Merge branch 'main' into feature/chat-api

README.md

@@ -68,6 +68,7 @@ unsloth studio -H 0.0.0.0 -p 8888
 ```
 
 #### Update
+To update, use the same install commands as above. Or run (does not work on Windows):
 ```bash
 unsloth studio update
 ```
@@ -152,7 +153,7 @@ unsloth studio -H 0.0.0.0 -p 8888
 ```
 Then to update :
 ```bash
-unsloth studio update --local
+unsloth studio update
 ```
 
 #### Developer installs: Windows PowerShell:
@@ -165,7 +166,7 @@ unsloth studio -H 0.0.0.0 -p 8888
 ```
 Then to update :
 ```bash
-unsloth studio update --local
+unsloth studio update
 ```
 
 #### Nightly: MacOS, Linux, WSL:

install.ps1

@@ -6,6 +6,7 @@
 
 function Install-UnslothStudio {
     $ErrorActionPreference = "Stop"
+    $script:UnslothVerbose = ($env:UNSLOTH_VERBOSE -eq "1")
 
     # ── Parse flags ──
     $StudioLocalInstall = $false
@@ -17,6 +18,8 @@ function Install-UnslothStudio {
         switch ($argList[$i]) {
             "--local"    { $StudioLocalInstall = $true }
             "--no-torch" { $SkipTorch = $true }
+            "--verbose"  { $script:UnslothVerbose = $true }
+            "-v"         { $script:UnslothVerbose = $true }
             "--package"  {
                 $i++
                 if ($i -ge $argList.Count) {
@@ -27,6 +30,12 @@ function Install-UnslothStudio {
             }
         }
     }
+    # Propagate to child processes so they also respect verbose mode.
+    # Process-scoped -- does not persist.
+    if ($script:UnslothVerbose) {
+        $env:UNSLOTH_VERBOSE = '1'
+    }
+
     if ($StudioLocalInstall) {
         $RepoRoot = (Resolve-Path (Split-Path -Parent $PSCommandPath)).Path
         if (-not (Test-Path (Join-Path $RepoRoot "pyproject.toml"))) {
@@ -39,10 +48,55 @@ function Install-UnslothStudio {
     $StudioHome = Join-Path $env:USERPROFILE ".unsloth\studio"
     $VenvDir = Join-Path $StudioHome "unsloth_studio"
 
+    $Rule = [string]::new([char]0x2500, 52)
+    $Sloth = [char]::ConvertFromUtf32(0x1F9A5)
+
+    function Enable-StudioVirtualTerminal {
+        if ($env:NO_COLOR) { return $false }
+        try {
+            if (-not ("StudioVT.Native" -as [type])) {
+                Add-Type -Namespace StudioVT -Name Native -MemberDefinition @'
+[DllImport("kernel32.dll")] public static extern IntPtr GetStdHandle(int nStdHandle);
+[DllImport("kernel32.dll")] public static extern bool GetConsoleMode(IntPtr h, out uint m);
+[DllImport("kernel32.dll")] public static extern bool SetConsoleMode(IntPtr h, uint m);
+'@ -ErrorAction Stop
+            }
+            $h = [StudioVT.Native]::GetStdHandle(-11)
+            [uint32]$mode = 0
+            if (-not [StudioVT.Native]::GetConsoleMode($h, [ref]$mode)) { return $false }
+            $mode = $mode -bor 0x0004
+            return [StudioVT.Native]::SetConsoleMode($h, $mode)
+        } catch {
+            return $false
+        }
+    }
+    $script:StudioVtOk = Enable-StudioVirtualTerminal
+
+    function Get-StudioAnsi {
+        param(
+            [Parameter(Mandatory = $true)]
+            [ValidateSet('Title', 'Dim', 'Ok', 'Warn', 'Err', 'Reset')]
+            [string]$Kind
+        )
+        $e = [char]27
+        switch ($Kind) {
+            'Title' { return "${e}[38;5;150m" }
+            'Dim'   { return "${e}[38;5;245m" }
+            'Ok'    { return "${e}[38;5;108m" }
+            'Warn'  { return "${e}[38;5;136m" }
+            'Err'   { return "${e}[91m" }
+            'Reset' { return "${e}[0m" }
+        }
+    }
+
     Write-Host ""
-    Write-Host "========================================="
-    Write-Host "   Unsloth Studio Installer (Windows)"
-    Write-Host "========================================="
+    if ($script:StudioVtOk -and -not $env:NO_COLOR) {
+        Write-Host ("  " + (Get-StudioAnsi Title) + $Sloth + " Unsloth Studio Installer (Windows)" + (Get-StudioAnsi Reset))
+        Write-Host ("  {0}{1}{2}" -f (Get-StudioAnsi Dim), $Rule, (Get-StudioAnsi Reset))
+    } else {
+        Write-Host ("  {0} Unsloth Studio Installer (Windows)" -f $Sloth) -ForegroundColor DarkGreen
+        Write-Host "  $Rule" -ForegroundColor DarkGray
+    }
     Write-Host ""
 
     # ── Helper: refresh PATH from registry (deduplicating entries) ──
@@ -62,13 +116,96 @@ function Install-UnslothStudio {
         $env:Path = $unique -join ";"
     }
 
+    function step {
+        param(
+            [Parameter(Mandatory = $true)][string]$Label,
+            [Parameter(Mandatory = $true)][string]$Value,
+            [string]$Color = "Green"
+        )
+        if ($script:StudioVtOk -and -not $env:NO_COLOR) {
+            $dim = Get-StudioAnsi Dim
+            $rst = Get-StudioAnsi Reset
+            $val = switch ($Color) {
+                'Green' { Get-StudioAnsi Ok }
+                'Yellow' { Get-StudioAnsi Warn }
+                'Red' { Get-StudioAnsi Err }
+                'DarkGray' { Get-StudioAnsi Dim }
+                default { Get-StudioAnsi Ok }
+            }
+            $padded = if ($Label.Length -ge 15) { $Label.Substring(0, 15) } else { $Label.PadRight(15) }
+            Write-Host ("  {0}{1}{2}{3}{4}{2}" -f $dim, $padded, $rst, $val, $Value)
+        } else {
+            $padded = if ($Label.Length -ge 15) { $Label.Substring(0, 15) } else { $Label.PadRight(15) }
+            Write-Host ("  {0}" -f $padded) -NoNewline -ForegroundColor DarkGray
+            $fc = switch ($Color) {
+                'Green' { 'DarkGreen' }
+                'Yellow' { 'Yellow' }
+                'Red' { 'Red' }
+                'DarkGray' { 'DarkGray' }
+                default { 'DarkGreen' }
+            }
+            Write-Host $Value -ForegroundColor $fc
+        }
+    }
+
+    function substep {
+        param(
+            [Parameter(Mandatory = $true)][string]$Message,
+            [string]$Color = "DarkGray"
+        )
+        if ($script:StudioVtOk -and -not $env:NO_COLOR) {
+            $msgCol = switch ($Color) {
+                'Yellow' { (Get-StudioAnsi Warn) }
+                'Red' { (Get-StudioAnsi Err) }
+                default { (Get-StudioAnsi Dim) }
+            }
+            $pad = "".PadRight(15)
+            Write-Host ("  {0}{1}{2}{3}" -f $msgCol, $pad, $Message, (Get-StudioAnsi Reset))
+        } else {
+            $fc = switch ($Color) {
+                'Yellow' { 'Yellow' }
+                'Red' { 'Red' }
+                default { 'DarkGray' }
+            }
+            Write-Host ("  {0,-15}{1}" -f "", $Message) -ForegroundColor $fc
+        }
+    }
+
+    # Run native commands quietly by default to match install.sh behavior.
+    # Full command output is shown only when --verbose / UNSLOTH_VERBOSE=1.
+    function Invoke-InstallCommand {
+        param(
+            [Parameter(Mandatory = $true)][ScriptBlock]$Command
+        )
+        $prevEap = $ErrorActionPreference
+        $ErrorActionPreference = "Continue"
+        try {
+            # Reset to avoid stale values from prior native commands.
+            $global:LASTEXITCODE = 0
+            if ($script:UnslothVerbose) {
+                # Merge stderr into stdout so progress/warning output stays visible
+                # without flipping $? on successful native commands (PS 5.1 treats
+                # stderr records as errors that set $? = $false even on exit code 0).
+                & $Command 2>&1 | Out-Host
+            } else {
+                $output = & $Command 2>&1 | Out-String
+                if ($LASTEXITCODE -ne 0) {
+                    Write-Host $output -ForegroundColor Red
+                }
+            }
+            return [int]$LASTEXITCODE
+        } finally {
+            $ErrorActionPreference = $prevEap
+        }
+    }
+
     function New-StudioShortcuts {
         param(
             [Parameter(Mandatory = $true)][string]$UnslothExePath
         )
 
         if (-not (Test-Path $UnslothExePath)) {
-            Write-Host "[WARN] Cannot create shortcuts: unsloth.exe not found at $UnslothExePath" -ForegroundColor Yellow
+            substep "cannot create shortcuts, unsloth.exe not found at $UnslothExePath" "Yellow"
             return
         }
         try {
@@ -81,7 +218,7 @@ function Install-UnslothStudio {
 
             $localAppDataDir = $env:LOCALAPPDATA
             if (-not $localAppDataDir -or [string]::IsNullOrWhiteSpace($localAppDataDir)) {
-                Write-Host "[WARN] LOCALAPPDATA path unavailable; skipped shortcut creation" -ForegroundColor Yellow
+                substep "LOCALAPPDATA path unavailable; skipped shortcut creation" "Yellow"
                 return
             }
             $appDir = Join-Path $localAppDataDir "Unsloth Studio"
@@ -104,10 +241,10 @@ function Install-UnslothStudio {
                 $null
             }
             if (-not $desktopLink) {
-                Write-Host "[WARN] Desktop path unavailable; skipped desktop shortcut creation" -ForegroundColor Yellow
+                substep "Desktop path unavailable; skipped desktop shortcut creation" "Yellow"
             }
             if (-not $startMenuLink) {
-                Write-Host "[WARN] APPDATA/Start Menu path unavailable; skipped Start menu shortcut creation" -ForegroundColor Yellow
+                substep "APPDATA/Start Menu path unavailable; skipped Start menu shortcut creation" "Yellow"
             }
             $iconPath = Join-Path $appDir "unsloth.ico"
             $bundledIcon = $null
@@ -362,27 +499,27 @@ shell.Run cmd, 0, False
                         $shortcut.Save()
                         $createdShortcutCount++
                     } catch {
-                        Write-Host "[WARN] Could not create shortcut at ${linkPath}: $($_.Exception.Message)" -ForegroundColor Yellow
+                        substep "could not create shortcut at ${linkPath}: $($_.Exception.Message)" "Yellow"
                     }
                 }
                 if ($createdShortcutCount -gt 0) {
-                    Write-Host "[OK] Created Unsloth Studio shortcut(s): $createdShortcutCount" -ForegroundColor Green
+                    substep "Created Unsloth Studio shortcut"
                 } else {
-                    Write-Host "[WARN] No Unsloth Studio shortcuts were created" -ForegroundColor Yellow
+                    substep "no Unsloth Studio shortcuts were created" "Yellow"
                 }
             } catch {
-                Write-Host "[WARN] Shortcut creation unavailable: $($_.Exception.Message)" -ForegroundColor Yellow
+                substep "shortcut creation unavailable: $($_.Exception.Message)" "Yellow"
             }
         } catch {
-            Write-Host "[WARN] Shortcut setup failed; skipping shortcuts: $($_.Exception.Message)" -ForegroundColor Yellow
+            substep "shortcut setup failed; skipping shortcuts: $($_.Exception.Message)" "Yellow"
         }
     }
 
     # ── Check winget ──
     if (-not (Get-Command winget -ErrorAction SilentlyContinue)) {
-        Write-Host "Error: winget is not available." -ForegroundColor Red
-        Write-Host "       Install it from https://aka.ms/getwinget" -ForegroundColor Yellow
-        Write-Host "       or install Python $PythonVersion and uv manually, then re-run." -ForegroundColor Yellow
+        step "winget" "not available" "Red"
+        substep "Install it from https://aka.ms/getwinget" "Yellow"
+        substep "or install Python $PythonVersion and uv manually, then re-run." "Yellow"
         return
     }
 
@@ -460,10 +597,10 @@ shell.Run cmd, 0, False
     # Find-CompatiblePython returns @{ Version = "3.13"; Path = "C:\...\python.exe" } or $null.
     $DetectedPython = Find-CompatiblePython
     if ($DetectedPython) {
-        Write-Host "==> Python already installed: Python $($DetectedPython.Version)"
+        step "python" "Python $($DetectedPython.Version) already installed"
     }
     if (-not $DetectedPython) {
-        Write-Host "==> Installing Python ${PythonVersion}..."
+        substep "installing Python ${PythonVersion}..."
         $pythonPackageId = "Python.Python.$PythonVersion"
         # Temporarily lower ErrorActionPreference so that winget stderr
         # (progress bars, warnings) does not become a terminating error
@@ -485,7 +622,7 @@ shell.Run cmd, 0, False
             # This handles both real failures AND "already installed" codes where
             # winget thinks Python is present but it's not actually on PATH
             # (e.g. user partially uninstalled, or installed via a different method).
-            Write-Host "    Python not found on PATH after winget. Retrying with --force..."
+            substep "Python not found on PATH after winget. Retrying with --force..." "Yellow"
             $ErrorActionPreference = "Continue"
             try {
                 winget install -e --id $pythonPackageId --accept-package-agreements --accept-source-agreements --force
@@ -507,7 +644,7 @@ shell.Run cmd, 0, False
 
     # ── Install uv if not present ──
     if (-not (Get-Command uv -ErrorAction SilentlyContinue)) {
-        Write-Host "==> Installing uv package manager..."
+        substep "installing uv package manager..."
         $prevEAP = $ErrorActionPreference
         $ErrorActionPreference = "Continue"
         try { winget install --id=astral-sh.uv -e --accept-package-agreements --accept-source-agreements } catch {}
@@ -515,15 +652,15 @@ shell.Run cmd, 0, False
         Refresh-SessionPath
         # Fallback: if winget didn't put uv on PATH, try the PowerShell installer
         if (-not (Get-Command uv -ErrorAction SilentlyContinue)) {
-            Write-Host "    Trying alternative uv installer..."
+            substep "trying alternative uv installer..." "Yellow"
             powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
             Refresh-SessionPath
         }
     }
 
     if (-not (Get-Command uv -ErrorAction SilentlyContinue)) {
-        Write-Host "Error: uv could not be installed." -ForegroundColor Red
-        Write-Host "       Install it from https://docs.astral.sh/uv/" -ForegroundColor Yellow
+        step "uv" "could not be installed" "Red"
+        substep "Install it from https://docs.astral.sh/uv/" "Yellow"
         return
     }
 
@@ -539,13 +676,13 @@ shell.Run cmd, 0, False
 
     if (Test-Path $VenvPython) {
         # New layout already exists -- nuke for fresh install
-        Write-Host "==> Removing existing environment for fresh install..."
+        substep "removing existing environment for fresh install..."
         Remove-Item -Recurse -Force $VenvDir
     } elseif (Test-Path (Join-Path $StudioHome ".venv\Scripts\python.exe")) {
         # Old layout (~/.unsloth/studio/.venv) exists -- validate before migrating
         $OldVenv = Join-Path $StudioHome ".venv"
         $OldPy = Join-Path $OldVenv "Scripts\python.exe"
-        Write-Host "==> Found legacy Studio environment, validating..."
+        substep "found legacy Studio environment, validating..."
         $prevEAP2 = $ErrorActionPreference
         $ErrorActionPreference = "Continue"
         try {
@@ -554,32 +691,34 @@ shell.Run cmd, 0, False
         } catch { $torchOk = $false }
         $ErrorActionPreference = $prevEAP2
         if ($torchOk) {
-            Write-Host "   Legacy environment is healthy -- migrating..."
+            substep "legacy environment is healthy -- migrating..."
             Move-Item -Path $OldVenv -Destination $VenvDir -Force
-            Write-Host "   Moved .venv -> unsloth_studio"
+            substep "moved .venv -> unsloth_studio"
             $_Migrated = $true
         } else {
-            Write-Host "   Legacy environment failed validation -- creating fresh environment"
+            substep "legacy environment failed validation -- creating fresh environment" "Yellow"
             Remove-Item -Recurse -Force $OldVenv -ErrorAction SilentlyContinue
         }
     } elseif (Test-Path (Join-Path $env:USERPROFILE "unsloth_studio\Scripts\python.exe")) {
         # CWD-relative venv from old install.ps1 -- migrate to absolute path
         $CwdVenv = Join-Path $env:USERPROFILE "unsloth_studio"
-        Write-Host "==> Found CWD-relative Studio environment, migrating to $VenvDir..."
+        substep "found CWD-relative Studio environment, migrating to $VenvDir..."
         Move-Item -Path $CwdVenv -Destination $VenvDir -Force
-        Write-Host "   Moved ~/unsloth_studio -> ~/.unsloth/studio/unsloth_studio"
+        substep "moved ~/unsloth_studio -> ~/.unsloth/studio/unsloth_studio"
         $_Migrated = $true
     }
 
     if (-not (Test-Path $VenvPython)) {
-        Write-Host "==> Creating Python $($DetectedPython.Version) virtual environment ($VenvDir)..."
-        uv venv $VenvDir --python "$($DetectedPython.Path)"
-        if ($LASTEXITCODE -ne 0) {
-            Write-Host "[ERROR] Failed to create virtual environment (exit code $LASTEXITCODE)" -ForegroundColor Red
+        step "venv" "creating Python $($DetectedPython.Version) virtual environment"
+        substep "$VenvDir"
+        $venvExit = Invoke-InstallCommand { uv venv $VenvDir --python "$($DetectedPython.Path)" }
+        if ($venvExit -ne 0) {
+            Write-Host "[ERROR] Failed to create virtual environment (exit code $venvExit)" -ForegroundColor Red
             return
         }
     } else {
-        Write-Host "==> Using migrated environment at $VenvDir"
+        step "venv" "using migrated environment"
+        substep "$VenvDir"
     }
 
     # ── Detect GPU (robust: PATH + hardcoded fallback paths, mirrors setup.ps1) ──
@@ -588,7 +727,7 @@ shell.Run cmd, 0, False
     try {
         $nvSmiCmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
         if ($nvSmiCmd) {
-            & $nvSmiCmd.Source 2>&1 | Out-Null
+            & $nvSmiCmd.Source *> $null
             if ($LASTEXITCODE -eq 0) { $HasNvidiaSmi = $true; $NvidiaSmiExe = $nvSmiCmd.Source }
         }
     } catch {}
@@ -599,18 +738,18 @@ shell.Run cmd, 0, False
         )) {
             if (Test-Path $p) {
                 try {
-                    & $p 2>&1 | Out-Null
+                    & $p *> $null
                     if ($LASTEXITCODE -eq 0) { $HasNvidiaSmi = $true; $NvidiaSmiExe = $p; break }
                 } catch {}
             }
         }
     }
     if ($HasNvidiaSmi) {
-        Write-Host "[OK] NVIDIA GPU detected" -ForegroundColor Green
+        step "gpu" "NVIDIA GPU detected"
     } else {
-        Write-Host "[WARN] No NVIDIA GPU detected. Studio will run in chat-only (GGUF) mode." -ForegroundColor Yellow
-        Write-Host "       Training and GPU inference require an NVIDIA GPU with drivers installed." -ForegroundColor Yellow
-        Write-Host "       https://www.nvidia.com/Download/index.aspx" -ForegroundColor Yellow
+        step "gpu" "none (chat-only / GGUF)" "Yellow"
+        substep "Training and GPU inference require an NVIDIA GPU with drivers installed." "Yellow"
+        substep "https://www.nvidia.com/Download/index.aspx" "Yellow"
     }
 
     # ── Choose the correct PyTorch index URL based on driver CUDA version ──
@@ -630,7 +769,7 @@ shell.Run cmd, 0, False
                 return "$baseUrl/cpu"
             }
         } catch {}
-        Write-Host "[WARN] Could not determine CUDA version from nvidia-smi, defaulting to cu126" -ForegroundColor Yellow
+        substep "could not determine CUDA version from nvidia-smi, defaulting to cu126" "Yellow"
         return "$baseUrl/cu126"
     }
     $TorchIndexUrl = Get-TorchIndexUrl
@@ -677,74 +816,101 @@ shell.Run cmd, 0, False
     if ($_Migrated) {
         # Migrated env: force-reinstall unsloth+unsloth-zoo to ensure clean state
         # in the new venv location, while preserving existing torch/CUDA
-        Write-Host "==> Upgrading unsloth in migrated environment..."
+        substep "upgrading unsloth in migrated environment..."
         if ($SkipTorch) {
             # No-torch: install unsloth + unsloth-zoo with --no-deps, then
             # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
-            uv pip install --python $VenvPython --no-deps --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
-            $NoTorchReq = Find-NoTorchRuntimeFile
-            if ($NoTorchReq) {
-                uv pip install --python $VenvPython --no-deps -r $NoTorchReq
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --no-deps --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo }
+            if ($baseInstallExit -eq 0) {
+                $NoTorchReq = Find-NoTorchRuntimeFile
+                if ($NoTorchReq) {
+                    $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --no-deps -r $NoTorchReq }
+                }
             }
         } else {
-            uv pip install --python $VenvPython --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo }
+        }
+        if ($baseInstallExit -ne 0) {
+            Write-Host "[ERROR] Failed to install unsloth (exit code $baseInstallExit)" -ForegroundColor Red
+            return
         }
         if ($StudioLocalInstall) {
-            Write-Host "==> Overlaying local repo (editable)..."
-            uv pip install --python $VenvPython -e $RepoRoot --no-deps
+            substep "overlaying local repo (editable)..."
+            $overlayExit = Invoke-InstallCommand { uv pip install --python $VenvPython -e $RepoRoot --no-deps }
+            if ($overlayExit -ne 0) {
+                Write-Host "[ERROR] Failed to overlay local repo (exit code $overlayExit)" -ForegroundColor Red
+                return
+            }
         }
     } elseif ($TorchIndexUrl) {
         if ($SkipTorch) {
-            Write-Host "==> Skipping PyTorch (--no-torch flag set)."
+            substep "skipping PyTorch (--no-torch flag set)." "Yellow"
         } else {
-            Write-Host "==> Installing PyTorch ($TorchIndexUrl)..."
-            uv pip install --python $VenvPython "torch>=2.4,<2.11.0" torchvision torchaudio --index-url $TorchIndexUrl
-            if ($LASTEXITCODE -ne 0) {
-                Write-Host "[ERROR] Failed to install PyTorch (exit code $LASTEXITCODE)" -ForegroundColor Red
+            substep "installing PyTorch ($TorchIndexUrl)..."
+            $torchInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython "torch>=2.4,<2.11.0" torchvision torchaudio --index-url $TorchIndexUrl }
+            if ($torchInstallExit -ne 0) {
+                Write-Host "[ERROR] Failed to install PyTorch (exit code $torchInstallExit)" -ForegroundColor Red
                 return
             }
         }
 
-        Write-Host "==> Installing unsloth (this may take a few minutes)..."
+        substep "installing unsloth (this may take a few minutes)..."
         if ($SkipTorch) {
             # No-torch: install unsloth + unsloth-zoo with --no-deps, then
             # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
-            uv pip install --python $VenvPython --no-deps --upgrade-package unsloth --upgrade-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
-            $NoTorchReq = Find-NoTorchRuntimeFile
-            if ($NoTorchReq) {
-                uv pip install --python $VenvPython --no-deps -r $NoTorchReq
-            }
-            if ($StudioLocalInstall) {
-                Write-Host "==> Overlaying local repo (editable)..."
-                uv pip install --python $VenvPython -e $RepoRoot --no-deps
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --no-deps --upgrade-package unsloth --upgrade-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo }
+            if ($baseInstallExit -eq 0) {
+                $NoTorchReq = Find-NoTorchRuntimeFile
+                if ($NoTorchReq) {
+                    $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --no-deps -r $NoTorchReq }
+                }
             }
         } elseif ($StudioLocalInstall) {
-            uv pip install --python $VenvPython --upgrade-package unsloth "unsloth>=2026.3.16" unsloth-zoo
-            Write-Host "==> Overlaying local repo (editable)..."
-            uv pip install --python $VenvPython -e $RepoRoot --no-deps
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --upgrade-package unsloth "unsloth>=2026.3.16" unsloth-zoo }
         } else {
-            uv pip install --python $VenvPython --upgrade-package unsloth "$PackageName"
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython --upgrade-package unsloth "$PackageName" }
+        }
+        if ($baseInstallExit -ne 0) {
+            Write-Host "[ERROR] Failed to install unsloth (exit code $baseInstallExit)" -ForegroundColor Red
+            return
+        }
+
+        if ($StudioLocalInstall) {
+            substep "overlaying local repo (editable)..."
+            $overlayExit = Invoke-InstallCommand { uv pip install --python $VenvPython -e $RepoRoot --no-deps }
+            if ($overlayExit -ne 0) {
+                Write-Host "[ERROR] Failed to overlay local repo (exit code $overlayExit)" -ForegroundColor Red
+                return
+            }
         }
     } else {
         # Fallback: GPU detection failed to produce a URL -- let uv resolve torch
-        Write-Host "==> Installing unsloth (this may take a few minutes)..."
+        substep "installing unsloth (this may take a few minutes)..."
         if ($StudioLocalInstall) {
-            uv pip install --python $VenvPython unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto
-            Write-Host "==> Overlaying local repo (editable)..."
-            uv pip install --python $VenvPython -e $RepoRoot --no-deps
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto }
+            if ($baseInstallExit -ne 0) {
+                Write-Host "[ERROR] Failed to install unsloth (exit code $baseInstallExit)" -ForegroundColor Red
+                return
+            }
+            substep "overlaying local repo (editable)..."
+            $overlayExit = Invoke-InstallCommand { uv pip install --python $VenvPython -e $RepoRoot --no-deps }
+            if ($overlayExit -ne 0) {
+                Write-Host "[ERROR] Failed to overlay local repo (exit code $overlayExit)" -ForegroundColor Red
+                return
+            }
         } else {
-            uv pip install --python $VenvPython "$PackageName" --torch-backend=auto
+            $baseInstallExit = Invoke-InstallCommand { uv pip install --python $VenvPython "$PackageName" --torch-backend=auto }
+            if ($baseInstallExit -ne 0) {
+                Write-Host "[ERROR] Failed to install unsloth (exit code $baseInstallExit)" -ForegroundColor Red
+                return
+            }
         }
     }
-    if ($LASTEXITCODE -ne 0) {
-        Write-Host "[ERROR] Failed to install unsloth (exit code $LASTEXITCODE)" -ForegroundColor Red
-        return
-    }
 
     # ── Run studio setup ──
     # setup.ps1 will handle installing Git, CMake, Visual Studio Build Tools,
     # CUDA Toolkit, Node.js, and other dependencies automatically via winget.
-    Write-Host "==> Running unsloth studio setup..."
