ($(& $PythonCmd --version 2>&1))" -ForegroundColor Green
 
-# Always create a .venv for isolation -- even for pip installs.
-# Created in the repo root (parent of studio/).
-$VenvDir = Join-Path $env:USERPROFILE ".unsloth\studio\.venv"
+# ── Studio home (configurable via UNSLOTH_STUDIO_HOME) ──
+$StudioHome = if ($env:UNSLOTH_STUDIO_HOME) { $env:UNSLOTH_STUDIO_HOME } else { Join-Path $env:USERPROFILE ".unsloth\studio" }
+# Persist for future `unsloth studio` runs (survives shell restarts)
+$UnslothDir = Join-Path $env:USERPROFILE ".unsloth"
+if (-not (Test-Path $UnslothDir)) { New-Item -ItemType Directory -Path $UnslothDir -Force | Out-Null }
+Set-Content -Path (Join-Path $UnslothDir "studio_home") -Value $StudioHome -NoNewline
+
+$VenvDir = Join-Path $StudioHome ".venv"
 if (-not (Test-Path $VenvDir)) {
     Write-Host "   Creating virtual environment at $VenvDir..." -ForegroundColor Cyan
     & $PythonCmd -m venv $VenvDir
@@ -901,7 +906,7 @@ $ErrorActionPreference = $prevEAP
 # The training subprocess just prepends .venv_t5/ to sys.path — instant switch.
 Write-Host ""
 Write-Host "   Pre-installing transformers 5.x for newer model support..." -ForegroundColor Cyan
-$VenvT5Dir = Join-Path $env:USERPROFILE ".unsloth\studio\.venv_t5"
+$VenvT5Dir = Join-Path $StudioHome ".venv_t5"
 if (Test-Path $VenvT5Dir) { Remove-Item -Recurse -Force $VenvT5Dir }
 New-Item -ItemType Directory -Path $VenvT5Dir -Force | Out-Null
 $prevEAP_t5 = $ErrorActionPreference

studio/setup.sh

@@ -227,8 +227,13 @@ if [ "$IS_COLAB" = true ]; then
     # Colab: install packages directly without venv
     install_python_stack
 else
-    # Local: create venv under ~/.unsloth/studio/ (shared location, not in repo)
-    STUDIO_HOME="$HOME/.unsloth/studio"
+    # Local: create venv under studio home (shared location, not in repo)
+    # Configurable via UNSLOTH_STUDIO_HOME; defaults to ~/.unsloth/studio
+    STUDIO_HOME="${UNSLOTH_STUDIO_HOME:-$HOME/.unsloth/studio}"
+    echo "   Studio home: $STUDIO_HOME"
+    # Persist for future `unsloth studio` runs (survives shell restarts)
+    mkdir -p "$HOME/.unsloth"
+    echo "$STUDIO_HOME" > "$HOME/.unsloth/studio_home"
     VENV_DIR="$STUDIO_HOME/.venv"
     VENV_T5_DIR="$STUDIO_HOME/.venv_t5"
     mkdir -p "$STUDIO_HOME"
@@ -270,133 +275,136 @@ else
 fi
 