+    step "setup" "running unsloth studio setup..."
     $UnslothExe = Join-Path $VenvDir "Scripts\unsloth.exe"
     if (-not (Test-Path $UnslothExe)) {
         Write-Host "[ERROR] unsloth CLI was not installed correctly." -ForegroundColor Red
@@ -754,17 +920,27 @@ shell.Run cmd, 0, False
         return
     }
     # Tell setup.ps1 to skip base package installation (install.ps1 already did it)
-    # Tell setup.ps1 to skip base package installation (install.ps1 already did it)
     $env:SKIP_STUDIO_BASE = "1"
     $env:STUDIO_PACKAGE_NAME = $PackageName
     $env:UNSLOTH_NO_TORCH = if ($SkipTorch) { "true" } else { "false" }
+    # Always set STUDIO_LOCAL_INSTALL explicitly to avoid stale values from
+    # a previous --local run in the same PowerShell session.
     if ($StudioLocalInstall) {
         $env:STUDIO_LOCAL_INSTALL = "1"
         $env:STUDIO_LOCAL_REPO = $RepoRoot
+    } else {
+        $env:STUDIO_LOCAL_INSTALL = "0"
+        Remove-Item Env:STUDIO_LOCAL_REPO -ErrorAction SilentlyContinue
     }
-    & $UnslothExe studio setup
-    if ($LASTEXITCODE -ne 0) {
-        Write-Host "[ERROR] unsloth studio setup failed (exit code $LASTEXITCODE)" -ForegroundColor Red
+    # Use 'studio setup' (not 'studio update') because 'update' pops
+    # SKIP_STUDIO_BASE, which would cause redundant package reinstallation
+    # and bypass the fast-path version check from PR #4667.
+    $studioArgs = @('studio', 'setup')
+    if ($script:UnslothVerbose) { $studioArgs += '--verbose' }
+    & $UnslothExe @studioArgs
+    $setupExit = $LASTEXITCODE
+    if ($setupExit -ne 0) {
+        Write-Host "[ERROR] unsloth studio setup failed (exit code $setupExit)" -ForegroundColor Red
         return
     }
 
@@ -780,27 +956,18 @@ shell.Run cmd, 0, False
             [System.Environment]::SetEnvironmentVariable("Path", "$ScriptsDir", "User")
         }
         Refresh-SessionPath
-        Write-Host "[OK] Added unsloth to PATH" -ForegroundColor Green
+        step "path" "added unsloth to PATH"
     }
 
-    Write-Host ""
-    Write-Host "========================================="
-    Write-Host "   Unsloth Studio installed!"
-    Write-Host "========================================="
-    Write-Host ""
-
     # Launch studio automatically in interactive terminals;
     # in non-interactive environments (CI, Docker) just print instructions.
     $IsInteractive = [Environment]::UserInteractive -and (-not [Console]::IsInputRedirected)
     if ($IsInteractive) {
-        Write-Host "==> Launching Unsloth Studio..."
-        Write-Host ""
         & $UnslothExe studio -H 0.0.0.0 -p 8888
     } else {
-        Write-Host "  To launch, run:"
-        Write-Host ""
-        Write-Host "    & `"$VenvDir\Scripts\Activate.ps1`""
-        Write-Host "    unsloth studio -H 0.0.0.0 -p 8888"
+        step "launch" "manual commands:"
+        substep "& `"$VenvDir\Scripts\Activate.ps1`""
+        substep "unsloth studio -H 0.0.0.0 -p 8888"
         Write-Host ""
     }
 }

install.sh

@@ -8,11 +8,36 @@
 # Usage (py):    ./install.sh --python 3.12  (override auto-detected Python version)
 set -e
 
+# ── Output style (aligned with studio/setup.sh) ──
+RULE=""
+_rule_i=0
+while [ "$_rule_i" -lt 52 ]; do
+    RULE="${RULE}─"
+    _rule_i=$((_rule_i + 1))
+done
+if [ -n "${NO_COLOR:-}" ]; then
+    C_TITLE= C_DIM= C_OK= C_WARN= C_ERR= C_RST=
+elif [ -t 1 ] || [ -n "${FORCE_COLOR:-}" ]; then
+    _ESC="$(printf '\033')"
+    C_TITLE="${_ESC}[38;5;150m"
+    C_DIM="${_ESC}[38;5;245m"
+    C_OK="${_ESC}[38;5;108m"
+    C_WARN="${_ESC}[38;5;136m"
+    C_ERR="${_ESC}[91m"
+    C_RST="${_ESC}[0m"
+else
+    C_TITLE= C_DIM= C_OK= C_WARN= C_ERR= C_RST=
+fi
+
+step()    { printf "  ${C_DIM}%-15.15s${C_RST}${3:-$C_OK}%s${C_RST}\n" "$1" "$2"; }
+substep() { printf "  ${C_DIM}%-15s${2:-$C_DIM}%s${C_RST}\n" "" "$1"; }
+
 # ── Parse flags ──
 STUDIO_LOCAL_INSTALL=false
 PACKAGE_NAME="unsloth"
 _USER_PYTHON=""
 _NO_TORCH_FLAG=false
+_VERBOSE=false
 _next_is_package=false
 _next_is_python=false
 for arg in "$@"; do
@@ -31,9 +56,44 @@ for arg in "$@"; do
         --package) _next_is_package=true ;;
         --python) _next_is_python=true ;;
         --no-torch) _NO_TORCH_FLAG=true ;;
+        --verbose|-v) _VERBOSE=true ;;
     esac
 done
 
+if [ "$_VERBOSE" = true ]; then
+    export UNSLOTH_VERBOSE=1
+fi
+
+_is_verbose() {
+    [ "${UNSLOTH_VERBOSE:-0}" = "1" ]
+}
+
+run_maybe_quiet() {
+    if _is_verbose; then
+        "$@"
+    else
+        "$@" > /dev/null 2>&1
+    fi
+}
+
+run_install_cmd() {
+    _label="$1"
+    shift
+    if _is_verbose; then
+        "$@" && return 0
+        _rc=$?
+        step "error" "$_label failed (exit code $_rc)" "$C_ERR" >&2
+        return "$_rc"
+    fi
+    _log=$(mktemp)
+    "$@" >"$_log" 2>&1 && { rm -f "$_log"; return 0; }
+    _rc=$?
+    step "error" "$_label failed (exit code $_rc)" "$C_ERR" >&2
+    cat "$_log" >&2
+    rm -f "$_log"
+    return $_rc
+}
+
 if [ "$_next_is_package" = true ]; then
     echo "❌ ERROR: --package requires an argument." >&2
     exit 1
@@ -643,14 +703,13 @@ WSLPS1_EOF
     fi
 
     if [ "$_css_created" -eq 1 ]; then
-        echo "[OK] Created Unsloth Studio shortcut(s)"
+        substep "Created Unsloth Studio shortcut"
     fi
 }
 
 echo ""
-echo "========================================="
-echo "   Unsloth Studio Installer"
-echo "========================================="
+printf "  ${C_TITLE}%s${C_RST}\n" "🦥 Unsloth Studio Installer"
+printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
 echo ""
 
 # ── Detect platform ──
@@ -660,7 +719,7 @@ if [ "$(uname)" = "Darwin" ]; then
 elif grep -qi microsoft /proc/version 2>/dev/null; then
     OS="wsl"
 fi
-echo "==> Platform: $OS"
+step "platform" "$OS"
 
 # ── Architecture detection & Python version ──
 _ARCH=$(uname -m)
@@ -740,8 +799,8 @@ MISSING=$(echo "$MISSING" | sed 's/^ *//')
 
 if [ -n "$MISSING" ]; then
     echo ""
-    echo "==> Unsloth Studio needs these packages: $MISSING"
-    echo "    These are needed to build the GGUF inference engine."
+    step "deps" "missing: $MISSING" "$C_WARN"
+    substep "These are needed to build the GGUF inference engine."
 
     case "$OS" in
         macos)
@@ -766,7 +825,7 @@ if [ -n "$MISSING" ]; then
     esac
     echo ""
 else
-    echo "==> All system dependencies found."
+    step "deps" "all system dependencies found"
 fi
 
 # ── Install uv ──
@@ -812,10 +871,10 @@ _uv_version_ok() {
 }
 
 if ! command -v uv >/dev/null 2>&1 || ! _uv_version_ok uv; then
-    echo "==> Installing uv package manager..."
+    substep "installing uv package manager..."
     _uv_tmp=$(mktemp)
     download "https://astral.sh/uv/install.sh" "$_uv_tmp"
-    sh "$_uv_tmp" </dev/null
+    run_maybe_quiet sh "$_uv_tmp" </dev/null
     rm -f "$_uv_tmp"
     if [ -f "$HOME/.local/bin/env" ]; then
         . "$HOME/.local/bin/env"
@@ -833,7 +892,7 @@ if [ -x "$VENV_DIR/bin/python" ]; then
     rm -rf "$VENV_DIR"
 elif [ -x "$STUDIO_HOME/.venv/bin/python" ]; then
     # Old layout exists — validate before migrating
-    echo "==> Found legacy Studio environment, validating..."
+    substep "found legacy Studio environment, validating..."
     if "$STUDIO_HOME/.venv/bin/python" -c "
 import torch
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
@@ -866,8 +925,9 @@ if [ "$SKIP_TORCH" = true ] && [ "$MAC_INTEL" = true ] && [ -z "$_USER_PYTHON" ]
 fi
 
 if [ ! -x "$VENV_DIR/bin/python" ]; then
-    echo "==> Creating Python ${PYTHON_VERSION} virtual environment (${VENV_DIR})..."
-    uv venv "$VENV_DIR" --python "$PYTHON_VERSION"
+    step "venv" "creating Python ${PYTHON_VERSION} virtual environment"
+    substep "$VENV_DIR"
+    run_install_cmd "create venv" uv venv "$VENV_DIR" --python "$PYTHON_VERSION"
 fi
 
 # Guard against Python 3.13.8 torch import bug on Apple Silicon
@@ -880,12 +940,13 @@ if [ -z "$_USER_PYTHON" ] && [ "$OS" = "macos" ] && [ "$_ARCH" = "arm64" ]; then
         echo "  Recreating venv with Python 3.12..."
         rm -rf "$VENV_DIR"
         PYTHON_VERSION="3.12"
-        uv venv "$VENV_DIR" --python "$PYTHON_VERSION"
+        run_install_cmd "recreate venv" uv venv "$VENV_DIR" --python "$PYTHON_VERSION"
     fi
 fi
 
 if [ -x "$VENV_DIR/bin/python" ]; then
-    echo "==> Using environment at ${VENV_DIR}"
+    step "venv" "using environment"
+    substep "${VENV_DIR}"
 fi
 
 # ── Resolve repo root (for --local installs) ──
@@ -960,71 +1021,71 @@ _VENV_PY="$VENV_DIR/bin/python"
 if [ "$_MIGRATED" = true ]; then
     # Migrated env: force-reinstall unsloth+unsloth-zoo to ensure clean state
     # in the new venv location, while preserving existing torch/CUDA
-    echo "==> Upgrading unsloth in migrated environment..."
+    substep "upgrading unsloth in migrated environment..."
     if [ "$SKIP_TORCH" = true ]; then
         # No-torch: install unsloth + unsloth-zoo with --no-deps (current
         # PyPI metadata still declares torch as a hard dep), then install
         # runtime deps (typer, safetensors, transformers, etc.) with --no-deps
         # to prevent transitive torch resolution.
-        uv pip install --python "$_VENV_PY" --no-deps \
+        run_install_cmd "install unsloth (migrated no-torch)" uv pip install --python "$_VENV_PY" --no-deps \
             --reinstall-package unsloth --reinstall-package unsloth-zoo \
             "unsloth>=2026.3.16" unsloth-zoo
         _NO_TORCH_RT="$(_find_no_torch_runtime)"
         if [ -n "$_NO_TORCH_RT" ]; then
-            uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
+            run_install_cmd "install no-torch runtime deps" uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
         fi
     else
-        uv pip install --python "$_VENV_PY" \
+        run_install_cmd "install unsloth (migrated)" uv pip install --python "$_VENV_PY" \
             --reinstall-package unsloth --reinstall-package unsloth-zoo \
             "unsloth>=2026.3.16" unsloth-zoo
     fi
     if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-        echo "==> Overlaying local repo (editable)..."
-        uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
+        substep "overlaying local repo (editable)..."
+        run_install_cmd "overlay local repo" uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
     fi
 elif [ -n "$TORCH_INDEX_URL" ]; then
     # Fresh: Step 1 - install torch from explicit index (skip when --no-torch or Intel Mac)
     if [ "$SKIP_TORCH" = true ]; then
-        echo "==> Skipping PyTorch (--no-torch or Intel Mac x86_64)."
+        substep "skipping PyTorch (--no-torch or Intel Mac x86_64)." "$C_WARN"
     else
-        echo "==> Installing PyTorch ($TORCH_INDEX_URL)..."
-        uv pip install --python "$_VENV_PY" "torch>=2.4,<2.11.0" torchvision torchaudio \
+        substep "installing PyTorch ($TORCH_INDEX_URL)..."
+        run_install_cmd "install PyTorch" uv pip install --python "$_VENV_PY" "torch>=2.4,<2.11.0" torchvision torchaudio \
             --index-url "$TORCH_INDEX_URL"
     fi
     # Fresh: Step 2 - install unsloth, preserving pre-installed torch
-    echo "==> Installing unsloth (this may take a few minutes)..."
+    substep "installing unsloth (this may take a few minutes)..."
     if [ "$SKIP_TORCH" = true ]; then
         # No-torch: install unsloth + unsloth-zoo with --no-deps, then
         # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
-        uv pip install --python "$_VENV_PY" --no-deps \
+        run_install_cmd "install unsloth (no-torch)" uv pip install --python "$_VENV_PY" --no-deps \
             --upgrade-package unsloth --upgrade-package unsloth-zoo \
             "unsloth>=2026.3.16" unsloth-zoo
         _NO_TORCH_RT="$(_find_no_torch_runtime)"
         if [ -n "$_NO_TORCH_RT" ]; then
-            uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
+            run_install_cmd "install no-torch runtime deps" uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
         fi
         if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-            echo "==> Overlaying local repo (editable)..."
-            uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
+            substep "overlaying local repo (editable)..."
+            run_install_cmd "overlay local repo" uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
         fi
     elif [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-        uv pip install --python "$_VENV_PY" \
+        run_install_cmd "install unsloth (local)" uv pip install --python "$_VENV_PY" \
             --upgrade-package unsloth "unsloth>=2026.3.16" unsloth-zoo
-        echo "==> Overlaying local repo (editable)..."
-        uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
+        substep "overlaying local repo (editable)..."
+        run_install_cmd "overlay local repo" uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
     else
-        uv pip install --python "$_VENV_PY" \
+        run_install_cmd "install unsloth" uv pip install --python "$_VENV_PY" \
             --upgrade-package unsloth "$PACKAGE_NAME"
     fi
 else
     # Fallback: GPU detection failed to produce a URL -- let uv resolve torch
-    echo "==> Installing unsloth (this may take a few minutes)..."
+    substep "installing unsloth (this may take a few minutes)..."
     if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-        uv pip install --python "$_VENV_PY" unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto
-        echo "==> Overlaying local repo (editable)..."
-        uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
+        run_install_cmd "install unsloth (auto torch backend)" uv pip install --python "$_VENV_PY" unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto
+        substep "overlaying local repo (editable)..."
+        run_install_cmd "overlay local repo" uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
     else
-        uv pip install --python "$_VENV_PY" "$PACKAGE_NAME" --torch-backend=auto
+        run_install_cmd "install unsloth (auto torch backend)" uv pip install --python "$_VENV_PY" "$PACKAGE_NAME" --torch-backend=auto
     fi
 fi
 
@@ -1059,19 +1120,33 @@ if [ -n "$VENV_ABS_BIN" ]; then
     export PATH="$VENV_ABS_BIN:$PATH"
 fi
 
-echo "==> Running unsloth setup..."
+if ! command -v bash >/dev/null 2>&1; then
+    step "setup" "bash is required to run studio setup" "$C_ERR"
+    substep "Please install bash and re-run install.sh"
+    exit 1
+fi
+
+step "setup" "running unsloth studio update..."
+# install.sh already installs base packages (unsloth + unsloth-zoo) and
+# no-torch-runtime.txt above, so tell install_python_stack.py to skip
+# the base step to avoid redundant reinstallation.
+_SKIP_BASE=1
+# Run setup.sh outside set -e so that a llama.cpp build failure (exit 1)
+# does not skip PATH setup, shortcuts, and launch below.  We capture the
+# exit code and propagate it after post-install steps finish.
+_SETUP_EXIT=0
 if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-    SKIP_STUDIO_BASE=1 \
+    SKIP_STUDIO_BASE="$_SKIP_BASE" \
     STUDIO_PACKAGE_NAME="$PACKAGE_NAME" \
     STUDIO_LOCAL_INSTALL=1 \
     STUDIO_LOCAL_REPO="$_REPO_ROOT" \
     UNSLOTH_NO_TORCH="$SKIP_TORCH" \
-    bash "$SETUP_SH" </dev/null
+    bash "$SETUP_SH" </dev/null || _SETUP_EXIT=$?
 else
-    SKIP_STUDIO_BASE=1 \
+    SKIP_STUDIO_BASE="$_SKIP_BASE" \
     STUDIO_PACKAGE_NAME="$PACKAGE_NAME" \
     UNSLOTH_NO_TORCH="$SKIP_TORCH" \
-    bash "$SETUP_SH" </dev/null
+    bash "$SETUP_SH" </dev/null || _SETUP_EXIT=$?
 fi
 
 # ── Make 'unsloth' available globally via ~/.local/bin ──
@@ -1096,7 +1171,7 @@ case ":$PATH:" in
                 echo '' >> "$_SHELL_PROFILE"
                 echo '# Added by Unsloth installer' >> "$_SHELL_PROFILE"
                 echo 'export PATH="$HOME/.local/bin:$PATH"' >> "$_SHELL_PROFILE"
-                echo "==> Added ~/.local/bin to PATH in $_SHELL_PROFILE"
+                step "path" "added ~/.local/bin to PATH in $_SHELL_PROFILE"
             fi
         fi
         export PATH="$_LOCAL_BIN:$PATH"
@@ -1105,17 +1180,30 @@ esac
 
 create_studio_shortcuts "$VENV_ABS_BIN/unsloth" "$OS"
 
+# If setup.sh failed, report and exit now.
+# PATH and shortcuts are already set up so the user can fix and retry.
+if [ "$_SETUP_EXIT" -ne 0 ]; then
+    echo ""
+    step "error" "studio setup failed (exit code $_SETUP_EXIT)" "$C_ERR"
+    substep "Check the output above for details, then re-run:"
+    if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
+        substep "  unsloth studio update --local"
+    else
+        substep "  unsloth studio update"
+    fi
+    echo ""
+    exit "$_SETUP_EXIT"
+fi
+
 echo ""
-echo "========================================="
-echo "   Unsloth Studio installed!"
-echo "========================================="
+printf "  ${C_TITLE}%s${C_RST}\n" "Unsloth Studio installed!"
+printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
 echo ""
 
 # Launch studio automatically in interactive terminals;
 # in non-interactive environments (Docker, CI, cloud-init) just print instructions.
 if [ -t 1 ]; then
-    echo "==> Launching Unsloth Studio..."
-    echo ""
+    step "launch" "starting Unsloth Studio..."
     "$VENV_DIR/bin/unsloth" studio -H 0.0.0.0 -p 8888
     _LAUNCH_EXIT=$?
     if [ "$_LAUNCH_EXIT" -ne 0 ] && [ "$_MIGRATED" = true ]; then
@@ -1130,13 +1218,10 @@ if [ -t 1 ]; then
     fi
     exit "$_LAUNCH_EXIT"
 else
-    echo "  To launch, run:"
-    echo ""
-    echo "    unsloth studio -H 0.0.0.0 -p 8888"
-    echo ""
-    echo "  Or activate the environment first:"
-    echo ""
-    echo "    source ${VENV_DIR}/bin/activate"
-    echo "    unsloth studio -H 0.0.0.0 -p 8888"
+    step "launch" "manual commands:"
+    substep "unsloth studio -H 0.0.0.0 -p 8888"
+    substep "or activate env first:"
+    substep "source ${VENV_DIR}/bin/activate"
+    substep "unsloth studio -H 0.0.0.0 -p 8888"
     echo ""
 fi

pyproject.toml

@@ -58,7 +58,8 @@ studio = [
 ]
 
 [tool.setuptools.packages.find]
-exclude = ["images*", "tests*", "kernels/moe*"]
+include = ["unsloth*", "unsloth_cli*", "studio", "studio.backend*"]
+exclude = ["images*", "tests*", "*.node_modules", "*.node_modules.*"]
 
 [project.optional-dependencies]
 triton = [

studio/backend/core/inference/inference.py

@@ -18,7 +18,14 @@ from typing import Optional, Union, Generator, Tuple
 from utils.models import ModelConfig, get_base_model_from_lora
 from utils.paths import is_model_cached
 from utils.utils import format_error_message
-from utils.hardware import get_device, clear_gpu_cache, log_gpu_memory
+from utils.hardware import (
+    get_device,
+    clear_gpu_cache,
+    log_gpu_memory,
+    get_device_map,
+    raise_if_offloaded,
+    get_visible_gpu_count,
+)
 from core.inference.audio_codecs import AudioCodecManager
 from io import StringIO
 import structlog
@@ -241,6 +248,7 @@ class InferenceBackend:
         load_in_4bit: bool = True,
         hf_token: Optional[str] = None,
         trust_remote_code: bool = False,
+        gpu_ids: Optional[list[int]] = None,
     ) -> bool:
         """
         Load any model: base, LoRA adapter, text, or vision.
@@ -260,6 +268,10 @@ class InferenceBackend:
                 return False
 
             self.loading_models.add(model_name)
+            device_map = get_device_map(gpu_ids, for_inference = True)
+            logger.info(
+                f"Using device_map='{device_map}' ({get_visible_gpu_count()} GPU(s) visible)"
+            )
 
             self.models[model_name] = {
                 "is_vision": config.is_vision,
@@ -290,6 +302,7 @@ class InferenceBackend:
                         config.path,
                         auto_model = CsmForConditionalGeneration,
                         load_in_4bit = False,
+                        device_map = device_map,
                         token = hf_token if hf_token and hf_token.strip() else None,
                         trust_remote_code = trust_remote_code,
                     )
@@ -325,6 +338,7 @@ class InferenceBackend:
                             config.path,
                             dtype = torch.float32,
                             load_in_4bit = False,
+                            device_map = device_map,
                             token = hf_token if hf_token and hf_token.strip() else None,
                             trust_remote_code = trust_remote_code,
                         )
@@ -345,6 +359,7 @@ class InferenceBackend:
                             llm_path,
                             dtype = torch.float32,
                             load_in_4bit = False,
+                            device_map = device_map,
                             token = hf_token if hf_token and hf_token.strip() else None,
                             trust_remote_code = trust_remote_code,
                         )
@@ -361,6 +376,7 @@ class InferenceBackend:
                         config.path,
                         max_seq_length = max_seq_length,
                         load_in_4bit = False,
+                        device_map = device_map,
                         token = hf_token if hf_token and hf_token.strip() else None,
                         trust_remote_code = trust_remote_code,
                     )
@@ -378,6 +394,7 @@ class InferenceBackend:
                         whisper_language = "English",
                         whisper_task = "transcribe",
                         load_in_4bit = False,
+                        device_map = device_map,
                         token = hf_token if hf_token and hf_token.strip() else None,
                         trust_remote_code = trust_remote_code,
                     )
@@ -405,6 +422,7 @@ class InferenceBackend:
                         model_name = config.path,
                         max_seq_length = max_seq_length,
                         load_in_4bit = False,
+                        device_map = device_map,
                         token = hf_token if hf_token and hf_token.strip() else None,
                         trust_remote_code = trust_remote_code,
                     )
@@ -420,6 +438,11 @@ class InferenceBackend:
                         audio_type, self.device, model_repo_path = model_repo_path
                     )
 
+                # Reject CPU/disk offload for audio models too
+                raise_if_offloaded(
+                    self.models[model_name]["model"], device_map, "Inference"
+                )
+
                 self.active_model_name = model_name
                 self.loading_models.discard(model_name)
                 logger.info(f"Successfully loaded audio model: {model_name}")
@@ -441,6 +464,7 @@ class InferenceBackend:
                     max_seq_length = max_seq_length,
                     dtype = dtype,
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     token = hf_token if hf_token and hf_token.strip() else None,
                     trust_remote_code = trust_remote_code,
                 )
@@ -497,6 +521,7 @@ class InferenceBackend:
                     max_seq_length = max_seq_length,
                     dtype = dtype,
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     token = hf_token if hf_token and hf_token.strip() else None,
                     trust_remote_code = trust_remote_code,
                 )
@@ -507,6 +532,10 @@ class InferenceBackend:
                 self.models[model_name]["model"] = model
                 self.models[model_name]["tokenizer"] = tokenizer
 
+            raise_if_offloaded(
+                self.models[model_name]["model"], device_map, "Inference"
+            )
+
             # Load chat template info
             self._load_chat_template_info(model_name)
 
@@ -615,6 +644,7 @@ class InferenceBackend:
         dtype = None,
         load_in_4bit: bool = True,
         hf_token: Optional[str] = None,
+        gpu_ids: Optional[list[int]] = None,
     ) -> Tuple[bool, Optional[str], Optional[str]]:
         """
         Final Corrected Version:
@@ -639,7 +669,12 @@ class InferenceBackend:
                     base_model_name, None, is_lora = False
                 )
                 if not self.load_model(
-                    base_config, max_seq_length, dtype, load_in_4bit, hf_token
+                    base_config,
+                    max_seq_length,
+                    dtype,
+                    load_in_4bit,
+                    hf_token,
+                    gpu_ids = gpu_ids,
                 ):
                     return False, None, None
 
@@ -1037,12 +1072,12 @@ class InferenceBackend:
                 input_text,
                 add_special_tokens = False,
                 return_tensors = "pt",
-            ).to(self.device)
+            ).to(model.device)
         else:
             # Text-only for vision model
             formatted_prompt = self.format_chat_prompt(messages, system_prompt)
             inputs = raw_tokenizer(formatted_prompt, return_tensors = "pt").to(
-                self.device
+                model.device
             )
 
         # Stream with TextIteratorStreamer + background thread
@@ -1182,7 +1217,7 @@ class InferenceBackend:
             return_dict = True,
             return_tensors = "pt",
             truncation = False,
-        ).to(self.device)
+        ).to(model.device)
 
         try:
             from transformers import TextIteratorStreamer

studio/backend/core/inference/llama_cpp.py

@@ -48,6 +48,7 @@ class LlamaCppBackend:
         self._healthy = False
         self._context_length: Optional[int] = None
         self._effective_context_length: Optional[int] = None
+        self._max_context_length: Optional[int] = None
         self._chat_template: Optional[str] = None
         self._supports_reasoning: bool = False
         self._reasoning_always_on: bool = False
@@ -100,6 +101,11 @@ class LlamaCppBackend:
         """Return the effective context length the server is running at."""
         return self._effective_context_length or self._context_length
 
+    @property
+    def max_context_length(self) -> Optional[int]:
+        """Return the maximum context currently available on this hardware."""
+        return self._max_context_length or self._context_length
+
     @property
     def chat_template(self) -> Optional[str]:
         return self._chat_template
@@ -287,7 +293,8 @@ class LlamaCppBackend:
                         continue
                     gpus.append((idx, free_mib))
             return gpus
-        except Exception:
+        except Exception as e:
+            logger.debug(f"Failed to query GPU free memory via nvidia-smi: {e}")
             return []
 
     @staticmethod
@@ -328,6 +335,11 @@ class LlamaCppBackend:
                 return sorted(selected), False
 
         # Model is too large even for all GPUs, let --fit handle it
+        logger.debug(
+            "Model does not fit in available GPU memory, falling back to --fit",
+            model_size_mib = round(model_size_mib, 2),
+            ranked_gpus = ranked,
+        )
         return None, True
 
     # ── KV cache VRAM estimation ─────────────────────────────────────
@@ -386,6 +398,11 @@ class LlamaCppBackend:
         If the model weights alone don't fit, returns min_ctx unchanged.
         """
         if not self._can_estimate_kv():
+            logger.debug(
+                "Skipping context fit because KV cache metadata is unavailable",
+                requested_ctx = requested_ctx,
+                available_mib = available_mib,
+            )
             return requested_ctx
 
         budget_bytes = available_mib * 1024 * 1024 * 0.70
@@ -399,6 +416,12 @@ class LlamaCppBackend:
         # Model weights alone exceed budget -- can't help by reducing ctx.
         # Return requested_ctx unchanged; --fit will handle VRAM management.
         if model_footprint >= budget_bytes:
+            logger.debug(
+                "Model footprint exceeds GPU budget before KV cache",
+                requested_ctx = requested_ctx,
+                available_mib = available_mib,
+                model_size_gb = round(model_footprint / (1024**3), 2),
+            )
             return requested_ctx
 
         # Binary search for max context that fits
@@ -960,7 +983,11 @@ class LlamaCppBackend:
 
             self._port = self._find_free_port()
 
-            # Select GPU(s) based on model size + estimated KV cache
+            # Select GPU(s) based on model size + estimated KV cache.
+            # Seed safe defaults before GPU probing so the except path
+            # still has valid state to publish.
+            effective_ctx = n_ctx if n_ctx > 0 else (self._context_length or 0)
+            max_available_ctx = self._context_length or effective_ctx
             try:
                 model_size = self._get_gguf_size_bytes(model_path)
                 gpus = self._get_gpu_free_memory()
@@ -975,6 +1002,9 @@ class LlamaCppBackend:
                 else:
                     effective_ctx = 0
                 original_ctx = effective_ctx
+                # Default UI ceiling to the model's native context length.
+                # GPU/VRAM-fit logic below may shrink this if hardware is limited.
+                max_available_ctx = self._context_length or effective_ctx
 
                 # Auto-cap context to fit in GPU VRAM and select GPUs.
                 #
@@ -993,6 +1023,29 @@ class LlamaCppBackend:
                 explicit_ctx = n_ctx > 0
 
                 if gpus and self._can_estimate_kv() and effective_ctx > 0:
+                    # Compute the largest hardware-aware cap from the model's
+                    # native context across all usable GPU subsets (for UI
+                    # bounds), independent of the currently requested context.
+                    native_ctx_for_cap = self._context_length or effective_ctx
+                    if native_ctx_for_cap > 0:
+                        ranked_for_cap = sorted(gpus, key = lambda g: g[1], reverse = True)
+                        best_cap = 0
+                        for n_gpus in range(1, len(ranked_for_cap) + 1):
+                            subset = ranked_for_cap[:n_gpus]
+                            pool_mib = sum(free for _, free in subset)
+                            capped = self._fit_context_to_vram(
+                                native_ctx_for_cap,
+                                pool_mib,
+                                model_size,
+                                cache_type_kv,
+                            )
+                            kv = self._estimate_kv_cache_bytes(capped, cache_type_kv)
+                            total_mib = (model_size + kv) / (1024 * 1024)
+                            if total_mib <= pool_mib * 0.70:
+                                best_cap = max(best_cap, capped)
+                        if best_cap > 0:
+                            max_available_ctx = best_cap
+
                     if explicit_ctx:
                         # Try to honor the user's requested context exactly.
                         requested_total = model_size + self._estimate_kv_cache_bytes(
@@ -1043,7 +1096,13 @@ class LlamaCppBackend:
                                 break
 
                 elif gpus:
-                    # Can't estimate KV -- fall back to file-size-only check
+                    # Can't estimate KV -- fall back to file-size-only check.
+                    # Without KV estimation we cannot prove a hardware cap, so
+                    # keep the ceiling at the native context (already the default).
+                    logger.debug(
+                        "Falling back to file-size-only GPU selection",
+                        model_size_gb = round(model_size / (1024**3), 2),
+                    )
                     gpu_indices, use_fit = self._select_gpus(model_size, gpus)
 
                 if effective_ctx < original_ctx:
@@ -1313,6 +1372,11 @@ class LlamaCppBackend:
             self._effective_context_length = (
                 effective_ctx if effective_ctx > 0 else self._context_length
             )
+            self._max_context_length = (
+                max_available_ctx
+                if max_available_ctx > 0
+                else self._effective_context_length
+            )
 
             # Wait for llama-server to become healthy
             if not self._wait_for_health(timeout = 600.0):
@@ -1347,6 +1411,7 @@ class LlamaCppBackend:
             self._healthy = False
             self._context_length = None
             self._effective_context_length = None
+            self._max_context_length = None
             self._chat_template = None
             self._supports_reasoning = False
             self._reasoning_always_on = False

studio/backend/core/inference/orchestrator.py

@@ -27,6 +27,7 @@ import uuid
 from io import BytesIO
 from pathlib import Path
 from typing import Any, Generator, Optional, Tuple, Union
+from utils.hardware import prepare_gpu_selection
 
 logger = get_logger(__name__)
 
@@ -571,6 +572,7 @@ class InferenceOrchestrator:
         load_in_4bit: bool = True,
         hf_token: Optional[str] = None,
         trust_remote_code: bool = False,
+        gpu_ids: Optional[list[int]] = None,
     ) -> bool:
         """Load a model for inference.
 
@@ -594,7 +596,16 @@ class InferenceOrchestrator:
                 "hf_token": hf_token or "",
                 "gguf_variant": getattr(config, "gguf_variant", None),
                 "trust_remote_code": trust_remote_code,
+                "gpu_ids": gpu_ids,
             }
+            resolved_gpu_ids, gpu_selection = prepare_gpu_selection(
+                gpu_ids,
+                model_name = model_name,
+                hf_token = hf_token,
+                load_in_4bit = load_in_4bit,
+            )
+            sub_config["resolved_gpu_ids"] = resolved_gpu_ids
+            sub_config["gpu_selection"] = gpu_selection
 
             # Always kill existing subprocess and spawn fresh.
             # Reusing a subprocess after unsloth patches torch internals

studio/backend/core/inference/worker.py

@@ -29,6 +29,7 @@ from pathlib import Path
 from typing import Any
 
 logger = get_logger(__name__)
+from utils.hardware import apply_gpu_ids
 
 
 def _activate_transformers_version(model_name: str) -> None:
@@ -178,6 +179,7 @@ def _handle_load(backend, config: dict, resp_queue: Any) -> None:
             load_in_4bit = load_in_4bit,
             hf_token = hf_token,
             trust_remote_code = trust_remote_code,
+            gpu_ids = config.get("resolved_gpu_ids"),
         )
 
         if success:
@@ -501,6 +503,8 @@ def run_inference_process(
         env = os.getenv("ENVIRONMENT_TYPE", "production"),
     )
 
+    apply_gpu_ids(config.get("resolved_gpu_ids"))
+
     model_name = config["model_name"]
 
     # ── 1. Activate correct transformers version BEFORE any ML imports ──

studio/backend/core/training/trainer.py

@@ -33,7 +33,14 @@ if sys.platform in ("win32", "darwin"):
         sys.path.insert(0, _compile_cache)
 
 import torch
-from utils.hardware import clear_gpu_cache, safe_num_proc, dataset_map_num_proc
+from utils.hardware import (
+    clear_gpu_cache,
+    safe_num_proc,
+    dataset_map_num_proc,
+    get_device_map,
+    raise_if_offloaded,
+    get_visible_gpu_count,
+)
 
 torch._dynamo.config.recompile_limit = 64
 from unsloth import FastLanguageModel, FastVisionModel, is_bfloat16_supported
@@ -487,6 +494,7 @@ class UnslothTrainer:
         is_dataset_audio: bool = False,
         trust_remote_code: bool = False,
         full_finetuning: bool = False,
+        gpu_ids: Optional[list[int]] = None,
     ) -> bool:
         """Load model for training (supports both text and vision models)"""
         self.load_in_4bit = load_in_4bit  # Store for training_meta.json
@@ -624,6 +632,11 @@ class UnslothTrainer:
                         self._update_progress(error = friendly, is_training = False)
                         return False
 
+            device_map = get_device_map(gpu_ids)
+            logger.info(
+                f"Using device_map='{device_map}' ({get_visible_gpu_count()} GPU(s) visible)"
+            )
+
             # Branch based on model type
             if self._audio_type == "csm":
                 # CSM: FastModel + auto_model=CsmForConditionalGeneration + load_in_4bit=False
@@ -636,6 +649,7 @@ class UnslothTrainer:
                     dtype = None,
                     auto_model = CsmForConditionalGeneration,
                     load_in_4bit = False,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -651,6 +665,7 @@ class UnslothTrainer:
                     model_name = model_name,
                     dtype = None,
                     load_in_4bit = False,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     auto_model = WhisperForConditionalGeneration,
                     whisper_language = "English",
@@ -672,6 +687,7 @@ class UnslothTrainer:
                     max_seq_length = max_seq_length,
                     dtype = None,
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -711,6 +727,7 @@ class UnslothTrainer:
                     max_seq_length = max_seq_length,
                     dtype = torch.float32,  # Spark-TTS requires float32
                     load_in_4bit = False,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -725,6 +742,7 @@ class UnslothTrainer:
                     model_name,
                     max_seq_length = max_seq_length,
                     load_in_4bit = False,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -741,6 +759,7 @@ class UnslothTrainer:
                     max_seq_length = max_seq_length,
                     dtype = None,
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -754,6 +773,7 @@ class UnslothTrainer:
                     max_seq_length = max_seq_length,
                     dtype = None,  # Auto-detect
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
@@ -786,12 +806,15 @@ class UnslothTrainer:
                     max_seq_length = max_seq_length,
                     dtype = None,  # Auto-detect
                     load_in_4bit = load_in_4bit,
+                    device_map = device_map,
                     full_finetuning = full_finetuning,
                     token = hf_token,
                     trust_remote_code = trust_remote_code,
                 )
                 logger.info("Loaded text model")
 
+            raise_if_offloaded(self.model, device_map, "Studio training")
+
             if self.should_stop:
                 return False
 
@@ -824,6 +847,7 @@ class UnslothTrainer:
                     is_dataset_audio = is_dataset_audio,
                     trust_remote_code = trust_remote_code,
                     full_finetuning = full_finetuning,
+                    gpu_ids = gpu_ids,
                 )
             error_msg = str(e)
             error_lower = error_msg.lower()

studio/backend/core/training/training.py

@@ -28,6 +28,7 @@ from pathlib import Path
 from typing import Optional, Tuple, Any
 
 import matplotlib.pyplot as plt
+from utils.hardware import prepare_gpu_selection
 
 logger = get_logger(__name__)
 
@@ -185,6 +186,7 @@ class TrainingBackend:
             "enable_tensorboard": kwargs.get("enable_tensorboard", False),
             "tensorboard_dir": kwargs.get("tensorboard_dir", "runs"),
             "trust_remote_code": kwargs.get("trust_remote_code", False),
+            "gpu_ids": kwargs.get("gpu_ids"),
         }
 
         # Derive load_in_4bit from training_type
@@ -192,6 +194,22 @@ class TrainingBackend:
             config["load_in_4bit"] = False
 
         # Spawn subprocess — use locals so state is untouched on failure
+        resolved_gpu_ids, gpu_selection = prepare_gpu_selection(
+            kwargs.get("gpu_ids"),
+            model_name = config["model_name"],
+            hf_token = config["hf_token"] or None,
+            training_type = config["training_type"],
+            load_in_4bit = config["load_in_4bit"],
+            batch_size = config.get("batch_size", 4),
+            max_seq_length = config.get("max_seq_length", 2048),
+            lora_rank = config.get("lora_r", 16),
+            target_modules = config.get("target_modules"),
+            gradient_checkpointing = config.get("gradient_checkpointing", "unsloth"),
+            optimizer = config.get("optim", "adamw_8bit"),
+        )
+        config["resolved_gpu_ids"] = resolved_gpu_ids
+        config["gpu_selection"] = gpu_selection
+
         from .worker import run_training_process
 
         event_queue = _CTX.Queue()

studio/backend/core/training/worker.py

@@ -29,6 +29,7 @@ import urllib.error
 import urllib.request
 
 logger = get_logger(__name__)
+from utils.hardware import apply_gpu_ids
 
 
 _CAUSAL_CONV1D_RELEASE_TAG = "v1.6.1.post4"
@@ -367,6 +368,8 @@ def run_training_process(
         env = os.getenv("ENVIRONMENT_TYPE", "production"),
     )
 
+    apply_gpu_ids(config.get("resolved_gpu_ids"))
+
     model_name = config["model_name"]
 
     # ── 1. Activate correct transformers version BEFORE any ML imports ──
@@ -682,6 +685,7 @@ def run_training_process(
             is_dataset_image = config.get("is_dataset_image", False),
             is_dataset_audio = config.get("is_dataset_audio", False),
             trust_remote_code = config.get("trust_remote_code", False),
+            gpu_ids = config.get("resolved_gpu_ids"),
         )
         if not success or trainer.should_stop:
             if trainer.should_stop:

studio/backend/main.py

@@ -68,7 +68,12 @@ from routes import (
 )
 from auth import storage
 from auth.authentication import get_current_subject
-from utils.hardware import detect_hardware, get_device, DeviceType
+from utils.hardware import (
+    detect_hardware,
+    get_device,
+    DeviceType,
+    get_backend_visible_gpu_info,
+)
 import utils.hardware.hardware as _hw_module
 
 from utils.cache_cleanup import clear_unsloth_compiled_cache
@@ -237,69 +242,14 @@ async def shutdown_server(
 async def get_system_info():
     """Get system information"""
     import platform
-    import subprocess
     import psutil
-    from utils.hardware import get_device, get_gpu_memory_info, DeviceType
+    from utils.hardware import get_device
 
-    # GPU Info — query nvidia-smi for physical GPUs, filtered by
-    # CUDA_VISIBLE_DEVICES when set (the frontend uses this for GGUF
-    # fit estimation and llama-server respects CVD too).
-    import os
-
-    gpu_info: dict = {"available": False, "devices": []}
-
-    device = get_device()
-    if device == DeviceType.CUDA:
-        # Parse CUDA_VISIBLE_DEVICES allowlist
-        allowed_indices = None
-        cvd = os.environ.get("CUDA_VISIBLE_DEVICES")
-        if cvd is not None and cvd.strip():
-            try:
-                allowed_indices = set(int(x.strip()) for x in cvd.split(","))
-            except ValueError:
-                pass  # Non-numeric (e.g. GPU-uuid), show all
-
-        try:
-            result = subprocess.run(
-                [
-                    "nvidia-smi",
-                    "--query-gpu=index,name,memory.total",
-                    "--format=csv,noheader,nounits",
-                ],
-                capture_output = True,
-                text = True,
-                timeout = 10,
-            )
-            if result.returncode == 0:
-                for line in result.stdout.strip().splitlines():
-                    parts = [p.strip() for p in line.split(",")]
-                    if len(parts) == 3:
-                        idx = int(parts[0])
-                        if allowed_indices is not None and idx not in allowed_indices:
-                            continue
-                        gpu_info["devices"].append(
-                            {
-                                "index": idx,
-                                "name": parts[1],
-                                "memory_total_gb": round(int(parts[2]) / 1024, 2),
-                            }
-                        )
-                gpu_info["available"] = len(gpu_info["devices"]) > 0
-        except Exception:
-            pass
-
-    # Fallback to torch-based single-GPU detection
-    if not gpu_info["available"]:
-        mem_info = get_gpu_memory_info()
-        if mem_info.get("available"):
-            gpu_info["available"] = True
-            gpu_info["devices"].append(
-                {
-                    "index": mem_info.get("device", 0),
-                    "name": mem_info.get("device_name", "Unknown"),
-                    "memory_total_gb": round(mem_info.get("total_gb", 0), 2),
-                }
-            )
+    visibility_info = get_backend_visible_gpu_info()
+    gpu_info = {
+        "available": visibility_info["available"],
+        "devices": visibility_info["devices"],
+    }
 
     # CPU & Memory
     memory = psutil.virtual_memory()
@@ -318,6 +268,13 @@ async def get_system_info():
     }
 
 
+@app.get("/api/system/gpu-visibility")
+async def get_gpu_visibility(
+    current_subject: str = Depends(get_current_subject),
+):
+    return get_backend_visible_gpu_info()
+
+
 @app.get("/api/system/hardware")
 async def get_hardware_info():
     """Return GPU name, total VRAM, and key ML package versions."""

studio/backend/models/inference.py

@@ -44,6 +44,10 @@ class LoadRequest(BaseModel):
         None,
         description = "KV cache data type for both K and V (e.g. 'f16', 'bf16', 'q8_0', 'q4_1', 'q5_1')",
     )
+    gpu_ids: Optional[List[int]] = Field(
+        None,
+        description = "Physical GPU indices to use, for example [0, 1]. Omit or pass [] to use automatic selection. Explicit gpu_ids are unsupported when the parent CUDA_VISIBLE_DEVICES uses UUID/MIG entries. Not supported for GGUF models.",
+    )
 
 
 class UnloadRequest(BaseModel):
@@ -132,6 +136,9 @@ class LoadResponse(BaseModel):
     context_length: Optional[int] = Field(
         None, description = "Model's native context length (from GGUF metadata)"
     )
+    max_context_length: Optional[int] = Field(
+        None, description = "Maximum context length currently available on this hardware"
+    )
     supports_reasoning: bool = Field(
         False,
         description = "Whether model supports thinking/reasoning mode (enable_thinking)",
@@ -206,6 +213,10 @@ class InferenceStatusResponse(BaseModel):
     context_length: Optional[int] = Field(
         None, description = "Context length of the active model"
     )
+    max_context_length: Optional[int] = Field(
+        None,
+        description = "Maximum context length currently available for the active model",
+    )
 
 
 # =====================================================================

studio/backend/models/training.py

@@ -128,6 +128,12 @@ class TrainingStartRequest(BaseModel):
     enable_tensorboard: bool = Field(False, description = "Enable TensorBoard logging")
     tensorboard_dir: Optional[str] = Field(None, description = "TensorBoard directory")
 
+    # GPU selection
+    gpu_ids: Optional[List[int]] = Field(
+        None,
+        description = "Physical GPU indices to use, for example [0, 1]. Omit or pass [] to use automatic selection. Explicit gpu_ids are unsupported when the parent CUDA_VISIBLE_DEVICES uses UUID/MIG entries.",
+    )
+
 
 class TrainingJobResponse(BaseModel):
     """Immediate response when training is initiated"""

studio/backend/routes/inference.py

@@ -160,6 +160,7 @@ async def load_model(
                     else False,
                     inference = inference_config,
                     context_length = llama_backend.context_length,
+                    max_context_length = llama_backend.max_context_length,
                     supports_reasoning = llama_backend.supports_reasoning,
                     reasoning_always_on = llama_backend.reasoning_always_on,
                     chat_template = llama_backend.chat_template,
@@ -210,8 +211,17 @@ async def load_model(
                 detail = f"Invalid model identifier: {request.model_path}",
             )
 
+        # Normalize gpu_ids: empty list means auto-selection, same as None
+        effective_gpu_ids = request.gpu_ids if request.gpu_ids else None
+
         # ── GGUF path: load via llama-server ──────────────────────
         if config.is_gguf:
+            if effective_gpu_ids is not None:
+                raise HTTPException(
+                    status_code = 400,
+                    detail = "gpu_ids is not supported for GGUF models yet.",
+                )
+
             llama_backend = get_llama_cpp_backend()
             unsloth_backend = get_inference_backend()
 
@@ -285,6 +295,7 @@ async def load_model(
                 has_audio_input = is_audio_input_type(_gguf_audio),
                 inference = inference_config,
                 context_length = llama_backend.context_length,
+                max_context_length = llama_backend.max_context_length,
                 supports_reasoning = llama_backend.supports_reasoning,
                 reasoning_always_on = llama_backend.reasoning_always_on,
                 supports_tools = llama_backend.supports_tools,
@@ -372,6 +383,7 @@ async def load_model(
             load_in_4bit = load_in_4bit,
             hf_token = request.hf_token,
             trust_remote_code = request.trust_remote_code,
+            gpu_ids = effective_gpu_ids,
         )
 
         if not success:
@@ -423,6 +435,9 @@ async def load_model(
 
     except HTTPException:
         raise
+    except ValueError as e:
+        logger.warning("Rejected inference GPU selection: %s", e)
+        raise HTTPException(status_code = 400, detail = str(e))
     except Exception as e:
         logger.error(f"Error loading model: {e}", exc_info = True)
         msg = str(e)
@@ -619,6 +634,7 @@ async def get_status(
                 reasoning_always_on = llama_backend.reasoning_always_on,
                 supports_tools = llama_backend.supports_tools,
                 context_length = llama_backend.context_length,
+                max_context_length = llama_backend.max_context_length,
             )
 
         # Otherwise, report Unsloth backend status

studio/backend/routes/training.py

@@ -88,14 +88,22 @@ async def get_hardware_utilization(
     Get a live snapshot of GPU hardware utilization.
 
     Designed to be polled by the frontend during training.
-    Returns GPU utilization %, temperature, VRAM usage, and power draw
-    via nvidia-smi for maximum accuracy.
+    Returns live GPU memory usage information for the active backend.
     """
     from utils.hardware import get_gpu_utilization
 
     return get_gpu_utilization()
 
 
+@router.get("/hardware/visible")
+async def get_visible_hardware_utilization(
+    current_subject: str = Depends(get_current_subject),
+):
+    from utils.hardware import get_visible_gpu_utilization
+
+    return get_visible_gpu_utilization()
+
+
 @router.post("/start")
 async def start_training(
     request: TrainingStartRequest,
@@ -202,6 +210,7 @@ async def start_training(
             "enable_tensorboard": request.enable_tensorboard,
             "tensorboard_dir": request.tensorboard_dir or "",
             "trust_remote_code": request.trust_remote_code,
+            "gpu_ids": request.gpu_ids,
         }
 
         # Training page has no trust_remote_code toggle — the value comes from
@@ -269,6 +278,9 @@ async def start_training(
             error = None,
         )
 
+    except ValueError as e:
+        logger.warning("Rejected training GPU selection: %s", e)
+        raise HTTPException(status_code = 400, detail = str(e))
     except Exception as e:
         logger.error(f"Error starting training: {e}", exc_info = True)
         raise HTTPException(

studio/backend/run.py

@@ -312,10 +312,10 @@ def run_server(
     if frontend_path:
         if setup_frontend(app, frontend_path):
             if not silent:
-                print(f"✅ Frontend loaded from {frontend_path}")
+                print(f"[OK] Frontend loaded from {frontend_path}")
         else:
             if not silent:
-                print(f"⚠️ Frontend not found at {frontend_path}")
+                print(f"[WARNING] Frontend not found at {frontend_path}")
 
     # Create the uvicorn server and expose it for signal handlers
     config = uvicorn.Config(

studio/backend/startup_banner.py

@@ -20,7 +20,7 @@ def stdout_supports_color() -> bool:
         return True
     try:
         return sys.stdout.isatty()
-    except Exception:
+    except (AttributeError, OSError, ValueError):
         return False
 
 
@@ -52,28 +52,36 @@ def print_studio_access_banner(
 
     ipv6_bind = bind_host in ("::", "::1")
     if ipv6_bind:
-        local_url = f"http://[::1]:{port}"
+        loopback_url = f"http://[::1]:{port}"
         alt_local = f"http://localhost:{port}"
     else:
-        local_url = f"http://127.0.0.1:{port}"
+        loopback_url = f"http://127.0.0.1:{port}"
         alt_local = f"http://localhost:{port}"
     if ":" in display_host:
         external_url = f"http://[{display_host}]:{port}"
     else:
         external_url = f"http://{display_host}:{port}"
+
     listen_all = bind_host in ("0.0.0.0", "::")
     loopback_bind = bind_host in ("127.0.0.1", "localhost", "::1")
-    api_base = local_url if listen_all or loopback_bind else external_url
+
+    # Use loopback URL only when the server is reachable on loopback;
+    # otherwise show the actual bound address.
+    primary_url = loopback_url if listen_all or loopback_bind else external_url
+    tip_url = alt_local if listen_all or loopback_bind else external_url
+    api_base = primary_url
 
     lines: list[str] = [
         "",
         style("🦥 Unsloth Studio is running", title),
         style("─" * 52, dim),
-        style("  On this machine — open this in your browser:", dim),
-        style(f"    {local_url}", local_url_style),
-        style(f"    (same as {alt_local})", dim),
+        style("  On this machine -- open this in your browser:", dim),
+        style(f"    {primary_url}", local_url_style),
     ]
 
+    if (listen_all or loopback_bind) and primary_url != alt_local:
+        lines.append(style(f"    (same as {alt_local})", dim))
+
     if listen_all and display_host not in (
         "127.0.0.1",
         "localhost",
@@ -88,7 +96,7 @@ def print_studio_access_banner(
                 style(f"    {external_url}", secondary),
             ]
         )
-    elif not listen_all and bind_host not in ("127.0.0.1", "localhost", "::1"):
+    elif not listen_all and not loopback_bind and external_url != primary_url:
         lines.extend(
             [
                 "",
@@ -105,7 +113,7 @@ def print_studio_access_banner(
             style(f"    {api_base}/api/health", secondary),
             style("─" * 52, dim),
             style(
-                "  Tip: if you are on the same computer, use the Local link above.",
+                f"  Tip: if you are on this computer, open {tip_url}/ in your browser.",
                 dim,
             ),
             "",

studio/backend/tests/test_gpu_selection_sandbox.py

@@ -0,0 +1,544 @@
+#!/usr/bin/env python3
+"""
+Sandbox test for multi-GPU selection logic.
+
+Tests the core GPU selection, memory estimation, and device_map logic
+in an isolated environment. Can be run on Linux, macOS, and Windows
+without requiring actual GPUs -- all hardware calls are mocked.