 # ── 8. Build llama.cpp binaries 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).
-#   - llama-server: for GGUF model inference
-#   - llama-quantize: for GGUF export quantization (symlinked to root for check_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 "$LLAMA_CPP_DIR"
-{
-    # Check prerequisites
-    if ! command -v cmake &>/dev/null; then
-        echo ""
-        echo "⚠️  cmake not found — skipping llama-server build (GGUF inference won't be available)"
-        echo "   Install cmake and re-run setup.sh to enable GGUF inference."
-    elif ! command -v git &>/dev/null; then
-        echo ""
-        echo "⚠️  git not found — skipping llama-server build (GGUF inference won't be available)"
-    else
-        echo ""
-        echo "Building llama-server for GGUF inference..."
-
-        BUILD_OK=true
-        run_quiet "clone llama.cpp" git clone --depth 1 https://github.com/ggml-org/llama.cpp.git "$LLAMA_CPP_DIR" || BUILD_OK=false
-
-        if [ "$BUILD_OK" = true ]; then
-            # Skip tests/examples we don't need (faster build)
-            CMAKE_ARGS="-DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_NATIVE=ON"
-
-            # Use ccache if available (dramatically faster rebuilds)
-            if command -v ccache &>/dev/null; then
-                CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache"
-                echo "   Using ccache for faster compilation"
-            fi
-
-            # Detect CUDA: check nvcc on PATH, then common install locations
-            NVCC_PATH=""
-            if command -v nvcc &>/dev/null; then
-                NVCC_PATH="$(command -v nvcc)"
-            elif [ -x /usr/local/cuda/bin/nvcc ]; then
-                NVCC_PATH="/usr/local/cuda/bin/nvcc"
-                export PATH="/usr/local/cuda/bin:$PATH"
-            elif ls /usr/local/cuda-*/bin/nvcc &>/dev/null 2>&1; then
-                # Pick the newest cuda-XX.X directory
-                NVCC_PATH="$(ls -d /usr/local/cuda-*/bin/nvcc 2>/dev/null | sort -V | tail -1)"
-                export PATH="$(dirname "$NVCC_PATH"):$PATH"
-            fi
-
-            if [ -n "$NVCC_PATH" ]; then
-                echo "   Building with CUDA support (nvcc: $NVCC_PATH)..."
-                CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON"
-
-                # Detect GPU compute capability and limit CUDA architectures
-                # Without this, cmake builds for ALL default archs (very slow)
-                CUDA_ARCHS=""
-                if command -v nvidia-smi &>/dev/null; then
-                    # Read all GPUs, deduplicate (handles mixed-GPU hosts)
-                    _raw_caps=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
-                    while IFS= read -r _cap; do
-                        _cap=$(echo "$_cap" | tr -d '[:space:]')
-                        if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
-                            _arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
-                            # Append if not already present
-                            case ";$CUDA_ARCHS;" in
-                                *";$_arch;"*) ;;
-                                *) CUDA_ARCHS="${CUDA_ARCHS:+$CUDA_ARCHS;}$_arch" ;;
-                            esac
-                        fi
-                    done <<< "$_raw_caps"
-                fi
-
-                if [ -n "$CUDA_ARCHS" ]; then
-                    echo "   GPU compute capabilities: ${CUDA_ARCHS//;/, } -- limiting build to detected archs"
-                    CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
-                else
-                    echo "   Could not detect GPU arch -- building for all default CUDA architectures (slower)"
-                fi
-
-                # Multi-threaded nvcc compilation (uses all CPU cores per .cu file)
-                CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
-            elif [ -d /usr/local/cuda ] || nvidia-smi &>/dev/null; then
-                echo "   CUDA driver detected but nvcc not found — building CPU-only"
-                echo "   To enable GPU: install cuda-toolkit or add nvcc to PATH"
-            else
-                echo "   Building CPU-only (no CUDA detected)..."
-            fi
-
-            NCPU=$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)
-
-            # Use Ninja if available (faster parallel builds than Make)
-            CMAKE_GENERATOR_ARGS=""
-            if command -v ninja &>/dev/null; then
-                CMAKE_GENERATOR_ARGS="-G Ninja"
-            fi
-
-            run_quiet "cmake llama.cpp" cmake $CMAKE_GENERATOR_ARGS -S "$LLAMA_CPP_DIR" -B "$LLAMA_CPP_DIR/build" $CMAKE_ARGS || BUILD_OK=false
-        fi
-
-        if [ "$BUILD_OK" = true ]; then
-            run_quiet "build llama-server" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-server -j"$NCPU" || BUILD_OK=false
-        fi
-
-        # Also build llama-quantize (needed by unsloth-zoo's GGUF export pipeline)
-        if [ "$BUILD_OK" = true ]; then
-            run_quiet "build llama-quantize" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-quantize -j"$NCPU" || true
-            # Symlink to llama.cpp root — check_llama_cpp() looks for the binary there
-            QUANTIZE_BIN="$LLAMA_CPP_DIR/build/bin/llama-quantize"
-            if [ -f "$QUANTIZE_BIN" ]; then
-                ln -sf build/bin/llama-quantize "$LLAMA_CPP_DIR/llama-quantize"
-            fi
-        fi
-
-        if [ "$BUILD_OK" = true ]; then
-            if [ -f "$LLAMA_SERVER_BIN" ]; then
-                echo "✅ llama-server built at $LLAMA_SERVER_BIN"
-            else
-                echo "⚠️  llama-server binary not found after build — GGUF inference won't be available"
-            fi
-            if [ -f "$LLAMA_CPP_DIR/llama-quantize" ]; then
-                echo "✅ llama-quantize available for GGUF export"
-            fi
-        else
-            echo "⚠️  llama-server build failed — GGUF inference won't be available, but everything else works"
-        fi
-    fi
-}
+# Disabled: llama.cpp build is commented out for now.
+# UNCOMMENT the block below to re-enable.
+#