+
+Usage:
+    python -m pytest studio/backend/tests/test_gpu_selection_sandbox.py -v
+    # or directly:
+    python studio/backend/tests/test_gpu_selection_sandbox.py
+"""
+
+import os
+import sys
+import unittest
+from pathlib import Path
+from unittest.mock import patch, MagicMock
+
+# Ensure backend is on sys.path
+_backend_root = Path(__file__).resolve().parent.parent
+if str(_backend_root) not in sys.path:
+    sys.path.insert(0, str(_backend_root))
+
+
+def _make_fake_config(
+    vocab_size = 32000,
+    hidden_size = 4096,
+    intermediate_size = 11008,
+    num_hidden_layers = 32,
+    num_attention_heads = 32,
+    num_key_value_heads = 8,
+    tie_word_embeddings = False,
+):
+    """Create a fake HF config-like object for estimation tests."""
+    from types import SimpleNamespace
+
+    return SimpleNamespace(
+        vocab_size = vocab_size,
+        hidden_size = hidden_size,
+        intermediate_size = intermediate_size,
+        num_hidden_layers = num_hidden_layers,
+        num_attention_heads = num_attention_heads,
+        num_key_value_heads = num_key_value_heads,
+        tie_word_embeddings = tie_word_embeddings,
+    )
+
+
+class TestEstimateFP16ModelSizeFromConfig(unittest.TestCase):
+    """Test the config-based model size estimation."""
+
+    def test_llama_8b_size_reasonable(self):
+        from utils.hardware.hardware import _estimate_fp16_model_size_bytes_from_config
+
+        config = _make_fake_config(
+            vocab_size = 128256,
+            hidden_size = 4096,
+            intermediate_size = 14336,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            tie_word_embeddings = False,
+        )
+        size = _estimate_fp16_model_size_bytes_from_config(config)
+        self.assertIsNotNone(size)
+        size_gb = size / (1024**3)
+        # Llama 3.1 8B should be ~15GB in fp16
+        self.assertGreater(size_gb, 12)
+        self.assertLess(size_gb, 20)
+
+    def test_small_model(self):
+        from utils.hardware.hardware import _estimate_fp16_model_size_bytes_from_config
+
+        config = _make_fake_config(
+            vocab_size = 32000,
+            hidden_size = 2048,
+            intermediate_size = 5504,
+            num_hidden_layers = 22,
+            num_attention_heads = 32,
+            num_key_value_heads = 4,
+        )
+        size = _estimate_fp16_model_size_bytes_from_config(config)
+        self.assertIsNotNone(size)
+        size_gb = size / (1024**3)
+        # ~1B model should be ~2GB in fp16
+        self.assertGreater(size_gb, 1)
+        self.assertLess(size_gb, 5)
+
+    def test_returns_none_for_incomplete_config(self):
+        from utils.hardware.hardware import _estimate_fp16_model_size_bytes_from_config
+        from types import SimpleNamespace
+
+        config = SimpleNamespace(vocab_size = 32000)  # Missing most fields
+        size = _estimate_fp16_model_size_bytes_from_config(config)
+        self.assertIsNone(size)
+
+    def test_moe_model(self):
+        from utils.hardware.hardware import _estimate_fp16_model_size_bytes_from_config
+        from types import SimpleNamespace
+
+        config = SimpleNamespace(
+            vocab_size = 152064,
+            hidden_size = 3584,
+            intermediate_size = 18944,
+            num_hidden_layers = 28,
+            num_attention_heads = 28,
+            num_key_value_heads = 4,
+            tie_word_embeddings = False,
+            num_local_experts = 64,
+            moe_intermediate_size = 2560,
+        )
+        size = _estimate_fp16_model_size_bytes_from_config(config)
+        self.assertIsNotNone(size)
+        size_gb = size / (1024**3)
+        # MoE model with 64 experts should be large
+        self.assertGreater(size_gb, 50)
+
+
+class TestEstimateRequiredModelMemory(unittest.TestCase):
+    """Test memory requirement estimation."""
+
+    def test_inference_fp16_uses_1_3x(self):
+        from utils.hardware.hardware import estimate_required_model_memory_gb
+
+        with patch(
+            "utils.hardware.hardware.estimate_fp16_model_size_bytes",
+            return_value = (10 * (1024**3), "config"),  # 10GB model
+        ):
+            required, meta = estimate_required_model_memory_gb(
+                "test/model",
+                training_type = None,  # inference
+                load_in_4bit = False,
+            )
+            self.assertIsNotNone(required)
+            self.assertAlmostEqual(required, 13.0, places = 0)
+            self.assertEqual(meta["mode"], "inference")
+
+    def test_inference_4bit_uses_reduced_estimate(self):
+        from utils.hardware.hardware import estimate_required_model_memory_gb
+
+        with patch(
+            "utils.hardware.hardware.estimate_fp16_model_size_bytes",
+            return_value = (30 * (1024**3), "config"),  # 30GB fp16 model
+        ):
+            required, meta = estimate_required_model_memory_gb(
+                "test/model",
+                training_type = None,  # inference
+                load_in_4bit = True,
+            )
+            self.assertIsNotNone(required)
+            # 4bit base = 30/3.2 = 9.375GB, required = 9.375 + max(9.375*0.3, 2) = 12.19GB
+            self.assertAlmostEqual(required, 12.2, places = 0)
+
+    def test_4bit_training_reduces_base(self):
+        from utils.hardware.hardware import estimate_required_model_memory_gb
+
+        with patch(
+            "utils.hardware.hardware.estimate_fp16_model_size_bytes",
+            return_value = (30 * (1024**3), "config"),  # 30GB fp16 model
+        ):
+            required, meta = estimate_required_model_memory_gb(
+                "test/model",
+                training_type = "LoRA/QLoRA",
+                load_in_4bit = True,
+            )
+            self.assertIsNotNone(required)
+            # fallback: base=30/3.2=9.375, lora=30*0.04=1.2, act=30*0.15=4.5, cuda=1.4
+            self.assertAlmostEqual(required, 16.5, places = 0)
+
+    def test_full_finetune_uses_3_5x(self):
+        from utils.hardware.hardware import estimate_required_model_memory_gb
+
+        with patch(
+            "utils.hardware.hardware.estimate_fp16_model_size_bytes",
+            return_value = (10 * (1024**3), "config"),  # 10GB model
+        ):
+            required, meta = estimate_required_model_memory_gb(
+                "test/model",
+                training_type = "Full Finetuning",
+            )
+            self.assertIsNotNone(required)
+            # fallback: 10 * 3.5 + 1.4 cuda overhead = 36.4
+            self.assertAlmostEqual(required, 36.4, places = 0)
+
+    def test_returns_none_when_unavailable(self):
+        from utils.hardware.hardware import estimate_required_model_memory_gb
+
+        with patch(
+            "utils.hardware.hardware.estimate_fp16_model_size_bytes",
+            return_value = (None, "unavailable"),
+        ):
+            required, meta = estimate_required_model_memory_gb("test/model")
+            self.assertIsNone(required)
+
+
+class TestAutoSelectGpuIds(unittest.TestCase):
+    """Test automatic GPU selection based on model size and free memory."""
+
+    def _make_utilization(self, devices):
+        """Create a fake utilization response."""
+        return {
+            "available": True,
+            "devices": [
+                {
+                    "index": idx,
+                    "vram_total_gb": total,
+                    "vram_used_gb": total - free,
+                }
+                for idx, total, free in devices
+            ],
+        }
+
+    def test_single_gpu_sufficient(self):
+        from utils.hardware.hardware import auto_select_gpu_ids
+        import utils.hardware.hardware as hw
+
+        with (
+            patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA),
+            patch.object(
+                hw,
+                "estimate_required_model_memory_gb",
+                return_value = (
+                    10.0,
+                    {
+                        "mode": "inference",
+                        "required_gb": 10.0,
+                        "model_size_source": "config",
+                        "model_size_gb": 7.7,
+                    },
+                ),
+            ),
+            patch.object(
+                hw,
+                "_get_parent_visible_gpu_spec",
+                return_value = {
+                    "raw": "0,1,2,3",
+                    "numeric_ids": [0, 1, 2, 3],
+                    "supports_explicit_gpu_ids": True,
+                },
+            ),
+            patch.object(hw, "get_parent_visible_gpu_ids", return_value = [0, 1, 2, 3]),
+            patch.object(
+                hw,
+                "get_visible_gpu_utilization",
+                return_value = self._make_utilization(
+                    [
+                        (0, 80.0, 75.0),
+                        (1, 80.0, 78.0),
+                        (2, 80.0, 70.0),
+                        (3, 80.0, 72.0),
+                    ]
+                ),
+            ),
+        ):
+            selected, meta = auto_select_gpu_ids("test/model")
+            # Should pick GPU 1 (most free memory: 78GB) -- enough for 10GB
+            self.assertEqual(len(selected), 1)
+            self.assertEqual(selected[0], 1)
+
+    def test_two_gpus_needed(self):
+        from utils.hardware.hardware import auto_select_gpu_ids
+        import utils.hardware.hardware as hw
+
+        with (
+            patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA),
+            patch.object(
+                hw,
+                "estimate_required_model_memory_gb",
+                return_value = (
+                    50.0,
+                    {
+                        "mode": "inference",
+                        "required_gb": 50.0,
+                        "model_size_source": "config",
+                        "model_size_gb": 38.0,
+                    },
+                ),
+            ),
+            patch.object(
+                hw,
+                "_get_parent_visible_gpu_spec",
+                return_value = {
+                    "raw": "0,1",
+                    "numeric_ids": [0, 1],
+                    "supports_explicit_gpu_ids": True,
+                },
+            ),
+            patch.object(hw, "get_parent_visible_gpu_ids", return_value = [0, 1]),
+            patch.object(
+                hw,
+                "get_visible_gpu_utilization",
+                return_value = self._make_utilization(
+                    [
+                        (0, 40.0, 30.0),  # 30GB free
+                        (1, 40.0, 35.0),  # 35GB free
+                    ]
+                ),
+            ),
+        ):
+            selected, meta = auto_select_gpu_ids("test/model")
+            # 35GB (first) + 30*0.85 (second) = 60.5GB > 50GB
+            self.assertEqual(len(selected), 2)
+
+    def test_non_cuda_returns_none(self):
+        from utils.hardware.hardware import auto_select_gpu_ids
+        import utils.hardware.hardware as hw
+
+        with patch.object(hw, "get_device", return_value = hw.DeviceType.CPU):
+            selected, meta = auto_select_gpu_ids("test/model")
+            self.assertIsNone(selected)
+            self.assertEqual(meta["selection_mode"], "non_cuda")
+
+
+class TestGetDeviceMap(unittest.TestCase):
+    """Test device_map string generation."""
+
+    def test_single_gpu_returns_sequential(self):
+        from utils.hardware.hardware import get_device_map
+        import utils.hardware.hardware as hw
+
+        with (
+            patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA),
+            patch.object(
+                hw,
+                "_get_parent_visible_gpu_spec",
+                return_value = {
+                    "raw": "0",
+                    "numeric_ids": [0],
+                    "supports_explicit_gpu_ids": True,
+                },
+            ),
+            patch.object(hw, "get_visible_gpu_count", return_value = 1),
+        ):
+            dm = get_device_map(gpu_ids = [0])
+            self.assertEqual(dm, "sequential")
+
+    def test_multi_gpu_returns_balanced(self):
+        from utils.hardware.hardware import get_device_map
+        import utils.hardware.hardware as hw
+
+        with patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA):
+            dm = get_device_map(gpu_ids = [0, 1])
+            self.assertEqual(dm, "balanced")
+
+    def test_cpu_returns_sequential(self):
+        from utils.hardware.hardware import get_device_map
+        import utils.hardware.hardware as hw
+
+        with patch.object(hw, "get_device", return_value = hw.DeviceType.CPU):
+            dm = get_device_map(gpu_ids = None)
+            self.assertEqual(dm, "sequential")
+
+
+class TestResolveRequestedGpuIds(unittest.TestCase):
+    """Test GPU ID validation."""
+
+    def test_none_returns_parent_visible(self):
+        from utils.hardware.hardware import resolve_requested_gpu_ids
+
+        with (
+            patch.dict(os.environ, {"CUDA_VISIBLE_DEVICES": "2,3"}, clear = False),
+            patch("utils.hardware.hardware.get_physical_gpu_count", return_value = 8),
+        ):
+            result = resolve_requested_gpu_ids(None)
+            self.assertEqual(result, [2, 3])
+
+    def test_empty_list_returns_parent_visible(self):
+        from utils.hardware.hardware import resolve_requested_gpu_ids
+
+        with (
+            patch.dict(os.environ, {"CUDA_VISIBLE_DEVICES": "2,3"}, clear = False),
+            patch("utils.hardware.hardware.get_physical_gpu_count", return_value = 8),
+        ):
+            result = resolve_requested_gpu_ids([])
+            self.assertEqual(result, [2, 3])
+
+    def test_duplicates_rejected(self):
+        from utils.hardware.hardware import resolve_requested_gpu_ids
+
+        with (
+            patch.dict(os.environ, {"CUDA_VISIBLE_DEVICES": "0,1,2"}, clear = False),
+            patch("utils.hardware.hardware.get_physical_gpu_count", return_value = 8),
+        ):
+            with self.assertRaises(ValueError):
+                resolve_requested_gpu_ids([1, 1])
+
+    def test_out_of_range_rejected(self):
+        from utils.hardware.hardware import resolve_requested_gpu_ids
+
+        with (
+            patch.dict(os.environ, {"CUDA_VISIBLE_DEVICES": "0,1"}, clear = False),
+            patch("utils.hardware.hardware.get_physical_gpu_count", return_value = 4),
+        ):
+            with self.assertRaises(ValueError):
+                resolve_requested_gpu_ids([5])
+
+    def test_uuid_env_var_rejects_explicit_ids(self):
+        from utils.hardware.hardware import resolve_requested_gpu_ids
+
+        with (
+            patch.dict(
+                os.environ, {"CUDA_VISIBLE_DEVICES": "GPU-abc,GPU-def"}, clear = False
+            ),
+            patch("utils.hardware.hardware.get_physical_gpu_count", return_value = 8),
+        ):
+            with self.assertRaises(ValueError):
+                resolve_requested_gpu_ids([0])
+
+
+class TestApplyGpuIds(unittest.TestCase):
+    """Test CUDA_VISIBLE_DEVICES environment variable setting."""
+
+    def test_apply_list(self):
+        from utils.hardware.hardware import apply_gpu_ids
+
+        with patch.dict(os.environ, {}, clear = False):
+            apply_gpu_ids([3, 5])
+            self.assertEqual(os.environ.get("CUDA_VISIBLE_DEVICES"), "3,5")
+
+    def test_apply_none_does_nothing(self):
+        from utils.hardware.hardware import apply_gpu_ids
+
+        original = os.environ.get("CUDA_VISIBLE_DEVICES")
+        apply_gpu_ids(None)
+        self.assertEqual(os.environ.get("CUDA_VISIBLE_DEVICES"), original)
+
+
+class TestMultiGpuOverheadAccounting(unittest.TestCase):
+    """Test that multi-GPU overhead is applied correctly.
+
+    The first GPU should keep its full free memory, and only
+    additional GPUs should have the overhead factor applied.
+    """
+
+    def _make_utilization(self, devices):
+        return {
+            "available": True,
+            "devices": [
+                {
+                    "index": idx,
+                    "vram_total_gb": total,
+                    "vram_used_gb": total - free,
+                }
+                for idx, total, free in devices
+            ],
+        }
+
+    def test_first_gpu_not_penalized(self):
+        """A model that just fits on 1 GPU should not require 2 GPUs."""
+        from utils.hardware.hardware import auto_select_gpu_ids
+        import utils.hardware.hardware as hw
+
+        # Model requires 79GB, GPU has 80GB free
+        with (
+            patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA),
+            patch.object(
+                hw,
+                "estimate_required_model_memory_gb",
+                return_value = (
+                    79.0,
+                    {
+                        "mode": "inference",
+                        "required_gb": 79.0,
+                        "model_size_source": "config",
+                        "model_size_gb": 60.0,
+                    },
+                ),
+            ),
+            patch.object(
+                hw,
+                "_get_parent_visible_gpu_spec",
+                return_value = {
+                    "raw": "0,1",
+                    "numeric_ids": [0, 1],
+                    "supports_explicit_gpu_ids": True,
+                },
+            ),
+            patch.object(hw, "get_parent_visible_gpu_ids", return_value = [0, 1]),
+            patch.object(
+                hw,
+                "get_visible_gpu_utilization",
+                return_value = self._make_utilization(
+                    [
+                        (0, 80.0, 80.0),
+                        (1, 80.0, 80.0),
+                    ]
+                ),
+            ),
+        ):
+            selected, meta = auto_select_gpu_ids("test/model")
+            # Should fit on 1 GPU (80GB >= 79GB)
+            self.assertEqual(len(selected), 1)
+
+    def test_second_gpu_has_overhead(self):
+        """When 2 GPUs are needed, the second one's contribution is reduced."""
+        from utils.hardware.hardware import auto_select_gpu_ids
+        import utils.hardware.hardware as hw
+
+        # Model requires 110GB. First GPU has 80GB, second has 40GB.
+        # With overhead: 80 + 40*0.85 = 114GB -- just enough
+        with (
+            patch.object(hw, "get_device", return_value = hw.DeviceType.CUDA),
+            patch.object(
+                hw,
+                "estimate_required_model_memory_gb",
+                return_value = (
+                    110.0,
+                    {
+                        "mode": "inference",
+                        "required_gb": 110.0,
+                        "model_size_source": "config",
+                        "model_size_gb": 85.0,
+                    },
+                ),
+            ),
+            patch.object(
+                hw,
+                "_get_parent_visible_gpu_spec",
+                return_value = {
+                    "raw": "0,1",
+                    "numeric_ids": [0, 1],
+                    "supports_explicit_gpu_ids": True,
+                },
+            ),
+            patch.object(hw, "get_parent_visible_gpu_ids", return_value = [0, 1]),
+            patch.object(
+                hw,
+                "get_visible_gpu_utilization",
+                return_value = self._make_utilization(
+                    [
+                        (0, 80.0, 80.0),
+                        (1, 80.0, 40.0),
+                    ]
+                ),
+            ),
+        ):
+            selected, meta = auto_select_gpu_ids("test/model")
+            # Should use both GPUs
+            self.assertEqual(len(selected), 2)
+
+
+if __name__ == "__main__":
+    unittest.main()

studio/backend/tests/test_utils.py

@@ -285,7 +285,7 @@ class TestLogGpuMemory:
     def test_does_not_raise(self):
         log_gpu_memory("test")
 
-    def test_logs_gpu_info_when_available(self, caplog):
+    def test_logs_gpu_info_when_available(self, capfd):
         fake_info = {
             "available": True,
             "backend": "cuda",
@@ -295,35 +295,27 @@ class TestLogGpuMemory:
             "utilization_pct": 12.5,
             "free_gb": 14.0,
         }
-        import structlog
-        from loggers import get_logger
 
-        with (
-            patch(
-                "utils.hardware.hardware.get_gpu_memory_info", return_value = fake_info
-            ),
-            caplog.at_level(logging.INFO, logger = "utils.hardware.hardware"),
+        with patch(
+            "utils.hardware.hardware.get_gpu_memory_info", return_value = fake_info
         ):
             log_gpu_memory("unit-test")
 
-        assert "unit-test" in caplog.text
-        assert "CUDA" in caplog.text
-        assert "FakeGPU" in caplog.text
+        captured = capfd.readouterr()
+        assert "unit-test" in captured.out
+        assert "CUDA" in captured.out
+        assert "FakeGPU" in captured.out
 
-    def test_logs_cpu_fallback_when_no_gpu(self, caplog):
+    def test_logs_cpu_fallback_when_no_gpu(self, capfd):
         fake_info = {"available": False, "backend": "cpu"}
-        import structlog
-        from loggers import get_logger
 
-        with (
-            patch(
-                "utils.hardware.hardware.get_gpu_memory_info", return_value = fake_info
-            ),
-            caplog.at_level(logging.INFO, logger = "utils.hardware.hardware"),
+        with patch(
+            "utils.hardware.hardware.get_gpu_memory_info", return_value = fake_info
         ):
             log_gpu_memory("cpu-test")
 
-        assert "No GPU available" in caplog.text
+        captured = capfd.readouterr()
+        assert "No GPU available" in captured.out
 
 
 # ========== format_error_message() ==========

studio/backend/tests/test_vram_estimation.py

@@ -0,0 +1,695 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved.
+
+import unittest
+from types import SimpleNamespace
+
+from utils.hardware.vram_estimation import (
+    ModelArchConfig,
+    TrainingVramConfig,
+    extract_arch_config,
+    compute_model_weights_bytes,
+    compute_total_params,
+    compute_lora_params,
+    compute_lora_adapter_bytes,
+    compute_optimizer_bytes,
+    compute_gradient_bytes,
+    compute_activation_bytes,
+    estimate_training_vram,
+    DEFAULT_TARGET_MODULES,
+)
+
+
+def _gb(b: int) -> float:
+    return b / (1024**3)
+
+
+LLAMA_8B = ModelArchConfig(
+    hidden_size = 4096,
+    num_hidden_layers = 32,
+    num_attention_heads = 32,
+    num_key_value_heads = 8,
+    intermediate_size = 14336,
+    vocab_size = 128256,
+    tie_word_embeddings = False,
+)
+
+QWEN_05B = ModelArchConfig(
+    hidden_size = 896,
+    num_hidden_layers = 24,
+    num_attention_heads = 14,
+    num_key_value_heads = 2,
+    intermediate_size = 4864,
+    vocab_size = 151936,
+    tie_word_embeddings = True,
+)
+
+MOE_CONFIG = ModelArchConfig(
+    hidden_size = 4096,
+    num_hidden_layers = 32,
+    num_attention_heads = 32,
+    num_key_value_heads = 8,
+    intermediate_size = 14336,
+    vocab_size = 32000,
+    tie_word_embeddings = False,
+    num_experts = 8,
+)
+
+DEEPSEEK_V3 = ModelArchConfig(
+    hidden_size = 7168,
+    num_hidden_layers = 61,
+    num_attention_heads = 128,
+    num_key_value_heads = 128,
+    intermediate_size = 18432,
+    vocab_size = 129280,
+    tie_word_embeddings = False,
+    num_experts = 256,
+    moe_intermediate_size = 2048,
+    n_shared_experts = 1,
+    num_dense_layers = 3,
+    q_lora_rank = 1536,
+    kv_lora_rank = 512,
+    qk_nope_head_dim = 128,
+    qk_rope_head_dim = 64,
+    v_head_dim = 128,
+)
+
+QWEN3_MOE_30B = ModelArchConfig(
+    hidden_size = 2048,
+    num_hidden_layers = 48,
+    num_attention_heads = 32,
+    num_key_value_heads = 4,
+    intermediate_size = 8192,
+    vocab_size = 151936,
+    tie_word_embeddings = True,
+    num_experts = 128,
+    moe_intermediate_size = 768,
+    n_shared_experts = 0,
+    num_dense_layers = 0,
+)
+
+GLM4_MOE = ModelArchConfig(
+    hidden_size = 4096,
+    num_hidden_layers = 46,
+    num_attention_heads = 96,
+    num_key_value_heads = 8,
+    intermediate_size = 10944,
+    vocab_size = 151552,
+    tie_word_embeddings = False,
+    num_experts = 128,
+    moe_intermediate_size = 1408,
+    n_shared_experts = 1,
+    num_dense_layers = 1,
+)
+
+GPT_OSS = ModelArchConfig(
+    hidden_size = 6144,
+    num_hidden_layers = 64,
+    num_attention_heads = 64,
+    num_key_value_heads = 8,
+    intermediate_size = 2880,
+    vocab_size = 200064,
+    tie_word_embeddings = False,
+    num_experts = 128,
+    moe_intermediate_size = None,
+    n_shared_experts = 0,
+    num_dense_layers = 0,
+)
+
+
+class TestExtractArchConfig(unittest.TestCase):
+    def test_basic_config(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 4096,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 128256,
+            tie_word_embeddings = False,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertIsNotNone(arch)
+        self.assertEqual(arch.hidden_size, 4096)
+        self.assertEqual(arch.num_hidden_layers, 32)
+        self.assertEqual(arch.num_key_value_heads, 8)
+        self.assertIsNone(arch.num_experts)
+
+    def test_vlm_text_config(self):
+        text_cfg = SimpleNamespace(
+            hidden_size = 2048,
+            num_hidden_layers = 24,
+            num_attention_heads = 16,
+            num_key_value_heads = 4,
+            intermediate_size = 8192,
+            vocab_size = 32000,
+            tie_word_embeddings = True,
+        )
+        hf_config = SimpleNamespace(text_config = text_cfg)
+        arch = extract_arch_config(hf_config)
+        self.assertIsNotNone(arch)
+        self.assertEqual(arch.hidden_size, 2048)
+
+    def test_moe_detection(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 4096,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 32000,
+            tie_word_embeddings = False,
+            num_local_experts = 8,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_experts, 8)
+
+    def test_missing_fields_returns_none(self):
+        hf_config = SimpleNamespace(hidden_size = 4096)
+        arch = extract_arch_config(hf_config)
+        self.assertIsNone(arch)
+
+    def test_intermediate_size_list(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 2048,
+            num_hidden_layers = 24,
+            num_attention_heads = 16,
+            num_key_value_heads = 4,
+            intermediate_size = [8192, 8192],
+            vocab_size = 32000,
+            tie_word_embeddings = True,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.intermediate_size, 8192)
+
+
+class TestModelWeightsBytes(unittest.TestCase):
+    def test_llama_8b_fp16(self):
+        weight_bytes = compute_model_weights_bytes(LLAMA_8B, "full", False)
+        weight_gb = _gb(weight_bytes)
+        self.assertGreater(weight_gb, 14.0)
+        self.assertLess(weight_gb, 18.0)
+
+    def test_llama_8b_qlora_4bit(self):
+        weight_bytes = compute_model_weights_bytes(LLAMA_8B, "qlora", True)
+        weight_gb = _gb(weight_bytes)
+        self.assertGreater(weight_gb, 4.0)
+        self.assertLess(weight_gb, 7.0)
+
+    def test_4bit_smaller_than_fp16(self):
+        fp16 = compute_model_weights_bytes(LLAMA_8B, "full", False)
+        q4 = compute_model_weights_bytes(LLAMA_8B, "qlora", True)
+        self.assertLess(q4, fp16)
+        ratio = fp16 / q4
+        self.assertGreater(ratio, 2.0)
+        self.assertLess(ratio, 4.0)
+
+    def test_moe_larger_than_dense(self):
+        dense = compute_model_weights_bytes(LLAMA_8B, "full", False)
+        moe = compute_model_weights_bytes(MOE_CONFIG, "full", False)
+        self.assertGreater(moe, dense * 3)
+
+
+class TestLoraParams(unittest.TestCase):
+    def test_llama_8b_default_modules_rank16(self):
+        lora_p = compute_lora_params(LLAMA_8B, 16, DEFAULT_TARGET_MODULES)
+        total_p = compute_total_params(LLAMA_8B)
+        ratio = lora_p / total_p
+        self.assertGreater(ratio, 0.005)
+        self.assertLess(ratio, 0.05)
+
+    def test_higher_rank_more_params(self):
+        r16 = compute_lora_params(LLAMA_8B, 16, DEFAULT_TARGET_MODULES)
+        r64 = compute_lora_params(LLAMA_8B, 64, DEFAULT_TARGET_MODULES)
+        self.assertAlmostEqual(r64 / r16, 4.0, places = 1)
+
+    def test_fewer_modules_fewer_params(self):
+        all_mods = compute_lora_params(LLAMA_8B, 16, DEFAULT_TARGET_MODULES)
+        qv_only = compute_lora_params(LLAMA_8B, 16, ["q_proj", "v_proj"])
+        self.assertLess(qv_only, all_mods)
+
+    def test_moe_mlp_modules_scale_with_experts(self):
+        dense_lora = compute_lora_params(
+            LLAMA_8B, 16, ["gate_proj", "up_proj", "down_proj"]
+        )
+        moe_lora = compute_lora_params(
+            MOE_CONFIG, 16, ["gate_proj", "up_proj", "down_proj"]
+        )
+        ratio = moe_lora / dense_lora
+        self.assertAlmostEqual(ratio, 8.0, delta = 0.5)
+
+    def test_attention_modules_same_for_moe(self):
+        dense_attn = compute_lora_params(
+            LLAMA_8B, 16, ["q_proj", "k_proj", "v_proj", "o_proj"]
+        )
+        moe_attn = compute_lora_params(
+            MOE_CONFIG, 16, ["q_proj", "k_proj", "v_proj", "o_proj"]
+        )
+        self.assertEqual(dense_attn, moe_attn)
+
+
+class TestOptimizerBytes(unittest.TestCase):
+    def test_adamw_8bit(self):
+        self.assertEqual(compute_optimizer_bytes(1_000_000, "adamw_8bit"), 4_000_000)
+
+    def test_adamw_torch(self):
+        self.assertEqual(compute_optimizer_bytes(1_000_000, "adamw_torch"), 6_000_000)
+
+    def test_sgd(self):
+        self.assertEqual(compute_optimizer_bytes(1_000_000, "sgd"), 4_000_000)
+
+    def test_unknown_defaults_to_4(self):
+        self.assertEqual(compute_optimizer_bytes(1_000_000, "some_new_opt"), 4_000_000)
+
+
+class TestGradientBytes(unittest.TestCase):
+    def test_fp16_gradients(self):
+        self.assertEqual(compute_gradient_bytes(1_000_000), 2_000_000)
+
+
+class TestActivationBytes(unittest.TestCase):
+    def test_no_gc_scales_with_layers(self):
+        act_none = compute_activation_bytes(LLAMA_8B, 2, 2048, "none")
+        act_gc = compute_activation_bytes(LLAMA_8B, 2, 2048, "true")
+        self.assertGreater(act_none, act_gc * 10)
+
+    def test_unsloth_gc_smaller_than_standard(self):
+        act_true = compute_activation_bytes(LLAMA_8B, 2, 2048, "true")
+        act_unsloth = compute_activation_bytes(LLAMA_8B, 2, 2048, "unsloth")
+        self.assertLess(act_unsloth, act_true)
+
+    def test_lora_activations_smaller_than_full_ft(self):
+        full_ft = compute_activation_bytes(LLAMA_8B, 2, 2048, "unsloth", is_lora = False)
+        lora = compute_activation_bytes(LLAMA_8B, 2, 2048, "unsloth", is_lora = True)
+        self.assertLess(lora, full_ft)
+
+    def test_scales_with_batch_size(self):
+        act_bsz2 = compute_activation_bytes(LLAMA_8B, 2, 2048, "unsloth")
+        act_bsz4 = compute_activation_bytes(LLAMA_8B, 4, 2048, "unsloth")
+        self.assertAlmostEqual(act_bsz4 / act_bsz2, 2.0, delta = 0.1)
+
+    def test_scales_with_seq_len(self):
+        act_2k = compute_activation_bytes(LLAMA_8B, 2, 2048, "unsloth")
+        act_4k = compute_activation_bytes(LLAMA_8B, 2, 4096, "unsloth")
+        self.assertAlmostEqual(act_4k / act_2k, 2.0, delta = 0.1)
+
+
+class TestEstimateTrainingVram(unittest.TestCase):
+    def test_llama_8b_qlora_reasonable_total(self):
+        config = TrainingVramConfig(
+            training_method = "qlora",
+            batch_size = 2,
+            max_seq_length = 2048,
+            lora_rank = 16,
+            gradient_checkpointing = "unsloth",
+            optimizer = "adamw_8bit",
+            load_in_4bit = True,
+        )
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        total_gb = _gb(breakdown.total)
+        self.assertGreater(total_gb, 5.0)
+        self.assertLess(total_gb, 12.0)
+
+    def test_llama_8b_full_ft_reasonable_total(self):
+        config = TrainingVramConfig(
+            training_method = "full",
+            batch_size = 2,
+            max_seq_length = 2048,
+            gradient_checkpointing = "unsloth",
+            optimizer = "adamw_8bit",
+            load_in_4bit = False,
+        )
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        total_gb = _gb(breakdown.total)
+        self.assertGreater(total_gb, 50.0)
+        self.assertLess(total_gb, 75.0)
+
+    def test_qlora_much_less_than_full_ft(self):
+        qlora_config = TrainingVramConfig(
+            training_method = "qlora",
+            load_in_4bit = True,
+            batch_size = 2,
+            max_seq_length = 2048,
+        )
+        full_config = TrainingVramConfig(
+            training_method = "full",
+            load_in_4bit = False,
+            batch_size = 2,
+            max_seq_length = 2048,
+        )
+        qlora = estimate_training_vram(LLAMA_8B, qlora_config)
+        full = estimate_training_vram(LLAMA_8B, full_config)
+        self.assertLess(qlora.total, full.total / 3)
+
+    def test_qwen_05b_qlora_fits_in_4gb(self):
+        config = TrainingVramConfig(
+            training_method = "qlora",
+            batch_size = 2,
+            max_seq_length = 2048,
+            lora_rank = 16,
+            gradient_checkpointing = "unsloth",
+            optimizer = "adamw_8bit",
+            load_in_4bit = True,
+        )
+        breakdown = estimate_training_vram(QWEN_05B, config)
+        total_gb = _gb(breakdown.total)
+        self.assertLess(total_gb, 5.0)
+
+    def test_breakdown_components_positive(self):
+        config = TrainingVramConfig(training_method = "qlora", load_in_4bit = True)
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        self.assertGreater(breakdown.model_weights, 0)
+        self.assertGreater(breakdown.lora_adapters, 0)
+        self.assertGreater(breakdown.optimizer_states, 0)
+        self.assertGreater(breakdown.gradients, 0)
+        self.assertGreater(breakdown.activations, 0)
+        self.assertGreater(breakdown.cuda_overhead, 0)
+
+    def test_full_ft_no_lora_adapters(self):
+        config = TrainingVramConfig(training_method = "full", load_in_4bit = False)
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        self.assertEqual(breakdown.lora_adapters, 0)
+
+    def test_to_gb_dict_keys(self):
+        config = TrainingVramConfig(training_method = "qlora", load_in_4bit = True)
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        gb_dict = breakdown.to_gb_dict()
+        expected_keys = {
+            "model_weights_gb",
+            "lora_adapters_gb",
+            "optimizer_states_gb",
+            "gradients_gb",
+            "activations_gb",
+            "cuda_overhead_gb",
+            "total_gb",
+        }
+        self.assertEqual(set(gb_dict.keys()), expected_keys)
+
+    def test_total_equals_sum_of_parts(self):
+        config = TrainingVramConfig(training_method = "qlora", load_in_4bit = True)
+        breakdown = estimate_training_vram(LLAMA_8B, config)
+        parts_sum = (
+            breakdown.model_weights
+            + breakdown.lora_adapters
+            + breakdown.optimizer_states
+            + breakdown.gradients
+            + breakdown.activations
+            + breakdown.cuda_overhead
+        )
+        self.assertEqual(breakdown.total, parts_sum)
+
+    def test_larger_batch_increases_total(self):
+        small = TrainingVramConfig(
+            training_method = "qlora",
+            load_in_4bit = True,
+            batch_size = 1,
+        )
+        large = TrainingVramConfig(
+            training_method = "qlora",
+            load_in_4bit = True,
+            batch_size = 8,
+        )
+        small_v = estimate_training_vram(LLAMA_8B, small)
+        large_v = estimate_training_vram(LLAMA_8B, large)
+        self.assertGreater(large_v.total, small_v.total)
+
+    def test_adamw_fp32_uses_more_optimizer_memory(self):
+        opt8 = TrainingVramConfig(
+            training_method = "full",
+            load_in_4bit = False,
+            optimizer = "adamw_8bit",
+        )
+        opt32 = TrainingVramConfig(
+            training_method = "full",
+            load_in_4bit = False,
+            optimizer = "adamw_torch",
+        )
+        v8 = estimate_training_vram(LLAMA_8B, opt8)
+        v32 = estimate_training_vram(LLAMA_8B, opt32)
+        self.assertAlmostEqual(
+            v32.optimizer_states / v8.optimizer_states, 1.5, delta = 0.1
+        )
+
+
+class TestExtractArchConfigMoE(unittest.TestCase):
+    def test_deepseek_v3_shared_experts(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 7168,
+            num_hidden_layers = 61,
+            num_attention_heads = 128,
+            num_key_value_heads = 128,
+            intermediate_size = 18432,
+            vocab_size = 129280,
+            tie_word_embeddings = False,
+            n_routed_experts = 256,
+            moe_intermediate_size = 2048,
+            n_shared_experts = 1,
+            first_k_dense_replace = 3,
+            q_lora_rank = 1536,
+            kv_lora_rank = 512,
+            qk_nope_head_dim = 128,
+            qk_rope_head_dim = 64,
+            v_head_dim = 128,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_experts, 256)
+        self.assertEqual(arch.n_shared_experts, 1)
+        self.assertEqual(arch.num_dense_layers, 3)
+        self.assertEqual(arch.q_lora_rank, 1536)
+        self.assertEqual(arch.kv_lora_rank, 512)
+
+    def test_qwen3_moe_decoder_sparse_step(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 2048,
+            num_hidden_layers = 48,
+            num_attention_heads = 32,
+            num_key_value_heads = 4,
+            intermediate_size = 8192,
+            vocab_size = 151936,
+            tie_word_embeddings = True,
+            num_local_experts = 128,
+            moe_intermediate_size = 768,
+            decoder_sparse_step = 1,
+            mlp_only_layers = [],
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_experts, 128)
+        self.assertEqual(arch.num_dense_layers, 0)
+        self.assertIsNone(arch.q_lora_rank)
+
+    def test_qwen3_moe_with_mlp_only_layers(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 2048,
+            num_hidden_layers = 24,
+            num_attention_heads = 16,
+            num_key_value_heads = 4,
+            intermediate_size = 8192,
+            vocab_size = 151936,
+            tie_word_embeddings = True,
+            num_local_experts = 60,
+            moe_intermediate_size = 1408,
+            decoder_sparse_step = 1,
+            mlp_only_layers = [0, 1, 2, 3],
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_dense_layers, 4)
+
+    def test_glm4_moe_first_k_dense(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 4096,
+            num_hidden_layers = 46,
+            num_attention_heads = 96,
+            num_key_value_heads = 8,
+            intermediate_size = 10944,
+            vocab_size = 151552,
+            tie_word_embeddings = False,
+            n_routed_experts = 128,
+            moe_intermediate_size = 1408,
+            n_shared_experts = 1,
+            first_k_dense_replace = 1,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_dense_layers, 1)
+        self.assertEqual(arch.n_shared_experts, 1)
+
+    def test_gpt_oss_no_moe_intermediate(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 6144,
+            num_hidden_layers = 64,
+            num_attention_heads = 64,
+            num_key_value_heads = 8,
+            intermediate_size = 2880,
+            vocab_size = 200064,
+            tie_word_embeddings = False,
+            num_local_experts = 128,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.num_experts, 128)
+        self.assertIsNone(arch.moe_intermediate_size)
+        self.assertEqual(arch.num_dense_layers, 0)
+
+    def test_backward_compat_no_new_fields(self):
+        hf_config = SimpleNamespace(
+            hidden_size = 4096,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 128256,
+            tie_word_embeddings = False,
+        )
+        arch = extract_arch_config(hf_config)
+        self.assertEqual(arch.n_shared_experts, 0)
+        self.assertEqual(arch.num_dense_layers, 0)
+        self.assertIsNone(arch.q_lora_rank)
+
+
+class TestSharedExperts(unittest.TestCase):
+    def test_shared_experts_increase_weight_bytes(self):
+        no_shared = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 32000,
+            tie_word_embeddings = False,
+            num_experts = 64,
+            moe_intermediate_size = 1407,
+            n_shared_experts = 0,
+        )
+        with_shared = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 32,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 32000,
+            tie_word_embeddings = False,
+            num_experts = 64,
+            moe_intermediate_size = 1407,
+            n_shared_experts = 2,
+        )
+        w_no = compute_model_weights_bytes(no_shared, "full", False)
+        w_yes = compute_model_weights_bytes(with_shared, "full", False)
+        self.assertGreater(w_yes, w_no)
+        delta_per_layer = 4096 * 1407 * 3 * 2
+        expected_delta = delta_per_layer * 32 * 2
+        actual_delta = w_yes - w_no
+        self.assertAlmostEqual(
+            actual_delta, expected_delta, delta = expected_delta * 0.01
+        )
+
+    def test_deepseek_v3_params_in_range(self):
+        total = compute_total_params(DEEPSEEK_V3)
+        total_b = total / 1e9
+        self.assertGreater(total_b, 600)
+        self.assertLess(total_b, 750)
+
+
+class TestMLA(unittest.TestCase):
+    def test_mla_different_from_standard(self):
+        from utils.hardware.vram_estimation import _compute_attn_elements
+
+        mla_arch = DEEPSEEK_V3
+        std_arch = ModelArchConfig(
+            hidden_size = 7168,
+            num_hidden_layers = 61,
+            num_attention_heads = 128,
+            num_key_value_heads = 128,
+            intermediate_size = 18432,
+            vocab_size = 129280,
+        )
+        mla_attn = _compute_attn_elements(mla_arch)
+        std_attn = _compute_attn_elements(std_arch)
+        self.assertNotEqual(mla_attn, std_attn)
+
+    def test_mla_lora_produces_values(self):
+        lora_p = compute_lora_params(DEEPSEEK_V3, 16, ["q_proj", "v_proj", "o_proj"])
+        self.assertGreater(lora_p, 0)
+
+
+class TestDenseMoEMix(unittest.TestCase):
+    def test_dense_layers_change_total(self):
+        all_moe = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 46,
+            num_attention_heads = 96,
+            num_key_value_heads = 8,
+            intermediate_size = 10944,
+            vocab_size = 151552,
+            tie_word_embeddings = False,
+            num_experts = 128,
+            moe_intermediate_size = 1408,
+            n_shared_experts = 1,
+            num_dense_layers = 0,
+        )
+        mixed = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 46,
+            num_attention_heads = 96,
+            num_key_value_heads = 8,
+            intermediate_size = 10944,
+            vocab_size = 151552,
+            tie_word_embeddings = False,
+            num_experts = 128,
+            moe_intermediate_size = 1408,
+            n_shared_experts = 1,
+            num_dense_layers = 1,
+        )
+        w_all = compute_model_weights_bytes(all_moe, "full", False)
+        w_mixed = compute_model_weights_bytes(mixed, "full", False)
+        self.assertNotEqual(w_all, w_mixed)
+
+    def test_glm4_moe_params_reasonable(self):
+        total = compute_total_params(GLM4_MOE)
+        total_b = total / 1e9
+        self.assertGreater(total_b, 80)
+        self.assertLess(total_b, 120)
+
+    def test_qwen3_moe_30b_params_reasonable(self):
+        total = compute_total_params(QWEN3_MOE_30B)
+        total_b = total / 1e9
+        self.assertGreater(total_b, 20)
+        self.assertLess(total_b, 50)
+
+    def test_gpt_oss_uses_intermediate_size(self):
+        total = compute_total_params(GPT_OSS)
+        total_b = total / 1e9
+        self.assertGreater(total_b, 350)
+        self.assertLess(total_b, 500)
+
+    def test_lora_dense_vs_moe_layers_differ(self):
+        all_moe = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 10,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 32000,
+            tie_word_embeddings = False,
+            num_experts = 8,
+            moe_intermediate_size = 1024,
+            num_dense_layers = 0,
+        )
+        mixed = ModelArchConfig(
+            hidden_size = 4096,
+            num_hidden_layers = 10,
+            num_attention_heads = 32,
+            num_key_value_heads = 8,
+            intermediate_size = 14336,
+            vocab_size = 32000,
+            tie_word_embeddings = False,
+            num_experts = 8,
+            moe_intermediate_size = 1024,
+            num_dense_layers = 5,
+        )
+        lora_all = compute_lora_params(
+            all_moe, 16, ["gate_proj", "up_proj", "down_proj"]
+        )
+        lora_mix = compute_lora_params(mixed, 16, ["gate_proj", "up_proj", "down_proj"])
+        self.assertNotEqual(lora_all, lora_mix)
+
+
+if __name__ == "__main__":
+    unittest.main()

studio/backend/utils/datasets/format_conversion.py

@@ -553,13 +553,9 @@ def convert_to_vlm_format(
                         batch_results[idx] = future.result()
                     except Exception as e:
                         failed_count += 1
-                        if failed_count == 1:
-                            print(
-                                f"⚠️ First VLM conversion failure: {type(e).__name__}: {e}"
-                            )
                         if failed_count == 1:
                             logger.info(
-                                f"⚠️ First VLM conversion failure: {type(e).__name__}: {e}"
+                                f"First VLM conversion failure: {type(e).__name__}: {e}"
                             )
 
             converted_list.extend(r for r in batch_results if r is not None)
@@ -583,13 +579,10 @@ def convert_to_vlm_format(
                 converted_list.append(_convert_single_sample(sample))
             except Exception as e:
                 failed_count += 1
-                if failed_count == 1:
-                    # Log the first failure to aid debugging
-                    print(f"⚠️ First VLM conversion failure: {type(e).__name__}: {e}")
                 if failed_count == 1:
                     # Log the first failure to aid debugging
                     logger.info(
-                        f"⚠️ First VLM conversion failure: {type(e).__name__}: {e}"
+                        f"First VLM conversion failure: {type(e).__name__}: {e}"
                     )
             pbar.set_postfix(ok = len(converted_list), failed = failed_count, refresh = False)
         pbar.close()

studio/backend/utils/hardware/VRAM_ESTIMATION.md

@@ -0,0 +1,161 @@
+# VRAM Estimation for Training
+
+```
+Total VRAM = Weights + LoRA Adapters + Optimizer + Gradients + Activations + CUDA Overhead
+```
+
+| Symbol | Meaning |
+|--------|---------|
+| `H` | `hidden_size` |
+| `L` | `num_hidden_layers` |
+| `V` | `vocab_size` |
+| `K` | `(H / num_attention_heads) * num_key_value_heads` |
+| `M` | `intermediate_size` (or `moe_intermediate_size`) |
+| `E` | `num_experts` (1 for dense) |
+| `r` | LoRA rank |
+| `B` | `per_device_train_batch_size` |
+| `S` | `max_seq_length` |
+
+---
+
+## 1. Model Weights
+
+```
+QKVO = (H + K + K + H) * H
+MLP  = H * M * 3 * E  +  (E * H if E > 1 else 0)
+
+Quantizable     = (QKVO + MLP) * L
+Non-quantizable = 2*H*L + V*H + (V*H if not tie_embeddings else 0)
+```
+
+| Mode | Bytes |
+|------|-------|
+| QLoRA 4-bit | `Quantizable * 2 / 3.2 + Non-quantizable * 2` |
+| LoRA / Full fp16 | `(Quantizable + Non-quantizable) * 2` |
+
+The 3.2 factor (`16/5`) accounts for BNB NF4 blockwise scales.
+
+## 2. LoRA Adapters
+
+| Module | A | B |
+|--------|---|---|
+| q_proj | `H×r` | `r×H` |
+| k_proj | `H×r` | `r×K` |
+| v_proj | `H×r` | `r×K` |
+| o_proj | `H×r` | `r×H` |
+| gate_proj | `H×r` | `r×M` |
+| up_proj | `H×r` | `r×M` |
+| down_proj | `M×r` | `r×H` |
+
+MLP modules multiply by `E` for MoE.
+
+```
+LoRA_bytes = sum(A + B per selected module) * L * 2
+```
+
+## 3. Optimizer States (calibrated)
+
+| Optimizer | Bytes/param | Notes |
+|-----------|------------|-------|
+| `adamw_8bit` | 4 | BNB upcasts to fp32 during step |
+| `adamw_torch` | 6 | Fused, no master copy |
+| `paged_adamw_32bit` | 8 | Full fp32 states |
+| `sgd` | 4 | |
+
+Trainable params = all params (Full FT) or LoRA params only.
+
+## 4. Gradients
+
+```
+Gradient_bytes = trainable_params * 2    (fp16, accumulated in-place)
+```
+
+## 5. Activations
+
+Per-layer (from `unsloth_zoo/vllm_utils.py`):
+```
+Per_layer = (S*B*(H+K+K) + S*B*2 + S*B*(M+M)) * 2 * 1.25
+```
+
+| GC Mode | Full FT | LoRA/QLoRA |
+|---------|---------|------------|
+| none | `L` layers | `L` layers |
+| true (HF) | 2.0 | 1.0 |
+| unsloth | 1.5 | 1.0 |
+
+## 6. Floors
+
+Gradients and activations have minimum floors at **15% of model weight memory** to account for autograd overhead, attention score matrices, NCCL buffers, mixed-precision scaling, and PyTorch fragmentation.
+
+```
+gradient_bytes  = max(computed, weights * 0.15)
+activation_bytes = max(computed, weights * 0.15 * B/2)
+```
+
+## 7. CUDA Overhead
+
+**1.4 GB** fixed — CUDA driver + PyTorch runtime, calibrated on RTX 5070 Ti.
+
+## 8. Multi-GPU Overhead
+
+When sharding across multiple GPUs, each additional GPU (beyond the first) contributes only **85%** of its free VRAM to the usable pool. The 15% discount accounts for NCCL all-reduce buffers, PCIe/NVLink transfer overhead, synchronization barriers, and memory fragmentation from non-uniform shard sizes. Calibrated empirically on 2-8 GPU setups with NVLink and PCIe topologies.
+
+```
+usable_gb = free[gpu_0] + sum(free[gpu_i] * 0.85 for i in 1..N)
+```
+
+---
+
+## Reference Table (bsz=2, seq=2048, rank=16, GC=unsloth, adamw_8bit)
+
+| Model | Weights | LoRA | Optim | Grad | Act | CUDA | Total |
+|-------|---------|------|-------|------|-----|------|-------|
+| 0.5B QLoRA | 0.5 | 0.0 | 0.0 | 0.1 | 0.1 | 1.4 | **2.1** |
+| 1B QLoRA | 1.1 | 0.0 | 0.0 | 0.2 | 0.2 | 1.4 | **2.9** |
+| 3B QLoRA | 2.4 | 0.0 | 0.1 | 0.5 | 0.5 | 1.4 | **4.9** |
+| 8B QLoRA | 6.0 | 0.1 | 0.2 | 1.2 | 1.2 | 1.4 | **10.1** |
+| 8B LoRA fp16 | 15.0 | 0.1 | 0.2 | 3.0 | 3.0 | 1.4 | **22.6** |
+| 8B Full FT | 15.0 | — | 29.9 | 15.0 | 3.0 | 1.4 | **64.2** |
+| 32B LoRA fp16 | 61.0 | 0.2 | 0.5 | 12.2 | 12.2 | 1.4 | **87.6** |
+| 72B QLoRA | 45.5 | 0.4 | 0.8 | 9.1 | 9.1 | 1.4 | **66.3** |
+
+## E2E Validation (Llama-3.2-1B, B200 emulating 24GB)
+
+| Config | Estimated | Actual (nvsmi) | Error |
+|--------|----------|----------------|-------|
+| QLoRA bsz=2 seq=512 | 2.55 GB | 2.65 GB | -3.7% |
+| QLoRA bsz=2 seq=2048 | 2.60 GB | 2.65 GB | -1.8% |
+| QLoRA bsz=4 seq=2048 | 2.65 GB | 2.65 GB | +0.0% |
+| LoRA fp16 bsz=2 | 3.84 GB | 3.88 GB | -1.0% |
+| Full FT adamw_8bit | 10.89 GB | 10.80 GB | +0.8% |
+| Full FT adamw_torch | 13.19 GB | 12.93 GB | +2.0% |
+
+*Note: e2e numbers predate the 15% floors, which add safety margin on top.*
+
+---
+
+## Parameter Flow
+
+```
+Frontend -> routes/{training,inference}.py
+         -> prepare_gpu_selection(gpu_ids, model_name, ...)
+            |
+            +-- gpu_ids is explicit (e.g. [5,6,7])
+            |     -> resolve_requested_gpu_ids: validate against parent-visible set
+            |     -> return all requested GPUs (model sharded across all of them)
+            |
+            +-- gpu_ids is None or []
+                  -> auto_select_gpu_ids: estimate VRAM, pick minimum GPUs needed
+                  -> estimate_required_model_memory_gb -> estimate_training_vram
+                  -> greedy selection: rank GPUs by free VRAM, add until model fits
+
+         -> get_device_map(resolved_gpu_ids)
+            -> "balanced" if >1 GPU, "sequential" otherwise
+
+         -> worker subprocess: apply_gpu_ids(resolved_gpu_ids)
+            -> sets CUDA_VISIBLE_DEVICES before torch/CUDA init
+```
+
+Threaded params: `batch_size`, `max_seq_length`, `lora_r`, `target_modules`, `gradient_checkpointing`, `optim`.