+# # 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())
+# 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 "$LLAMA_CPP_DIR"
+# {
+#     # Check prerequisites
+#     if ! command -v cmake &>/dev/null; then
+#         echo ""
+#         echo "⚠️  cmake not found — skipping llama-server build (GGUF inference won't be available)"
+#         echo "   Install cmake and re-run setup.sh to enable GGUF inference."
+#     elif ! command -v git &>/dev/null; then
+#         echo ""
+#         echo "⚠️  git not found — skipping llama-server build (GGUF inference won't be available)"
+#     else
+#         echo ""
+#         echo "Building llama-server for GGUF inference..."
+#
+#         BUILD_OK=true
+#         run_quiet "clone llama.cpp" git clone --depth 1 https://github.com/ggml-org/llama.cpp.git "$LLAMA_CPP_DIR" || BUILD_OK=false
+#
+#         if [ "$BUILD_OK" = true ]; then
+#             # Skip tests/examples we don't need (faster build)
+#             CMAKE_ARGS="-DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_NATIVE=ON"
+#
+#             # Use ccache if available (dramatically faster rebuilds)
+#             if command -v ccache &>/dev/null; then
+#                 CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache"
+#                 echo "   Using ccache for faster compilation"
+#             fi
+#
+#             # Detect CUDA: check nvcc on PATH, then common install locations
+#             NVCC_PATH=""
+#             if command -v nvcc &>/dev/null; then
+#                 NVCC_PATH="$(command -v nvcc)"
+#             elif [ -x /usr/local/cuda/bin/nvcc ]; then
+#                 NVCC_PATH="/usr/local/cuda/bin/nvcc"
+#                 export PATH="/usr/local/cuda/bin:$PATH"
+#             elif ls /usr/local/cuda-*/bin/nvcc &>/dev/null 2>&1; then
+#                 # Pick the newest cuda-XX.X directory
+#                 NVCC_PATH="$(ls -d /usr/local/cuda-*/bin/nvcc 2>/dev/null | sort -V | tail -1)"
+#                 export PATH="$(dirname "$NVCC_PATH"):$PATH"
+#             fi
+#
+#             if [ -n "$NVCC_PATH" ]; then
+#                 echo "   Building with CUDA support (nvcc: $NVCC_PATH)..."
+#                 CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON"
+#
+#                 # Detect GPU compute capability and limit CUDA architectures
+#                 # Without this, cmake builds for ALL default archs (very slow)
+#                 CUDA_ARCHS=""
+#                 if command -v nvidia-smi &>/dev/null; then
+#                     # Read all GPUs, deduplicate (handles mixed-GPU hosts)
+#                     _raw_caps=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
+#                     while IFS= read -r _cap; do
+#                         _cap=$(echo "$_cap" | tr -d '[:space:]')
+#                         if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
+#                             _arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
+#                             # Append if not already present
+#                             case ";$CUDA_ARCHS;" in
+#                                 *";$_arch;"*) ;;
+#                                 *) CUDA_ARCHS="${CUDA_ARCHS:+$CUDA_ARCHS;}$_arch" ;;
+#                             esac
+#                         fi
+#                     done <<< "$_raw_caps"
+#                 fi
+#
+#                 if [ -n "$CUDA_ARCHS" ]; then
+#                     echo "   GPU compute capabilities: ${CUDA_ARCHS//;/, } -- limiting build to detected archs"
+#                     CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
+#                 else
+#                     echo "   Could not detect GPU arch -- building for all default CUDA architectures (slower)"
+#                 fi
+#
+#                 # Multi-threaded nvcc compilation (uses all CPU cores per .cu file)
+#                 CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
+#             elif [ -d /usr/local/cuda ] || nvidia-smi &>/dev/null; then
+#                 echo "   CUDA driver detected but nvcc not found — building CPU-only"
+#                 echo "   To enable GPU: install cuda-toolkit or add nvcc to PATH"
+#             else
+#                 echo "   Building CPU-only (no CUDA detected)..."
+#             fi
+#
+#             NCPU=$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)
+#
+#             # Use Ninja if available (faster parallel builds than Make)
+#             CMAKE_GENERATOR_ARGS=""
+#             if command -v ninja &>/dev/null; then
+#                 CMAKE_GENERATOR_ARGS="-G Ninja"
+#             fi
+#
+#             run_quiet "cmake llama.cpp" cmake $CMAKE_GENERATOR_ARGS -S "$LLAMA_CPP_DIR" -B "$LLAMA_CPP_DIR/build" $CMAKE_ARGS || BUILD_OK=false
+#         fi
+#
+#         if [ "$BUILD_OK" = true ]; then
+#             run_quiet "build llama-server" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-server -j"$NCPU" || BUILD_OK=false
+#         fi
+#
+#         # Also build llama-quantize (needed by unsloth-zoo's GGUF export pipeline)
+#         if [ "$BUILD_OK" = true ]; then
+#             run_quiet "build llama-quantize" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-quantize -j"$NCPU" || true
+#             # Symlink to llama.cpp root — check_llama_cpp() looks for the binary there
+#             QUANTIZE_BIN="$LLAMA_CPP_DIR/build/bin/llama-quantize"
+#             if [ -f "$QUANTIZE_BIN" ]; then
+#                 ln -sf build/bin/llama-quantize "$LLAMA_CPP_DIR/llama-quantize"
+#             fi
+#         fi
+#
+#         if [ "$BUILD_OK" = true ]; then
+#             if [ -f "$LLAMA_SERVER_BIN" ]; then
+#                 echo "✅ llama-server built at $LLAMA_SERVER_BIN"
+#             else
+#                 echo "⚠️  llama-server binary not found after build — GGUF inference won't be available"
+#             fi
+#             if [ -f "$LLAMA_CPP_DIR/llama-quantize" ]; then
+#                 echo "✅ llama-quantize available for GGUF export"
+#             fi
+#         else
+#             echo "⚠️  llama-server build failed — GGUF inference won't be available, but everything else works"
+#         fi
+#     fi
+# }
 
 echo ""
 if [ "$IS_COLAB" = true ]; then