+
+Source: `studio/backend/utils/hardware/vram_estimation.py`

studio/backend/utils/hardware/__init__.py

@@ -18,11 +18,31 @@ from .hardware import (
     get_gpu_summary,
     get_package_versions,
     get_gpu_utilization,
+    get_visible_gpu_utilization,
+    get_backend_visible_gpu_info,
     get_physical_gpu_count,
     get_visible_gpu_count,
+    get_parent_visible_gpu_ids,
+    resolve_requested_gpu_ids,
+    estimate_fp16_model_size_bytes,
+    estimate_required_model_memory_gb,
+    auto_select_gpu_ids,
+    prepare_gpu_selection,
     safe_num_proc,
     safe_thread_num_proc,
     dataset_map_num_proc,
+    get_device_map,
+    get_offloaded_device_map_entries,
+    raise_if_offloaded,
+    apply_gpu_ids,
+)
+
+from .vram_estimation import (
+    ModelArchConfig,
+    TrainingVramConfig,
+    VramBreakdown,
+    extract_arch_config,
+    estimate_training_vram,
 )
 
 __all__ = [
@@ -38,9 +58,26 @@ __all__ = [
     "get_gpu_summary",
     "get_package_versions",
     "get_gpu_utilization",
+    "get_visible_gpu_utilization",
+    "get_backend_visible_gpu_info",
     "get_physical_gpu_count",
     "get_visible_gpu_count",
+    "get_parent_visible_gpu_ids",
+    "resolve_requested_gpu_ids",
+    "estimate_fp16_model_size_bytes",
+    "estimate_required_model_memory_gb",
+    "auto_select_gpu_ids",
+    "prepare_gpu_selection",
     "safe_num_proc",
     "safe_thread_num_proc",
     "dataset_map_num_proc",
+    "get_device_map",
+    "get_offloaded_device_map_entries",
+    "raise_if_offloaded",
+    "apply_gpu_ids",
+    "ModelArchConfig",
+    "TrainingVramConfig",
+    "VramBreakdown",
+    "extract_arch_config",
+    "estimate_training_vram",
 ]

studio/backend/utils/hardware/nvidia.py

@@ -0,0 +1,279 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import subprocess
+from typing import Any, Optional
+
+from loggers import get_logger
+
+logger = get_logger(__name__)
+
+
+def _parse_smi_value(raw: str):
+    raw = raw.strip()
+    if not raw or raw == "[N/A]":
+        return None
+    try:
+        return float(raw)
+    except (ValueError, TypeError):
+        return None
+
+
+def _build_gpu_metrics(
+    vram_used_mb,
+    vram_total_mb,
+    power_draw,
+    power_limit,
+    **extra,
+) -> dict[str, Any]:
+    return {
+        **extra,
+        "vram_used_gb": round(vram_used_mb / 1024, 2)
+        if vram_used_mb is not None
+        else None,
+        "vram_total_gb": round(vram_total_mb / 1024, 2)
+        if vram_total_mb is not None
+        else None,
+        "vram_utilization_pct": round((vram_used_mb / vram_total_mb) * 100, 1)
+        if vram_used_mb is not None and vram_total_mb and vram_total_mb > 0
+        else None,
+        "power_draw_w": power_draw,
+        "power_limit_w": power_limit,
+        "power_utilization_pct": round((power_draw / power_limit) * 100, 1)
+        if power_draw is not None and power_limit and power_limit > 0
+        else None,
+    }
+
+
+def _visible_ordinal_map(
+    parent_visible_ids: Optional[list[int]],
+) -> Optional[dict[int, int]]:
+    if parent_visible_ids is None:
+        return None
+    return {gpu_id: ordinal for ordinal, gpu_id in enumerate(parent_visible_ids)}
+
+
+def get_physical_gpu_count() -> Optional[int]:
+    """Return physical GPU count via nvidia-smi, or None on failure."""
+    try:
+        result = subprocess.run(
+            ["nvidia-smi", "-L"],
+            capture_output = True,
+            text = True,
+            timeout = 5,
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            return len(result.stdout.strip().splitlines())
+        logger.warning(
+            "nvidia-smi -L returned code %d; caller should fall back to torch",
+            result.returncode,
+        )
+    except Exception as e:
+        logger.warning("nvidia-smi -L failed: %s; caller should fall back to torch", e)
+    return None
+
+
+def get_primary_gpu_utilization() -> dict[str, Any]:
+    try:
+        result = subprocess.run(
+            [
+                "nvidia-smi",
+                "--query-gpu=utilization.gpu,temperature.gpu,"
+                "memory.used,memory.total,power.draw,power.limit",
+                "--format=csv,noheader,nounits",
+            ],
+            capture_output = True,
+            text = True,
+            timeout = 5,
+        )
+    except (OSError, subprocess.TimeoutExpired) as e:
+        logger.warning("nvidia-smi query failed in get_primary_gpu_utilization: %s", e)
+        return {"available": False}
+    if result.returncode != 0 or not result.stdout.strip():
+        return {"available": False}
+
+    first_line = result.stdout.strip().splitlines()[0]
+    parts = [p.strip() for p in first_line.split(",")]
+    if len(parts) < 6:
+        return {"available": False}
+
+    return _build_gpu_metrics(
+        vram_used_mb = _parse_smi_value(parts[2]),
+        vram_total_mb = _parse_smi_value(parts[3]),
+        power_draw = _parse_smi_value(parts[4]),
+        power_limit = _parse_smi_value(parts[5]),
+        available = True,
+        gpu_utilization_pct = _parse_smi_value(parts[0]),
+        temperature_c = _parse_smi_value(parts[1]),
+    )
+
+
+def get_visible_gpu_utilization(
+    parent_visible_ids: Optional[list[int]],
+    parent_cuda_visible_devices: Optional[str] = None,
+) -> dict[str, Any]:
+    # When parent_visible_ids is None (UUID/MIG mask), we cannot safely
+    # map nvidia-smi rows to the process's visible devices. Return empty
+    # instead of exposing all physical GPUs.
+    if parent_visible_ids is None:
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": parent_cuda_visible_devices,
+            "parent_visible_gpu_ids": [],
+            "devices": [],
+            "index_kind": "unresolved",
+        }
+    visible_ordinals = _visible_ordinal_map(parent_visible_ids)
+    try:
+        result = subprocess.run(
+            [
+                "nvidia-smi",
+                "--query-gpu=index,utilization.gpu,temperature.gpu,"
+                "memory.used,memory.total,power.draw,power.limit",
+                "--format=csv,noheader,nounits",
+            ],
+            capture_output = True,
+            text = True,
+            timeout = 5,
+        )
+    except (OSError, subprocess.TimeoutExpired) as e:
+        logger.warning("nvidia-smi query failed in get_visible_gpu_utilization: %s", e)
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": parent_cuda_visible_devices,
+            "parent_visible_gpu_ids": parent_visible_ids or [],
+            "devices": [],
+            "index_kind": "physical",
+        }
+    if result.returncode != 0 or not result.stdout.strip():
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": parent_cuda_visible_devices,
+            "parent_visible_gpu_ids": parent_visible_ids or [],
+            "devices": [],
+            "index_kind": "physical",
+        }
+
+    devices = []
+    for line in result.stdout.strip().splitlines():
+        parts = [p.strip() for p in line.split(",")]
+        if len(parts) < 7:
+            continue
+
+        try:
+            idx = int(parts[0])
+        except (ValueError, TypeError):
+            continue
+
+        if visible_ordinals is not None and idx not in visible_ordinals:
+            continue
+
+        devices.append(
+            _build_gpu_metrics(
+                vram_used_mb = _parse_smi_value(parts[3]),
+                vram_total_mb = _parse_smi_value(parts[4]),
+                power_draw = _parse_smi_value(parts[5]),
+                power_limit = _parse_smi_value(parts[6]),
+                index = idx,
+                index_kind = "physical",
+                visible_ordinal = (
+                    visible_ordinals[idx]
+                    if visible_ordinals is not None
+                    else len(devices)
+                ),
+                gpu_utilization_pct = _parse_smi_value(parts[1]),
+                temperature_c = _parse_smi_value(parts[2]),
+            )
+        )
+
+    return {
+        "available": len(devices) > 0,
+        "backend_cuda_visible_devices": parent_cuda_visible_devices,
+        "parent_visible_gpu_ids": parent_visible_ids or [],
+        "devices": devices,
+        "index_kind": "physical",
+    }
+
+
+def get_backend_visible_gpu_info(
+    parent_visible_ids: Optional[list[int]],
+    backend_cuda_visible_devices: Optional[str],
+) -> dict[str, Any]:
+    # When parent_visible_ids is None (UUID/MIG mask), we cannot safely
+    # map nvidia-smi rows to the process's visible devices.
+    if parent_visible_ids is None:
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": backend_cuda_visible_devices,
+            "parent_visible_gpu_ids": [],
+            "devices": [],
+            "index_kind": "unresolved",
+        }
+    visible_ordinals = _visible_ordinal_map(parent_visible_ids)
+    try:
+        result = subprocess.run(
+            [
+                "nvidia-smi",
+                "--query-gpu=index,name,memory.total",
+                "--format=csv,noheader,nounits",
+            ],
+            capture_output = True,
+            text = True,
+            timeout = 10,
+        )
+    except (OSError, subprocess.TimeoutExpired) as e:
+        logger.warning("nvidia-smi query failed in get_backend_visible_gpu_info: %s", e)
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": backend_cuda_visible_devices,
+            "parent_visible_gpu_ids": parent_visible_ids or [],
+            "devices": [],
+            "index_kind": "physical",
+        }
+    if result.returncode != 0:
+        return {
+            "available": False,
+            "backend_cuda_visible_devices": backend_cuda_visible_devices,
+            "parent_visible_gpu_ids": parent_visible_ids or [],
+            "devices": [],
+            "index_kind": "physical",
+        }
+
+    devices = []
+    for line in result.stdout.strip().splitlines():
+        parts = [p.strip() for p in line.split(",")]
+        if len(parts) < 3:
+            continue
+        try:
+            idx = int(parts[0])
+        except (ValueError, TypeError):
+            continue
+        if visible_ordinals is not None and idx not in visible_ordinals:
+            continue
+        # Use split with limit to handle GPU names containing commas
+        name = parts[1] if len(parts) == 3 else ", ".join(parts[1:-1])
+        try:
+            mem_total_mb = int(parts[-1])
+        except (ValueError, TypeError):
+            continue
+        devices.append(
+            {
+                "index": idx,
+                "index_kind": "physical",
+                "visible_ordinal": (
+                    visible_ordinals[idx]
+                    if visible_ordinals is not None
+                    else len(devices)
+                ),
+                "name": name,
+                "memory_total_gb": round(mem_total_mb / 1024, 2),
+            }
+        )
+
+    return {
+        "available": len(devices) > 0,
+        "backend_cuda_visible_devices": backend_cuda_visible_devices,
+        "parent_visible_gpu_ids": parent_visible_ids or [],
+        "devices": devices,
+        "index_kind": "physical",
+    }

studio/backend/utils/hardware/vram_estimation.py

@@ -0,0 +1,501 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved.
+
+"""
+Training VRAM estimation.
+
+Total VRAM = weights + LoRA adapters + optimizer states + gradients
+           + activations + CUDA overhead.
+Activation formula from unsloth_zoo/vllm_utils.py.
+All constants empirically calibrated against Llama-3.2-1B on B200.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Dict, Optional
+
+QUANT_4BIT_FACTOR = 16 / 5
+CUDA_OVERHEAD_BYTES = int(1.4 * 1024**3)  # calibrated on RTX 5070 Ti
+
+DEFAULT_TARGET_MODULES = [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj",
+]
+
+# Empirically calibrated bytes/param — see VRAM_ESTIMATION.md for rationale.
+OPTIMIZER_BYTES_PER_PARAM: Dict[str, int] = {
+    "adamw_8bit": 4,  # BNB upcasts to fp32 during step
+    "paged_adamw_8bit": 4,
+    "adamw_bnb_8bit": 4,
+    "paged_adamw_32bit": 8,
+    "adamw_torch": 6,  # fused, no master copy
+    "adamw_torch_fused": 6,
+    "sgd": 4,
+}
+
+# (full_ft_multiplier, lora_multiplier) — fraction of num_layers.
+# LoRA: frozen base layers skip activation storage, but you always need
+# at least ~1 layer in flight during backprop recomputation.
+GC_LAYER_MULTIPLIERS = {
+    "none": (None, None),
+    "true": (2.0, 1.0),
+    "unsloth": (1.5, 1.0),
+}
+
+
+@dataclass
+class ModelArchConfig:
+    hidden_size: int
+    num_hidden_layers: int
+    num_attention_heads: int
+    num_key_value_heads: int
+    intermediate_size: int
+    vocab_size: int
+    tie_word_embeddings: bool = True
+    num_experts: Optional[int] = None
+    moe_intermediate_size: Optional[int] = None
+    n_shared_experts: int = 0
+    num_dense_layers: int = 0
+    q_lora_rank: Optional[int] = None
+    kv_lora_rank: Optional[int] = None
+    qk_nope_head_dim: Optional[int] = None
+    qk_rope_head_dim: Optional[int] = None
+    v_head_dim: Optional[int] = None
+
+
+@dataclass
+class TrainingVramConfig:
+    training_method: str = "qlora"
+    batch_size: int = 4
+    max_seq_length: int = 2048
+    lora_rank: int = 16
+    target_modules: list = field(default_factory = lambda: list(DEFAULT_TARGET_MODULES))
+    gradient_checkpointing: str = "unsloth"
+    optimizer: str = "adamw_8bit"
+    load_in_4bit: bool = True
+
+
+@dataclass
+class VramBreakdown:
+    model_weights: int
+    lora_adapters: int
+    optimizer_states: int
+    gradients: int
+    activations: int
+    cuda_overhead: int
+    # The computed (formula-based) activation cost before floors.
+    # This is the true per-layer cost that doesn't shard across GPUs.
+    activations_computed: int = 0
+
+    @property
+    def total(self) -> int:
+        return (
+            self.model_weights
+            + self.lora_adapters
+            + self.optimizer_states
+            + self.gradients
+            + self.activations
+            + self.cuda_overhead
+        )
+
+    def min_gpu_vram(self, n_gpus: int) -> int:
+        """Minimum VRAM a single GPU needs: its shard + non-shardable costs.
+
+        Weights/LoRA/optimizer/gradients shard across GPUs.
+        The computed activation cost does NOT shard (one GPU runs the layer).
+        The floor portion (activations - computed) is overhead that shards.
+        """
+        shardable = (
+            self.model_weights
+            + self.lora_adapters
+            + self.optimizer_states
+            + self.gradients
+            + (self.activations - self.activations_computed)  # floor overhead shards
+        )
+        per_gpu_fixed = self.activations_computed + self.cuda_overhead
+        return shardable // max(n_gpus, 1) + per_gpu_fixed
+
+    def to_gb_dict(self) -> Dict[str, float]:
+        return {
+            "model_weights_gb": round(self.model_weights / (1024**3), 3),
+            "lora_adapters_gb": round(self.lora_adapters / (1024**3), 3),
+            "optimizer_states_gb": round(self.optimizer_states / (1024**3), 3),
+            "gradients_gb": round(self.gradients / (1024**3), 3),
+            "activations_gb": round(self.activations / (1024**3), 3),
+            "cuda_overhead_gb": round(self.cuda_overhead / (1024**3), 3),
+            "total_gb": round(self.total / (1024**3), 3),
+        }
+
+
+def _compute_num_dense_layers(text_config, total_layers: int) -> int:
+    """Count how many layers use dense MLP instead of MoE."""
+    first_k = getattr(text_config, "first_k_dense_replace", None)
+    if first_k is not None:
+        return min(int(first_k), total_layers)
+
+    sparse_step = getattr(text_config, "decoder_sparse_step", None)
+    mlp_only = getattr(text_config, "mlp_only_layers", None) or []
+    if sparse_step is not None and sparse_step > 0:
+        mlp_only_set = set(mlp_only)
+        moe_count = sum(
+            1
+            for i in range(total_layers)
+            if i not in mlp_only_set and (i + 1) % sparse_step == 0
+        )
+        return total_layers - moe_count
+
+    return 0
+
+
+def extract_arch_config(hf_config) -> Optional[ModelArchConfig]:
+    text_config = getattr(hf_config, "text_config", None) or hf_config
+
+    hidden_size = getattr(text_config, "hidden_size", None)
+    num_layers = getattr(text_config, "num_hidden_layers", None)
+    num_heads = getattr(text_config, "num_attention_heads", None)
+    intermediate_size = getattr(text_config, "intermediate_size", None)
+    vocab_size = getattr(text_config, "vocab_size", None)
+
+    if isinstance(intermediate_size, (list, tuple)):
+        intermediate_size = intermediate_size[0] if intermediate_size else None
+    if intermediate_size is None and hidden_size is not None:
+        intermediate_size = hidden_size * 4
+
+    if not all(
+        v is not None
+        for v in (hidden_size, num_layers, num_heads, intermediate_size, vocab_size)
+    ):
+        return None
+    if num_heads <= 0:
+        return None
+
+    num_kv_heads = getattr(text_config, "num_key_value_heads", num_heads)
+
+    num_experts = None
+    for attr in ("num_local_experts", "num_experts", "n_routed_experts"):
+        num_experts = getattr(text_config, attr, None)
+        if num_experts is not None:
+            break
+
+    moe_intermediate = getattr(text_config, "moe_intermediate_size", None)
+    n_shared_experts = getattr(text_config, "n_shared_experts", None) or 0
+
+    num_dense_layers = 0
+    if num_experts is not None and num_experts > 1:
+        num_dense_layers = _compute_num_dense_layers(text_config, num_layers)
+
+    q_lora_rank = getattr(text_config, "q_lora_rank", None)
+    kv_lora_rank = getattr(text_config, "kv_lora_rank", None)
+    qk_nope_head_dim = getattr(text_config, "qk_nope_head_dim", None)
+    qk_rope_head_dim = getattr(text_config, "qk_rope_head_dim", None)
+    v_head_dim = getattr(text_config, "v_head_dim", None)
+
+    return ModelArchConfig(
+        hidden_size = hidden_size,
+        num_hidden_layers = num_layers,
+        num_attention_heads = num_heads,
+        num_key_value_heads = num_kv_heads,
+        intermediate_size = intermediate_size,
+        vocab_size = vocab_size,
+        tie_word_embeddings = getattr(text_config, "tie_word_embeddings", True),
+        num_experts = num_experts,
+        moe_intermediate_size = moe_intermediate,
+        n_shared_experts = n_shared_experts,
+        num_dense_layers = num_dense_layers,
+        q_lora_rank = q_lora_rank,
+        kv_lora_rank = kv_lora_rank,
+        qk_nope_head_dim = qk_nope_head_dim,
+        qk_rope_head_dim = qk_rope_head_dim,
+        v_head_dim = v_head_dim,
+    )
+
+
+def _get_kv_size(arch: ModelArchConfig) -> int:
+    return (arch.hidden_size // arch.num_attention_heads) * arch.num_key_value_heads
+
+
+def _get_mlp_size(arch: ModelArchConfig) -> int:
+    if arch.moe_intermediate_size is not None:
+        return arch.moe_intermediate_size
+    return arch.intermediate_size
+
+
+def _get_num_experts(arch: ModelArchConfig) -> int:
+    return arch.num_experts if arch.num_experts and arch.num_experts > 1 else 1
+
+
+def _compute_attn_elements(arch: ModelArchConfig) -> int:
+    """Attention weight elements per layer."""
+    hd = arch.hidden_size
+    if arch.q_lora_rank is not None:
+        nh = arch.num_attention_heads
+        qk_head = arch.qk_nope_head_dim + arch.qk_rope_head_dim
+        q_a = hd * arch.q_lora_rank
+        q_b = arch.q_lora_rank * (nh * qk_head)
+        kv_a = hd * (arch.kv_lora_rank + arch.qk_rope_head_dim)
+        kv_b = arch.kv_lora_rank * (nh * (arch.qk_nope_head_dim + arch.v_head_dim))
+        o = (nh * arch.v_head_dim) * hd
+        norms = arch.q_lora_rank + arch.kv_lora_rank
+        return q_a + q_b + kv_a + kv_b + o + norms
+    kv_size = _get_kv_size(arch)
+    return (hd + kv_size + kv_size + hd) * hd
+
+
+def _compute_dense_mlp_elements(arch: ModelArchConfig) -> int:
+    return arch.hidden_size * arch.intermediate_size * 3
+
+
+def _compute_moe_mlp_elements(arch: ModelArchConfig) -> int:
+    hd = arch.hidden_size
+    mlp_size = _get_mlp_size(arch)
+    n_experts = _get_num_experts(arch)
+    return hd * mlp_size * 3 * (n_experts + arch.n_shared_experts) + n_experts * hd
+
+
+def _compute_layer_elements(arch: ModelArchConfig):
+    """Return (total_quantizable, layernorms_per_layer, embed, lm_head) element counts.
+
+    total_quantizable is summed across ALL layers (not per-layer).
+    """
+    hd = arch.hidden_size
+    n_layers = arch.num_hidden_layers
+    n_experts = _get_num_experts(arch)
+
+    attn_total = _compute_attn_elements(arch) * n_layers
+
+    if n_experts > 1:
+        n_dense = arch.num_dense_layers
+        n_moe = n_layers - n_dense
+        mlp_total = (
+            _compute_moe_mlp_elements(arch) * n_moe
+            + _compute_dense_mlp_elements(arch) * n_dense
+        )
+    else:
+        mlp_total = _compute_dense_mlp_elements(arch) * n_layers
+
+    layernorms = 2 * hd
+    embed_tokens = arch.vocab_size * hd
+    lm_head = 0 if arch.tie_word_embeddings else arch.vocab_size * hd
+    return attn_total + mlp_total, layernorms, embed_tokens, lm_head
+
+
+def compute_model_weights_bytes(
+    arch: ModelArchConfig,
+    training_method: str,
+    load_in_4bit: bool,
+) -> int:
+    total_quantizable, layernorms, embed_tokens, lm_head = _compute_layer_elements(arch)
+    n_layers = arch.num_hidden_layers
+    non_quantizable = layernorms * n_layers + embed_tokens + lm_head
+
+    if training_method == "qlora" and load_in_4bit:
+        return int(total_quantizable * 2 / QUANT_4BIT_FACTOR + non_quantizable * 2)
+
+    return int((total_quantizable + non_quantizable) * 2)
+
+
+def compute_total_params(arch: ModelArchConfig) -> int:
+    total_quantizable, layernorms, embed_tokens, lm_head = _compute_layer_elements(arch)
+    n_layers = arch.num_hidden_layers
+    return total_quantizable + layernorms * n_layers + embed_tokens + lm_head
+
+
+def _lora_attn_elements(
+    arch: ModelArchConfig,
+    r: int,
+    target_modules: list,
+) -> int:
+    hd = arch.hidden_size
+    if arch.q_lora_rank is not None:
+        # MLA: q_proj->q_b, k_proj->kv_a, v_proj->kv_b, o_proj->o
+        nh = arch.num_attention_heads
+        qk_head = arch.qk_nope_head_dim + arch.qk_rope_head_dim
+        kv_out = nh * (arch.qk_nope_head_dim + arch.v_head_dim)
+        o_in = nh * arch.v_head_dim
+        dims = {
+            "q_proj": (arch.q_lora_rank, nh * qk_head),
+            "k_proj": (hd, arch.kv_lora_rank + arch.qk_rope_head_dim),
+            "v_proj": (arch.kv_lora_rank, kv_out),
+            "o_proj": (o_in, hd),
+        }
+    else:
+        kv_size = _get_kv_size(arch)
+        dims = {
+            "q_proj": (hd, hd),
+            "k_proj": (hd, kv_size),
+            "v_proj": (hd, kv_size),
+            "o_proj": (hd, hd),
+        }
+    total = 0
+    for name, (in_dim, out_dim) in dims.items():
+        if name in target_modules:
+            total += in_dim * r + r * out_dim
+    return total
+
+
+def _lora_mlp_elements(
+    hd: int,
+    mlp_size: int,
+    r: int,
+    target_modules: list,
+    expert_mult: int,
+) -> int:
+    module_ab = {
+        "gate_proj": (hd * r, r * mlp_size),
+        "up_proj": (hd * r, r * mlp_size),
+        "down_proj": (mlp_size * r, r * hd),
+    }
+    total = 0
+    for name, (a, b) in module_ab.items():
+        if name in target_modules:
+            total += (a + b) * expert_mult
+    return total
+
+
+def compute_lora_params(
+    arch: ModelArchConfig,
+    lora_rank: int,
+    target_modules: list,
+) -> int:
+    hd = arch.hidden_size
+    r = lora_rank
+    n_layers = arch.num_hidden_layers
+    n_experts = _get_num_experts(arch)
+
+    attn_total = _lora_attn_elements(arch, r, target_modules) * n_layers
+
+    if n_experts > 1:
+        n_dense = arch.num_dense_layers
+        n_moe = n_layers - n_dense
+        # Include shared experts alongside routed experts
+        moe_expert_mult = n_experts + arch.n_shared_experts
+        moe_mlp = _lora_mlp_elements(
+            hd,
+            _get_mlp_size(arch),
+            r,
+            target_modules,
+            moe_expert_mult,
+        )
+        dense_mlp = _lora_mlp_elements(
+            hd,
+            arch.intermediate_size,
+            r,
+            target_modules,
+            1,
+        )
+        mlp_total = moe_mlp * n_moe + dense_mlp * n_dense
+    else:
+        mlp_total = (
+            _lora_mlp_elements(
+                hd,
+                arch.intermediate_size,
+                r,
+                target_modules,
+                1,
+            )
+            * n_layers
+        )
+
+    return attn_total + mlp_total
+
+
+def compute_lora_adapter_bytes(lora_params: int) -> int:
+    return lora_params * 2
+
+
+def compute_optimizer_bytes(trainable_params: int, optimizer: str) -> int:
+    optimizer_key = optimizer.lower().replace("-", "_")
+    bytes_per_param = OPTIMIZER_BYTES_PER_PARAM.get(optimizer_key, 4)
+    return trainable_params * bytes_per_param
+
+
+def compute_gradient_bytes(trainable_params: int) -> int:
+    return trainable_params * 2
+
+
+def compute_activation_bytes(
+    arch: ModelArchConfig,
+    batch_size: int,
+    seq_len: int,
+    gradient_checkpointing: str,
+    is_lora: bool = False,
+) -> int:
+    hd = arch.hidden_size
+    kv_size = _get_kv_size(arch)
+    mlp_size = _get_mlp_size(arch)
+    bsz = batch_size
+    n_layers = arch.num_hidden_layers
+
+    activation_qkv = seq_len * bsz * (hd + kv_size + kv_size)
+    residual_memory = (seq_len * bsz) * 2
+    activation_mlp = seq_len * bsz * (mlp_size + mlp_size)
+
+    per_layer_bytes = (activation_qkv + residual_memory + activation_mlp) * 2
+    per_layer_bytes = int(per_layer_bytes * 1.25)
+
+    gc_key = gradient_checkpointing.lower()
+    gc_entry = GC_LAYER_MULTIPLIERS.get(gc_key, (None, None))
+    full_ft_mult, lora_mult = gc_entry
+    gc_multiplier = lora_mult if is_lora else full_ft_mult
+
+    if gc_multiplier is None:
+        effective_layers = n_layers
+    else:
+        effective_layers = gc_multiplier
+
+    return int(per_layer_bytes * effective_layers)
+
+
+def estimate_training_vram(
+    arch: ModelArchConfig,
+    config: TrainingVramConfig,
+) -> VramBreakdown:
+    method = config.training_method.lower()
+    is_lora = method in ("qlora", "lora")
+    load_in_4bit = config.load_in_4bit or method == "qlora"
+
+    model_weights = compute_model_weights_bytes(arch, method, load_in_4bit)
+
+    lora_params = 0
+    lora_adapter_bytes = 0
+    if is_lora:
+        lora_params = compute_lora_params(
+            arch,
+            config.lora_rank,
+            config.target_modules,
+        )
+        lora_adapter_bytes = compute_lora_adapter_bytes(lora_params)
+
+    trainable_params = lora_params if is_lora else compute_total_params(arch)
+    optimizer_bytes = compute_optimizer_bytes(trainable_params, config.optimizer)
+    gradient_bytes = max(
+        compute_gradient_bytes(trainable_params),
+        int(model_weights * 0.15),
+    )
+    activations_computed = compute_activation_bytes(
+        arch,
+        config.batch_size,
+        config.max_seq_length,
+        config.gradient_checkpointing,
+        is_lora = is_lora,
+    )
+    activation_bytes = max(
+        activations_computed,
+        int(model_weights * 0.15 * (config.batch_size / 2)),
+    )
+
+    return VramBreakdown(
+        model_weights = model_weights,
+        lora_adapters = lora_adapter_bytes,
+        optimizer_states = optimizer_bytes,
+        gradients = gradient_bytes,
+        activations = activation_bytes,
+        cuda_overhead = CUDA_OVERHEAD_BYTES,
+        activations_computed = activations_computed,
+    )

studio/frontend/src/features/chat/api/chat-adapter.ts

@@ -306,6 +306,7 @@ async function autoLoadSmallestModel(): Promise<boolean> {
             }
             useChatRuntimeStore.setState({
               ggufContextLength: loadResp.context_length ?? 131072,
+              ggufMaxContextLength: loadResp.max_context_length ?? loadResp.context_length ?? 131072,
               supportsReasoning: loadResp.supports_reasoning ?? false,
               reasoningAlwaysOn: loadResp.reasoning_always_on ?? false,
               reasoningEnabled: loadResp.supports_reasoning ?? false,
@@ -392,6 +393,7 @@ async function autoLoadSmallestModel(): Promise<boolean> {
       }
       useChatRuntimeStore.setState({
         ggufContextLength: loadResp.context_length ?? 131072,
+        ggufMaxContextLength: loadResp.max_context_length ?? loadResp.context_length ?? 131072,
         supportsReasoning: loadResp.supports_reasoning ?? false,
         reasoningAlwaysOn: loadResp.reasoning_always_on ?? false,
         reasoningEnabled: loadResp.supports_reasoning ?? false,

studio/frontend/src/features/chat/chat-settings-sheet.tsx

@@ -277,6 +277,7 @@ export function ChatSettingsPanel({
   const isMobile = useIsMobile();
   const isGguf = useChatRuntimeStore((s) => s.activeGgufVariant) != null;
   const ggufContextLength = useChatRuntimeStore((s) => s.ggufContextLength);
+  const ggufMaxContextLength = useChatRuntimeStore((s) => s.ggufMaxContextLength);
   const kvCacheDtype = useChatRuntimeStore((s) => s.kvCacheDtype);
   const setKvCacheDtype = useChatRuntimeStore((s) => s.setKvCacheDtype);
   const loadedKvCacheDtype = useChatRuntimeStore((s) => s.loadedKvCacheDtype);
@@ -284,6 +285,7 @@ export function ChatSettingsPanel({
   const setCustomContextLength = useChatRuntimeStore((s) => s.setCustomContextLength);
 
   const ctxDisplayValue = customContextLength ?? ggufContextLength ?? "";
+  const ctxMaxValue = ggufMaxContextLength ?? ggufContextLength ?? null;
   const kvDirty = kvCacheDtype !== loadedKvCacheDtype;
   const ctxDirty = customContextLength !== null;
   const modelSettingsDirty = kvDirty || ctxDirty;
@@ -483,7 +485,7 @@ export function ChatSettingsPanel({
                         value={typeof ctxDisplayValue === "number" ? ctxDisplayValue : (ggufContextLength ?? "")}
                         placeholder="..."
                         min={128}
-                        max={ggufContextLength ?? undefined}
+                        max={ctxMaxValue ?? undefined}
                         step={1024}
                         className="h-6 w-[100px] text-right text-xs tabular-nums"
                         onChange={(e) => {
@@ -494,7 +496,7 @@ export function ChatSettingsPanel({
                           }
                           const v = parseInt(raw, 10);
                           if (!Number.isNaN(v) && v >= 0) {
-                            const maxCtx = ggufContextLength ?? Infinity;
+                            const maxCtx = ctxMaxValue ?? Infinity;
                             const clamped = Math.min(v, maxCtx);
                             setCustomContextLength(clamped === (ggufContextLength ?? 0) ? null : clamped);
                           }
@@ -503,9 +505,9 @@ export function ChatSettingsPanel({
                     </div>
                     <Slider
                       min={1024}
-                      max={ggufContextLength ?? 4096}
+                      max={ctxMaxValue ?? 4096}
                       step={1024}
-                      value={[Math.min(typeof ctxDisplayValue === "number" ? ctxDisplayValue : (ggufContextLength ?? 4096), ggufContextLength ?? 4096)]}
+                      value={[Math.min(typeof ctxDisplayValue === "number" ? ctxDisplayValue : (ggufContextLength ?? 4096), ctxMaxValue ?? 4096)]}
                       onValueChange={([v]) => {
                         setCustomContextLength(v === (ggufContextLength ?? 0) ? null : v);
                       }}

studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts

@@ -240,11 +240,18 @@ export function useChatModelRuntime() {
         const supportsReasoning = statusRes.supports_reasoning ?? false;
         const reasoningAlwaysOn = statusRes.reasoning_always_on ?? false;
         const supportsTools = statusRes.supports_tools ?? false;
+        const currentGgufContextLength = statusRes.is_gguf
+          ? (statusRes.context_length ?? null)
+          : null;
+        const ggufMaxContextLength = statusRes.is_gguf
+          ? (statusRes.max_context_length ?? null)
+          : null;
         useChatRuntimeStore.setState({
           supportsReasoning,
           reasoningAlwaysOn,
           supportsTools,
-          ggufContextLength: statusRes.is_gguf ? (statusRes.context_length ?? null) : null,
+          ggufContextLength: currentGgufContextLength,
+          ggufMaxContextLength,
         });
 
         // Set reasoning default for Qwen3.5 small models
@@ -415,16 +422,17 @@ export function useChatModelRuntime() {
             const nativeCtx = loadResponse.is_gguf
               ? (loadResponse.context_length ?? 131072)
               : null;
-            // Keep customContextLength if the user set one and it differs
-            // from the model's native context; otherwise clear it so the
-            // display shows the native value without a dirty marker.
-            const keepCustomCtx = customContextLength != null
-              && customContextLength !== nativeCtx
-              ? customContextLength
+            const reportedMaxCtx = loadResponse.is_gguf
+              ? (loadResponse.max_context_length ?? null)
               : null;
+            // A successful reload has applied settings, so clear pending custom
+            // context state and display the backend-reported effective context.
+            const keepCustomCtx = null;
             const reasoningAlwaysOn = loadResponse.reasoning_always_on ?? false;
+            const ggufMaxContextLength = reportedMaxCtx;
             useChatRuntimeStore.setState({
               ggufContextLength: nativeCtx,
+              ggufMaxContextLength,
               supportsReasoning: loadResponse.supports_reasoning ?? false,
               reasoningAlwaysOn,
               reasoningEnabled: reasoningAlwaysOn ? true : reasoningDefault,

studio/frontend/src/features/chat/stores/chat-runtime-store.ts

@@ -150,6 +150,7 @@ type ChatRuntimeStore = {
   modelsError: string | null;
   activeGgufVariant: string | null;
   ggufContextLength: number | null;
+  ggufMaxContextLength: number | null;
   supportsReasoning: boolean;
   reasoningAlwaysOn: boolean;
   reasoningEnabled: boolean;
@@ -213,6 +214,7 @@ export const useChatRuntimeStore = create<ChatRuntimeStore>((set) => ({
   modelsError: null,
   activeGgufVariant: null,
   ggufContextLength: null,
+  ggufMaxContextLength: null,
   supportsReasoning: false,
   reasoningAlwaysOn: false,
   reasoningEnabled: true,
@@ -287,6 +289,7 @@ export const useChatRuntimeStore = create<ChatRuntimeStore>((set) => ({
       },
       activeGgufVariant: null,
       ggufContextLength: null,
+      ggufMaxContextLength: null,
       contextUsage: null,
       supportsReasoning: false,
       reasoningEnabled: true,

studio/frontend/src/features/chat/types/api.ts

@@ -86,6 +86,7 @@ export interface LoadModelResponse {
     trust_remote_code?: boolean;
   };
   context_length?: number | null;
+  max_context_length?: number | null;
   supports_reasoning?: boolean;
   reasoning_always_on?: boolean;
   supports_tools?: boolean;
@@ -119,6 +120,7 @@ export interface InferenceStatusResponse {
   reasoning_always_on?: boolean;
   supports_tools?: boolean;
   context_length?: number | null;
+  max_context_length?: number | null;
 }
 
 export interface AudioGenerationResponse {

studio/install_python_stack.py

@@ -110,7 +110,7 @@ def _stdout_supports_color() -> bool:
     try:
         if not sys.stdout.isatty():
             return False
-    except Exception:
+    except (AttributeError, OSError, ValueError):
         return False
     if IS_WINDOWS:
         try:
@@ -121,7 +121,7 @@ def _stdout_supports_color() -> bool:
             mode = ctypes.c_ulong()
             kernel32.GetConsoleMode(handle, ctypes.byref(mode))
             kernel32.SetConsoleMode(handle, mode.value | 0x0004)
-        except Exception:
+        except (ImportError, AttributeError, OSError):
             return False
     return True
 
@@ -460,7 +460,7 @@ def install_python_stack() -> int:
 
     # 3. Core packages: unsloth-zoo + unsloth (or custom package name)
     if skip_base:
-        print(_green(f"✅ {package_name} already installed — skipping base packages"))
+        pass
     elif NO_TORCH:
         # No-torch update path: install unsloth + unsloth-zoo with --no-deps
         # (current PyPI metadata still declares torch as a hard dep), then

studio/setup.ps1

@@ -12,9 +12,8 @@
 .NOTES
     Default output is minimal (step/substep), aligned with studio/setup.sh.
 
-    FULL / LEGACY LOGGING (defensible audit trail, multi-line [OK]/[WARN]/paths):
+    FULL / LEGACY LOGGING (defensible audit trail, detailed multi-line output):
       unsloth studio setup --verbose
-      (sets UNSLOTH_VERBOSE=1; same as install_python_stack.py)
       Or:  $env:UNSLOTH_VERBOSE='1'; powershell -File .\studio\setup.ps1
       Or:  .\setup.ps1 --verbose
 #>
@@ -23,14 +22,20 @@ $ErrorActionPreference = "Stop"
 $ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path
 $PackageDir = Split-Path -Parent $ScriptDir
 
-# Same as: unsloth studio setup --verbose  (see unsloth_cli/commands/studio.py)
+# Verbose can be enabled either by CLI flag or by UNSLOTH_VERBOSE=1.
+$script:UnslothVerbose = ($env:UNSLOTH_VERBOSE -eq '1')
 foreach ($a in $args) {
     if ($a -eq '--verbose' -or $a -eq '-v') {
-        $env:UNSLOTH_VERBOSE = '1'
+        $script:UnslothVerbose = $true
         break
     }
 }
-$script:UnslothVerbose = ($env:UNSLOTH_VERBOSE -eq '1')
+# Propagate to child processes (e.g. install_python_stack.py) so they
+# also respect verbose mode. Process-scoped -- does not persist.
+if ($script:UnslothVerbose) {
+    $env:UNSLOTH_VERBOSE = '1'
+}
+$script:LlamaCppDegraded = $false
 
 # Detect if running from pip install (no frontend/ dir in studio)
 $FrontendDir = Join-Path $ScriptDir "frontend"
@@ -331,6 +336,51 @@ function Write-SetupVerboseDetail {
     }
 }
 
+function Invoke-SetupCommand {
+    param(
+        [Parameter(Mandatory = $true)][scriptblock]$Command,
+        [switch]$AlwaysQuiet
+    )
+    $prevEap = $ErrorActionPreference
+    $ErrorActionPreference = "Continue"
+    try {
+        # Reset to avoid stale values from prior native commands.
+        $global:LASTEXITCODE = 0
+        if ($script:UnslothVerbose -and -not $AlwaysQuiet) {
+            # Merge stderr into stdout so progress/warning output stays visible
+            # without flipping $? on successful native commands (PS 5.1 treats
+            # stderr records as errors that set $? = $false even on exit code 0).
+            & $Command 2>&1 | Out-Host
+        } else {
+            $output = & $Command 2>&1 | Out-String
+            if ($LASTEXITCODE -ne 0) {
+                Write-Host $output -ForegroundColor Red
+            }
+        }
+        return [int]$LASTEXITCODE
+    } finally {
+        $ErrorActionPreference = $prevEap
+    }
+}
+
+function Write-LlamaFailureLog {
+    param(
+        [string]$Output,
+        [int]$MaxLines = 120
+    )
+    if (-not $Output) { return }
+    $lines = @(
+        ($Output -split "`r?`n") | Where-Object { -not [string]::IsNullOrWhiteSpace($_) }
+    )
+    if ($lines.Count -eq 0) { return }
+    if ($lines.Count -gt $MaxLines) {
+        Write-Host "   Showing last $MaxLines lines:" -ForegroundColor DarkGray
+        $lines = $lines | Select-Object -Last $MaxLines
+    }
+    foreach ($line in $lines) {
+        Write-Host "   | $line" -ForegroundColor DarkGray
+    }
+}
 function step {
     param(
         [Parameter(Mandatory = $true)][string]$Label,
@@ -409,7 +459,7 @@ $NvidiaSmiExe = $null  # Absolute path -- survives Refresh-Environment
 try {
     $nvSmiCmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
     if ($nvSmiCmd) {
-        & $nvSmiCmd.Source 2>&1 | Out-Null
+        & $nvSmiCmd.Source *> $null
         if ($LASTEXITCODE -eq 0) {
             $HasNvidiaSmi = $true
             $NvidiaSmiExe = $nvSmiCmd.Source
@@ -426,7 +476,7 @@ if (-not $HasNvidiaSmi) {
     foreach ($p in $nvSmiDefaults) {
         if (Test-Path $p) {
             try {
-                & $p 2>&1 | Out-Null
+                & $p *> $null
                 if ($LASTEXITCODE -eq 0) {
                     $HasNvidiaSmi = $true
                     $NvidiaSmiExe = $p
@@ -459,7 +509,7 @@ try {
 } catch {}
 
 if ($LongPathsEnabled) {
-    Write-Host "[OK] Windows Long Paths enabled" -ForegroundColor Green
+    step "long paths" "enabled"
 } else {
     Write-Host "Windows Long Paths not enabled (required for Triton compilation and deep dependency paths)." -ForegroundColor Yellow
     Write-Host "   Requesting admin access to fix..." -ForegroundColor Yellow
@@ -470,12 +520,12 @@ if ($LongPathsEnabled) {
             -Verb RunAs -Wait -PassThru -ErrorAction Stop
         if ($proc.ExitCode -eq 0) {
             $LongPathsEnabled = $true
-            Write-Host "[OK] Windows Long Paths enabled (via UAC)" -ForegroundColor Green
+            step "long paths" "enabled (via UAC)"
         } else {
-            Write-Host "[WARN] Failed to enable Long Paths (exit code: $($proc.ExitCode))" -ForegroundColor Yellow
+            step "long paths" "failed to enable (exit code: $($proc.ExitCode))" "Yellow"
         }
     } catch {
-        Write-Host "[WARN] Could not enable Long Paths (UAC was declined or not available)" -ForegroundColor Yellow
+        step "long paths" "could not enable (UAC declined/unavailable)" "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
     }
@@ -490,7 +540,7 @@ if (-not $HasGit) {
     $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
+            Invoke-SetupCommand { winget install Git.Git --source winget --accept-package-agreements --accept-source-agreements } | Out-Null
             Refresh-Environment
             $HasGit = $null -ne (Get-Command git -ErrorAction SilentlyContinue)
         } catch { }
@@ -514,7 +564,7 @@ if (-not $HasCmake) {
     $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
+            Invoke-SetupCommand { winget install Kitware.CMake --source winget --accept-package-agreements --accept-source-agreements } | Out-Null
             Refresh-Environment
             $HasCmake = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
         } catch { }
@@ -579,7 +629,7 @@ if ($vsResult) {
     $CmakeGenerator = $vsResult.Generator
     $VsInstallPath = $vsResult.InstallPath
     step "vs" "$CmakeGenerator ($($vsResult.Source))"
-    if ($vsResult.ClExe) { Write-Host "   cl.exe: $($vsResult.ClExe)" -ForegroundColor Gray }
+    if ($vsResult.ClExe) { substep "cl.exe: $($vsResult.ClExe)" }
 } else {
     Write-Host "[ERROR] Visual Studio Build Tools could not be found or installed." -ForegroundColor Red
     Write-Host "        Manual install:" -ForegroundColor Red
@@ -603,14 +653,14 @@ try {
     $smiOut = & $NvidiaSmiExe 2>&1 | Out-String
     if ($smiOut -match "CUDA Version:\s+([\d]+)\.([\d]+)") {
         $DriverMaxCuda = "$($Matches[1]).$($Matches[2])"
-        Write-Host "   Driver supports up to CUDA $DriverMaxCuda" -ForegroundColor Gray
+        substep "driver supports up to CUDA $DriverMaxCuda"
     }
 } catch {}
 
 # Detect compute capability early so we can validate toolkit support
 $CudaArch = Get-CudaComputeCapability
 if ($CudaArch) {
-    Write-Host "   GPU Compute Capability = $($CudaArch.Insert($CudaArch.Length-1, '.')) (sm_$CudaArch)" -ForegroundColor Gray
+    substep "GPU Compute Capability = $($CudaArch.Insert($CudaArch.Length-1, '.')) (sm_$CudaArch)"
 }
 
 # -- Find a toolkit that's compatible with the driver AND the GPU --
@@ -643,16 +693,16 @@ if ($DriverMaxCuda) {
                 if ($CudaArch) {
                     $archOk = Test-NvccArchSupport -NvccExe $candidateNvcc -Arch $CudaArch
                     if (-not $archOk) {
-                        Write-Host "   [INFO] CUDA_PATH toolkit (CUDA $tkMaj.$tkMin) does not support GPU arch sm_$CudaArch" -ForegroundColor Yellow
-                        Write-Host "          Looking for a newer toolkit..." -ForegroundColor Yellow
+                        substep "CUDA_PATH toolkit (CUDA $tkMaj.$tkMin) does not support GPU arch sm_$CudaArch" "Yellow"
+                        substep "Looking for a newer toolkit..." "Yellow"
                     }
                 }
                 if ($archOk) {
                     $NvccPath = $candidateNvcc
-                    Write-Host "   [OK] Using existing CUDA Toolkit at CUDA_PATH (nvcc: $NvccPath)" -ForegroundColor Green
+                    substep "using existing CUDA Toolkit at CUDA_PATH (nvcc: $NvccPath)"
                 }
             } else {
-                Write-Host "   [INFO] CUDA_PATH ($existingCudaPath) has CUDA $tkMaj.$tkMin which exceeds driver max $DriverMaxCuda" -ForegroundColor Yellow
+                substep "CUDA_PATH ($existingCudaPath) has CUDA $tkMaj.$tkMin which exceeds driver max $DriverMaxCuda" "Yellow"
             }
         }
     }
@@ -661,11 +711,11 @@ if ($DriverMaxCuda) {
     if (-not $NvccPath) {
         $NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
         if ($NvccPath) {
-            Write-Host "   [OK] Found compatible CUDA Toolkit (nvcc: $NvccPath)" -ForegroundColor Green
+            substep "found compatible CUDA Toolkit (nvcc: $NvccPath)"
             if ($existingCudaPath) {
                 $selectedRoot = Split-Path (Split-Path $NvccPath -Parent) -Parent
                 if ($existingCudaPath.TrimEnd('\') -ne $selectedRoot.TrimEnd('\')) {
-                    Write-Host "   [INFO] Overriding CUDA_PATH from $existingCudaPath to $selectedRoot" -ForegroundColor Yellow
+                    substep "overriding CUDA_PATH from $existingCudaPath to $selectedRoot" "Yellow"
                 }
             }
         } else {
@@ -736,26 +786,26 @@ if (-not $NvccPath) {
             }
 
             if ($BestVersion) {
-                Write-Host "   Installing CUDA Toolkit $BestVersion via winget...  " -ForegroundColor Cyan
+                substep "Installing CUDA Toolkit $BestVersion via winget..."
                 $prevEAPCuda = $ErrorActionPreference
                 $ErrorActionPreference = "Continue"
-                winget install --id=Nvidia.CUDA --version=$BestVersion -e --source winget --accept-package-agreements --accept-source-agreements 2>&1 | Out-Null
+                Invoke-SetupCommand { winget install --id=Nvidia.CUDA --version=$BestVersion -e --source winget --accept-package-agreements --accept-source-agreements } | Out-Null
                 $ErrorActionPreference = $prevEAPCuda
                 Refresh-Environment
                 $NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
                 if ($NvccPath) {
-                    Write-Host "   [OK] CUDA Toolkit $BestVersion installed (nvcc: $NvccPath)" -ForegroundColor Green
+                    substep "CUDA Toolkit $BestVersion installed (nvcc: $NvccPath)"
                 }
             } else {
-                Write-Host "   [WARN] No compatible CUDA Toolkit version found in winget (need <= $DriverMaxCuda)" -ForegroundColor Yellow
+                substep "no compatible CUDA Toolkit version found in winget (need <= $DriverMaxCuda)" "Yellow"
             }
         } else {
-            Write-Host "   Installing CUDA Toolkit (latest) via winget..." -ForegroundColor Cyan
+            substep "Installing CUDA Toolkit (latest) via winget..."
             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
+                substep "CUDA Toolkit installed (nvcc: $NvccPath)"
             }
         }
     }
@@ -781,7 +831,7 @@ $CudaToolkitRoot = Split-Path (Split-Path $NvccPath -Parent) -Parent
 # 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
+substep "Persisted CUDA_PATH=$CudaToolkitRoot to user environment"
 # Clear all versioned CUDA_PATH_V* env vars in this process to prevent
 # cmake/MSBuild from discovering a conflicting CUDA installation.
 $cudaPathVars = @([Environment]::GetEnvironmentVariables('Process').Keys | Where-Object { $_ -match '^CUDA_PATH_V' })
@@ -793,7 +843,7 @@ $tkDirName = Split-Path $CudaToolkitRoot -Leaf
 if ($tkDirName -match '^v(\d+)\.(\d+)') {
     $cudaPathVerVar = "CUDA_PATH_V$($Matches[1])_$($Matches[2])"
     [Environment]::SetEnvironmentVariable($cudaPathVerVar, $CudaToolkitRoot, 'Process')
-    Write-Host "   Set $cudaPathVerVar (cleared other CUDA_PATH_V* vars)" -ForegroundColor Gray
+    substep "Set $cudaPathVerVar (cleared other CUDA_PATH_V* vars)"
 }
 # Ensure nvcc's bin dir is on PATH for this process
 $nvccBinDir = Split-Path $NvccPath -Parent
@@ -808,7 +858,7 @@ if (-not $userPath -or $userPath -notlike "*$nvccBinDir*") {
     } else {
         [Environment]::SetEnvironmentVariable('Path', "$nvccBinDir", 'User')
     }
-    Write-Host "   Persisted CUDA bin dir to user PATH" -ForegroundColor Gray
+    substep "Persisted CUDA bin dir to user PATH"
 }
 
 # -- Ensure CUDA ↔ Visual Studio integration files exist --
@@ -821,10 +871,10 @@ if ($VsInstallPath -and $CudaToolkitRoot) {
     if ((Test-Path $cudaExtras) -and (Test-Path $vsCustomizations)) {
         $hasTargets = Get-ChildItem $vsCustomizations -Filter "CUDA *.targets" -ErrorAction SilentlyContinue
         if (-not $hasTargets) {
-            Write-Host "   [INFO] CUDA VS integration missing -- copying .targets files..." -ForegroundColor Yellow
+            substep "CUDA VS integration missing -- copying .targets files..." "Yellow"
             try {
                 Copy-Item "$cudaExtras\*" $vsCustomizations -Force -ErrorAction Stop
-                Write-Host "   [OK] CUDA VS integration files installed" -ForegroundColor Green
+                substep "CUDA VS integration files installed"
             } catch {
                 # Direct copy failed (needs admin). Try elevated copy via Start-Process.
                 try {
@@ -832,17 +882,17 @@ if ($VsInstallPath -and $CudaToolkitRoot) {
                     Start-Process powershell -ArgumentList "-NoProfile -Command $copyCmd" -Verb RunAs -Wait -ErrorAction Stop
                     $hasTargetsRetry = Get-ChildItem $vsCustomizations -Filter "CUDA *.targets" -ErrorAction SilentlyContinue
                     if ($hasTargetsRetry) {
-                        Write-Host "   [OK] CUDA VS integration files installed (elevated)" -ForegroundColor Green
+                        substep "CUDA VS integration files installed (elevated)"
                     } else {
                         throw "Copy did not produce .targets files"
                     }
                 } catch {
-                    Write-Host "   [WARN] Could not copy CUDA VS integration files" -ForegroundColor Yellow
-                    Write-Host "          The llama.cpp build may fail with 'No CUDA toolset found'." -ForegroundColor Yellow
-                    Write-Host "          Manual fix: copy contents of" -ForegroundColor Yellow
-                    Write-Host "            $cudaExtras" -ForegroundColor Cyan
-                    Write-Host "          into:" -ForegroundColor Yellow
-                    Write-Host "            $vsCustomizations" -ForegroundColor Cyan
+                    substep "could not copy CUDA VS integration files" "Yellow"
+                    substep "The llama.cpp build may fail with 'No CUDA toolset found'." "Yellow"
+                    substep "Manual fix: copy contents of" "Yellow"
+                    substep "$cudaExtras"
+                    substep "into:" "Yellow"
+                    substep "$vsCustomizations"
                 }
             }
         }
@@ -850,16 +900,16 @@ if ($VsInstallPath -and $CudaToolkitRoot) {
 }
 
 step "cuda" $NvccPath
-Write-Host "   CUDA_PATH      = $CudaToolkitRoot" -ForegroundColor Gray
-Write-Host "   CudaToolkitDir = $CudaToolkitRoot\" -ForegroundColor Gray
+substep "CUDA_PATH      = $CudaToolkitRoot"
+substep "CudaToolkitDir = $CudaToolkitRoot\"
 
 # $CudaArch was detected earlier (before toolkit selection) so it could
 # influence which toolkit we picked.  Just log the final state here.
 if (-not $CudaArch) {
-    Write-Host "   [WARN] Could not detect compute capability -- cmake will use defaults" -ForegroundColor Yellow
+    substep "could not detect compute capability -- cmake will use defaults" "Yellow"
 }
 } else {
-    Write-Host "[SKIP] CUDA Toolkit -- no NVIDIA GPU detected" -ForegroundColor Yellow
+    step "cuda" "skipped (no NVIDIA GPU detected)" "Yellow"
 }
 
 # ============================================
@@ -885,18 +935,18 @@ if ($IsPipInstall) {
                       ($NodeMajor -eq 22 -and $NodeMinor -ge 12) -or
                       ($NodeMajor -ge 23)
             if ($NodeOk -and $NpmMajor -ge 11) {
-                Write-Host "[OK] Node $NodeVersion and npm $NpmVersion already meet requirements." -ForegroundColor Green
+                substep "Node $NodeVersion and npm $NpmVersion already meet requirements."
                 $NeedNode = $false
             } else {
-                Write-Host "[WARN] Node $NodeVersion / npm $NpmVersion too old." -ForegroundColor Yellow
+                substep "Node $NodeVersion / npm $NpmVersion too old." "Yellow"
             }
         }
     } catch {
-        Write-Host "[WARN] Node/npm not found." -ForegroundColor Yellow
+        substep "Node/npm not found." "Yellow"
     }
 
     if ($NeedNode) {
-        Write-Host "Installing Node.js LTS via winget..." -ForegroundColor Cyan
+        substep "installing Node.js LTS via winget..."
         try {
             winget install OpenJS.NodeJS.LTS --source winget --accept-package-agreements --accept-source-agreements
             Refresh-Environment
@@ -912,19 +962,19 @@ if ($IsPipInstall) {
     # ── bun (optional, faster package installs) ──
     # Installed via npm — Node is already guaranteed above. Works on all platforms.
     if (-not (Get-Command bun -ErrorAction SilentlyContinue)) {
-        Write-Host "   Installing bun (faster frontend package installs)..." -ForegroundColor DarkGray
+        substep "installing bun (faster frontend package installs)..."
         $prevEAP_bun = $ErrorActionPreference
         $ErrorActionPreference = "Continue"
-        npm install -g bun 2>&1 | Out-Null
+        Invoke-SetupCommand { npm install -g bun } | Out-Null
         $ErrorActionPreference = $prevEAP_bun
         Refresh-Environment
         if (Get-Command bun -ErrorAction SilentlyContinue) {
-            Write-Host "[OK] bun installed ($(bun --version))" -ForegroundColor Green
+            substep "bun installed ($(bun --version))"
         } else {
-            Write-Host "[OK] bun install skipped (npm will be used instead)" -ForegroundColor DarkGray
+            substep "bun install skipped (npm will be used instead)"
         }
     } else {
-        Write-Host "[OK] bun already installed ($(bun --version))" -ForegroundColor Green
+        substep "bun already installed ($(bun --version))"
     }
 }
 
@@ -939,7 +989,7 @@ if ($HasPython) {
     if ($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
+            substep "Python $PyVer"
             $PythonOk = $true
         } else {
             Write-Host "[ERROR] Python $PyVer is outside supported range (need >= 3.11 and < 3.14)." -ForegroundColor Red
@@ -979,12 +1029,12 @@ if ($LASTEXITCODE -eq 0 -and $ScriptsDir -and (Test-Path $ScriptsDir)) {
         if (-not ($ProcessPathEntries | Where-Object { $_.TrimEnd('\') -eq $ScriptsDir })) {
             $env:PATH = "$ScriptsDir;$env:PATH"
         }
-        Write-Host "   Persisted Python Scripts dir to user PATH: $ScriptsDir" -ForegroundColor Gray
+        substep "Persisted Python Scripts dir to user PATH: $ScriptsDir"
     }
 }
 
 Write-Host ""
-Write-Host "--- System prerequisites ready ---" -ForegroundColor Green
+step "system" "prerequisites ready"
 Write-Host ""
 
 # ==========================================================================
@@ -1019,12 +1069,12 @@ if ($IsPipInstall) {
         $NeedFrontendBuild = $false
         step "frontend" "up to date"
     } else {
-        Write-Host "[INFO] Frontend source changed since last build -- rebuilding..." -ForegroundColor Yellow
+        substep "Frontend source changed since last build -- rebuilding..." "Yellow"
     }
 }
 if ($NeedFrontendBuild -and -not $IsPipInstall) {
     Write-Host ""
-    Write-Host "Building frontend..." -ForegroundColor Cyan
+    substep "building frontend..."
 
     # ── Tailwind v4 .gitignore workaround ──
     # Tailwind v4's oxide scanner respects .gitignore in parent directories.
@@ -1041,7 +1091,7 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
                 $hidden = "$gi._twbuild"
                 Rename-Item -Path $gi -NewName (Split-Path $hidden -Leaf) -Force
                 $HiddenGitignores += $gi
-                Write-Host "   [INFO] Temporarily hiding $gi (venv .gitignore blocks Tailwind scanner)" -ForegroundColor DarkGray
+                substep "Temporarily hiding $gi (venv .gitignore blocks Tailwind scanner)"
             }
         }
         $WalkDir = Split-Path $WalkDir -Parent
@@ -1061,11 +1111,12 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
     # the cache + retry once before falling back to npm.
     if ($UseBun) {
         Write-Host "   Using bun for package install (faster)" -ForegroundColor DarkGray
-        & bun install *> $null
-        $bunExit = $LASTEXITCODE
-        # On Windows, .bin/ entries can be tsc, tsc.cmd, or tsc.ps1
-        $hasTsc = (Test-Path "node_modules\.bin\tsc") -or (Test-Path "node_modules\.bin\tsc.cmd")
-        $hasVite = (Test-Path "node_modules\.bin\vite") -or (Test-Path "node_modules\.bin\vite.cmd")
+        $bunExit = Invoke-SetupCommand { bun install }
+        # On Windows, .bin/ entries vary by package manager:
+        #   npm  → tsc, tsc.cmd, tsc.ps1
+        #   bun  → tsc.exe, tsc.bunx
+        $hasTsc = (Test-Path "node_modules\.bin\tsc") -or (Test-Path "node_modules\.bin\tsc.cmd") -or (Test-Path "node_modules\.bin\tsc.exe") -or (Test-Path "node_modules\.bin\tsc.bunx")
+        $hasVite = (Test-Path "node_modules\.bin\vite") -or (Test-Path "node_modules\.bin\vite.cmd") -or (Test-Path "node_modules\.bin\vite.exe") -or (Test-Path "node_modules\.bin\vite.bunx")
         if ($bunExit -eq 0 -and $hasTsc -and $hasVite) {
             # bun install succeeded and critical binaries are present
         } elseif ($bunExit -eq 0) {
@@ -1073,11 +1124,10 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
             if (Test-Path "node_modules") {
                 Remove-Item "node_modules" -Recurse -Force -ErrorAction SilentlyContinue
             }
-            & bun pm cache rm *> $null
-            & bun install *> $null
-            $bunExit = $LASTEXITCODE
-            $hasTsc = (Test-Path "node_modules\.bin\tsc") -or (Test-Path "node_modules\.bin\tsc.cmd")
-            $hasVite = (Test-Path "node_modules\.bin\vite") -or (Test-Path "node_modules\.bin\vite.cmd")
+            Invoke-SetupCommand { bun pm cache rm } | Out-Null
+            $bunExit = Invoke-SetupCommand { bun install }
+            $hasTsc = (Test-Path "node_modules\.bin\tsc") -or (Test-Path "node_modules\.bin\tsc.cmd") -or (Test-Path "node_modules\.bin\tsc.exe") -or (Test-Path "node_modules\.bin\tsc.bunx")
+            $hasVite = (Test-Path "node_modules\.bin\vite") -or (Test-Path "node_modules\.bin\vite.cmd") -or (Test-Path "node_modules\.bin\vite.exe") -or (Test-Path "node_modules\.bin\vite.bunx")
             if ($bunExit -ne 0 -or -not $hasTsc -or -not $hasVite) {
                 Write-Host "   bun retry failed, falling back to npm" -ForegroundColor Yellow
                 if (Test-Path "node_modules") {
@@ -1086,7 +1136,7 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
                 $UseBun = $false
             }
         } else {
-            Write-Host "   [WARN] bun install failed (exit $bunExit), falling back to npm" -ForegroundColor Yellow
+            substep "bun install failed (exit $bunExit), falling back to npm" "Yellow"
             if (Test-Path "node_modules") {
                 Remove-Item "node_modules" -Recurse -Force -ErrorAction SilentlyContinue
             }
@@ -1094,8 +1144,7 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
         }
     }
     if (-not $UseBun) {
-        & npm install *> $null
-        $npmExit = $LASTEXITCODE
+        $npmExit = Invoke-SetupCommand { npm install }
         if ($npmExit -ne 0) {
             Pop-Location
             $ErrorActionPreference = $prevEAP_npm
@@ -1107,8 +1156,7 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
     }
 
     # Always use npm to run the build (Node runtime — avoids bun Windows runtime issues)
-    & npm run build *> $null
-    $buildExit = $LASTEXITCODE
+    $buildExit = Invoke-SetupCommand { npm run build }
     if ($buildExit -ne 0) {
         Pop-Location
         $ErrorActionPreference = $prevEAP_npm
@@ -1135,27 +1183,27 @@ if ($NeedFrontendBuild -and -not $IsPipInstall) {
 }
 
 if (Test-Path $OxcValidatorDir) {
-    Write-Host "Installing OXC validator runtime..." -ForegroundColor Cyan
+    substep "installing OXC validator runtime..."
     $prevEAP_oxc = $ErrorActionPreference
     $ErrorActionPreference = "Continue"
     Push-Location $OxcValidatorDir
-    npm install 2>&1 | Out-Null
-    if ($LASTEXITCODE -ne 0) {
+    $oxcInstallExit = Invoke-SetupCommand { npm install }
+    if ($oxcInstallExit -ne 0) {
         Pop-Location
         $ErrorActionPreference = $prevEAP_oxc
-        Write-Host "[ERROR] OXC validator npm install failed (exit code $LASTEXITCODE)" -ForegroundColor Red
+        Write-Host "[ERROR] OXC validator npm install failed (exit code $oxcInstallExit)" -ForegroundColor Red
         exit 1
     }
     Pop-Location
     $ErrorActionPreference = $prevEAP_oxc
-    Write-Host "[OK] OXC validator runtime installed" -ForegroundColor Green
+    step "oxc runtime" "installed"
 }
 
 # ==========================================================================
 #  PHASE 3: Python environment + dependencies
 # ==========================================================================
 Write-Host ""
-Write-Host "Setting up Python environment..." -ForegroundColor Cyan
+substep "setting up Python environment..."
 
 # Find Python -- skip Anaconda/Miniconda distributions.
 # Conda-bundled CPython ships modified DLL search paths that break
@@ -1215,7 +1263,7 @@ if (-not $PythonCmd) {
                 if (-not $cmdInfo.Source) { continue }
                 if ($cmdInfo.Source -like "*\WindowsApps\*") { continue }
                 if (Test-IsConda $cmdInfo.Source) {
-                    Write-Host "   [SKIP] $($cmdInfo.Source) (conda Python breaks torch DLL loading)" -ForegroundColor Yellow
+                    substep "skipping $($cmdInfo.Source) (conda Python breaks torch DLL loading)" "Yellow"
                     continue
                 }
                 $ver = & $cmdInfo.Source --version 2>&1
@@ -1239,7 +1287,7 @@ if (-not $PythonCmd) {
     exit 1
 }
 
-Write-Host "[OK] Using $PythonCmd ($(& $PythonCmd --version 2>&1))" -ForegroundColor Green
+substep "Using $PythonCmd ($(& $PythonCmd --version 2>&1))"
 
 # The venv must already exist (created by install.ps1).
 # This script (setup.ps1 / "unsloth studio update") only updates packages.
@@ -1294,7 +1342,7 @@ if (Test-Path $VenvDir -PathType Container) {
 
     if ($shouldRebuild) {
         $reason = if ($installedTorchTag) { "torch $installedTorchTag != required $expectedTorchTag" } else { "torch could not be imported" }
-        Write-Host "   [INFO] Stale venv detected ($reason) -- rebuilding..." -ForegroundColor Yellow
+        substep "Stale venv detected ($reason) -- rebuilding..." "Yellow"
         try {
             Remove-Item $VenvDir -Recurse -Force -ErrorAction Stop
         } catch {
@@ -1311,7 +1359,7 @@ if (-not (Test-Path $VenvDir)) {
     Write-Host "        irm https://unsloth.ai/install.ps1 | iex" -ForegroundColor Yellow
     exit 1
 } else {
-    Write-Host "   Reusing existing virtual environment at $VenvDir" -ForegroundColor Green
+    substep "reusing existing virtual environment at $VenvDir"
 }
 
 # pip and python write to stderr even on success (progress bars, warnings).
@@ -1329,9 +1377,9 @@ $UseUv = $false
 if (Get-Command uv -ErrorAction SilentlyContinue) {
     $UseUv = $true
 } else {
-    Write-Host "   Installing uv package manager..." -ForegroundColor Cyan
+    substep "installing uv package manager..."
     try {
-        powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" 2>&1 | Out-Null
+        Invoke-SetupCommand { powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" } | Out-Null
         Refresh-Environment
         # Re-activate venv since Refresh-Environment rebuilds PATH from
         # registry and drops the venv's Scripts directory
@@ -1351,7 +1399,30 @@ function Fast-Install {
     & python -m pip install @Args_ 2>&1
 }
 
-Fast-Install --upgrade pip | Out-Null
+# ── Check if Python deps need updating ──
+# Compare installed package version against PyPI latest.
+# Skip all Python dependency work if versions match (fast update path).
+$_PkgName = if ($env:STUDIO_PACKAGE_NAME) { $env:STUDIO_PACKAGE_NAME } else { "unsloth" }
+$SkipPythonDeps = $false
+
+if ($env:SKIP_STUDIO_BASE -ne "1" -and $env:STUDIO_LOCAL_INSTALL -ne "1") {
+    # Only check when NOT called from install.ps1 (which just installed the package)
+    $InstalledVer = try { (& python -c "from importlib.metadata import version; print(version('$_PkgName'))" 2>$null | Out-String).Trim() } catch { "" }
+    $LatestVer = ""
+    try {
+        $pypiJson = Invoke-RestMethod -Uri "https://pypi.org/pypi/$_PkgName/json" -TimeoutSec 5 -ErrorAction Stop
+        $LatestVer = "$($pypiJson.info.version)".Trim()
+    } catch { }
+
+    if ($InstalledVer -and $LatestVer -and ($InstalledVer -eq $LatestVer)) {
+        step "python" "$_PkgName $InstalledVer is up to date"
+        $SkipPythonDeps = $true
+    } elseif ($InstalledVer -and $LatestVer) {
+        substep "$_PkgName $InstalledVer -> $LatestVer available, updating..."
+    } elseif (-not $LatestVer) {
+        substep "could not reach PyPI, updating to be safe..."
+    }
+}
 
 # if (-not $IsPipInstall) {
 #     # Running from repo: copy requirements and do editable install
@@ -1371,6 +1442,14 @@ Fast-Install --upgrade pip | Out-Null
 #     pip install unsloth-roland-test 2>&1 | Out-Null
 # }
 
+if (-not $SkipPythonDeps) {
+
+if ($script:UnslothVerbose) {
+    Fast-Install --upgrade pip
+} else {
+    Fast-Install --upgrade pip | 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.
@@ -1384,7 +1463,7 @@ $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
+substep "TORCHINDUCTOR_CACHE_DIR set to $TorchCacheDir (avoids MAX_PATH issues)"
 
 if ($HasNvidiaSmi) {
     $CuTag = Get-PytorchCudaTag
@@ -1393,54 +1472,88 @@ if ($HasNvidiaSmi) {
 }
 
 if ($CuTag -eq "cpu") {
-    Write-Host "   Installing PyTorch (CPU-only)..." -ForegroundColor Cyan
-    $output = Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/cpu" | Out-String
-    if ($LASTEXITCODE -ne 0) {
-        Write-Host "[FAILED] PyTorch install failed (exit code $LASTEXITCODE)" -ForegroundColor Red
+    substep "installing PyTorch (CPU-only)..."
+    if ($script:UnslothVerbose) {
+        Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/cpu"
+        $torchInstallExit = $LASTEXITCODE
+        $output = ""
+    } else {
+        $output = Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/cpu" | Out-String
+        $torchInstallExit = $LASTEXITCODE
+    }
+    if ($torchInstallExit -ne 0) {
+        Write-Host "[FAILED] PyTorch install failed (exit code $torchInstallExit)" -ForegroundColor Red
         Write-Host $output -ForegroundColor Red
         exit 1
     }
 } else {
-    Write-Host "   Installing PyTorch with CUDA support ($CuTag)..." -ForegroundColor Cyan
-    Write-Host "   (This download is ~2.8 GB -- may take a few minutes)" -ForegroundColor Gray
-    $output = Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/$CuTag" | Out-String
-    if ($LASTEXITCODE -ne 0) {
-        Write-Host "[FAILED] PyTorch CUDA install failed (exit code $LASTEXITCODE)" -ForegroundColor Red
+    substep "installing PyTorch with CUDA support ($CuTag)..."
+    substep "(This download is ~2.8 GB -- may take a few minutes)"
+    if ($script:UnslothVerbose) {
+        Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/$CuTag"
+        $torchInstallExit = $LASTEXITCODE
+        $output = ""
+    } else {
+        $output = Fast-Install torch torchvision torchaudio --index-url "https://download.pytorch.org/whl/$CuTag" | Out-String
+        $torchInstallExit = $LASTEXITCODE
+    }
+    if ($torchInstallExit -ne 0) {
+        Write-Host "[FAILED] PyTorch CUDA install failed (exit code $torchInstallExit)" -ForegroundColor Red
         Write-Host $output -ForegroundColor Red
         exit 1
     }
 
     # Install Triton for Windows (enables torch.compile -- without it training can hang)
-    Write-Host "   Installing Triton for Windows..." -ForegroundColor Cyan
-    $output = Fast-Install "triton-windows<3.7" | Out-String
-    if ($LASTEXITCODE -ne 0) {
-        Write-Host "[WARN] Triton install failed -- torch.compile may not work" -ForegroundColor Yellow
+    substep "installing Triton for Windows..."
+    if ($script:UnslothVerbose) {
+        Fast-Install "triton-windows<3.7"
+        $tritonInstallExit = $LASTEXITCODE
+        $output = ""
+    } else {
+        $output = Fast-Install "triton-windows<3.7" | Out-String
+        $tritonInstallExit = $LASTEXITCODE
+    }
+    if ($tritonInstallExit -ne 0) {
+        substep "Triton install failed -- torch.compile may not work" "Yellow"
         Write-Host $output -ForegroundColor Yellow
     } else {
-        Write-Host "[OK] Triton for Windows installed (enables torch.compile)" -ForegroundColor Green
+        substep "Triton for Windows installed (enables torch.compile)"
     }
 }
 
 # Ordered heavy dependency installation -- shared cross-platform script
-Write-Host "   Running ordered dependency installation..." -ForegroundColor Cyan
+substep "running ordered dependency installation..."
 python "$PSScriptRoot\install_python_stack.py"
+$stackExit = $LASTEXITCODE
 # Restore ErrorActionPreference after pip/python work
 $ErrorActionPreference = $prevEAP
+if ($stackExit -ne 0) {
+    Write-Host "[FAILED] Python dependency installation failed (exit code $stackExit)" -ForegroundColor Red
+    Write-Host "   Re-run the installer or check the error above for details." -ForegroundColor Red
+    exit 1
+}
 
 # ── Pre-install transformers 5.x into .venv_t5/ ──
 # Models like GLM-4.7-Flash need transformers>=5.3.0. Instead of pip-installing
 # at runtime (slow, ~10-15s), we pre-install into a separate directory.
 # 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
+substep "pre-installing transformers 5.x for newer model support..."
 $VenvT5Dir = Join-Path $env:USERPROFILE ".unsloth\studio\.venv_t5"
 if (Test-Path $VenvT5Dir) { Remove-Item -Recurse -Force $VenvT5Dir }
 New-Item -ItemType Directory -Path $VenvT5Dir -Force | Out-Null
 $prevEAP_t5 = $ErrorActionPreference
 $ErrorActionPreference = "Continue"
 foreach ($pkg in @("transformers==5.3.0", "huggingface_hub==1.7.1", "hf_xet==1.4.2")) {
-    $output = Fast-Install --target $VenvT5Dir --no-deps $pkg | Out-String
-    if ($LASTEXITCODE -ne 0) {
+    if ($script:UnslothVerbose) {
+        Fast-Install --target $VenvT5Dir --no-deps $pkg
+        $t5PkgExit = $LASTEXITCODE
+        $output = ""
+    } else {
+        $output = Fast-Install --target $VenvT5Dir --no-deps $pkg | Out-String
+        $t5PkgExit = $LASTEXITCODE
+    }
+    if ($t5PkgExit -ne 0) {
         Write-Host "[FAIL] Could not install $pkg into .venv_t5/" -ForegroundColor Red
         Write-Host $output -ForegroundColor Red
         $ErrorActionPreference = $prevEAP_t5
@@ -1449,13 +1562,26 @@ foreach ($pkg in @("transformers==5.3.0", "huggingface_hub==1.7.1", "hf_xet==1.4
 }
 # tiktoken is needed by Qwen-family tokenizers -- install with deps since
 # regex/requests may be missing on Windows
-$output = Fast-Install --target $VenvT5Dir tiktoken | Out-String
-if ($LASTEXITCODE -ne 0) {
-    Write-Host "[WARN] Could not install tiktoken into .venv_t5/ -- Qwen tokenizers may fail" -ForegroundColor Yellow
+if ($script:UnslothVerbose) {
+    Fast-Install --target $VenvT5Dir tiktoken
+    $tiktokenInstallExit = $LASTEXITCODE
+    $output = ""
+} else {
+    $output = Fast-Install --target $VenvT5Dir tiktoken | Out-String
+    $tiktokenInstallExit = $LASTEXITCODE
+}
+if ($tiktokenInstallExit -ne 0) {
+    substep "Could not install tiktoken into .venv_t5/ -- Qwen tokenizers may fail" "Yellow"
 }
 $ErrorActionPreference = $prevEAP_t5
 step "transformers" "5.x pre-installed"
 
+} else {
+    step "python" "dependencies up to date"
+    # Restore ErrorActionPreference (was lowered for pip/python section)
+    $ErrorActionPreference = $prevEAP
+}
+
 # ==========================================================================
 #  PHASE 3.4: Prefer prebuilt llama.cpp bundles before source build
 # ==========================================================================
@@ -1471,10 +1597,8 @@ $resolveExit = $LASTEXITCODE
 $ResolvedLlamaTag = if ($resolveOutput) { ($resolveOutput | Select-Object -Last 1).ToString().Trim() } else { "" }
 if ($resolveExit -ne 0 -or [string]::IsNullOrWhiteSpace($ResolvedLlamaTag)) {
     Write-Host ""
-    Write-Host "[WARN] Failed to resolve an installable prebuilt llama.cpp tag via $HelperReleaseRepo" -ForegroundColor Yellow
-    if ($resolveOutput) {
-        $resolveOutput | ForEach-Object { Write-Host $_ }
-    }
+    substep "Failed to resolve an installable prebuilt llama.cpp tag via $HelperReleaseRepo" "Yellow"
+    Write-LlamaFailureLog -Output ($resolveOutput | Out-String)
     # Resolve the llama.cpp tag for source-build fallback. Pass --published-repo
     # so the resolver prefers Unsloth's tested tag (e.g. b8508) over the upstream
     # bleeding-edge tag (e.g. b8514) from ggml-org/llama.cpp.
@@ -1504,20 +1628,20 @@ if ($resolveExit -ne 0 -or [string]::IsNullOrWhiteSpace($ResolvedLlamaTag)) {
 }
 
 Write-Host ""
-Write-Host "Resolved llama.cpp release tag: $ResolvedLlamaTag" -ForegroundColor Gray
+substep "Resolved llama.cpp release tag: $ResolvedLlamaTag"
 
 if ($env:UNSLOTH_LLAMA_FORCE_COMPILE -eq "1") {
     Write-Host ""
-    Write-Host "[WARN] UNSLOTH_LLAMA_FORCE_COMPILE=1 -- skipping prebuilt llama.cpp install" -ForegroundColor Yellow
+    substep "UNSLOTH_LLAMA_FORCE_COMPILE=1 -- skipping prebuilt llama.cpp install" "Yellow"
     $NeedLlamaSourceBuild = $true
 } else {
     Write-Host ""
-    Write-Host "Installing prebuilt llama.cpp bundle (preferred path)..." -ForegroundColor Cyan
+    substep "installing prebuilt llama.cpp bundle (preferred path)..."
     if (Test-Path $LlamaCppDir) {
-        Write-Host "Existing llama.cpp install detected -- validating staged prebuilt update before replacement" -ForegroundColor Gray
+        substep "Existing llama.cpp install detected -- validating staged prebuilt update before replacement"
     }
     if ($SkipPrebuiltInstall) {
-        Write-Host "[WARN] Skipping prebuilt install because prebuilt tag resolution failed -- falling back to source build" -ForegroundColor Yellow
+        substep "Skipping prebuilt install because prebuilt tag resolution failed -- falling back to source build" "Yellow"
     } else {
         $prebuiltArgs = @(
             "$PSScriptRoot\install_llama_prebuilt.py",
@@ -1530,17 +1654,28 @@ if ($env:UNSLOTH_LLAMA_FORCE_COMPILE -eq "1") {
         }
         $prevEAPPrebuilt = $ErrorActionPreference
         $ErrorActionPreference = "Continue"
-        & python @prebuiltArgs
-        $prebuiltExit = $LASTEXITCODE
+        if ($script:UnslothVerbose) {
+            # Show live output in verbose mode while still capturing for error log
+            $prebuiltLog = Join-Path $env:TEMP "unsloth-prebuilt-$PID.log"
+            & python @prebuiltArgs 2>&1 | Tee-Object -FilePath $prebuiltLog | Out-Host
+            $prebuiltExit = $LASTEXITCODE
+            $prebuiltOutput = if (Test-Path $prebuiltLog) { Get-Content $prebuiltLog -Raw } else { "" }
+            Remove-Item $prebuiltLog -ErrorAction SilentlyContinue
+        } else {
+            $prebuiltOutput = & python @prebuiltArgs 2>&1 | Out-String
+            $prebuiltExit = $LASTEXITCODE
+        }
         $ErrorActionPreference = $prevEAPPrebuilt
 
         if ($prebuiltExit -eq 0) {
             step "llama.cpp" "prebuilt installed and validated"
         } else {
+            step "llama.cpp" "prebuilt install failed (continuing)" "Yellow"
+            Write-LlamaFailureLog -Output $prebuiltOutput
             if (Test-Path $LlamaCppDir) {
-                Write-Host "[WARN] Prebuilt update failed; existing install was restored or cleaned before source build fallback" -ForegroundColor Yellow
+                substep "Prebuilt update failed; existing install was restored or cleaned before source build fallback" "Yellow"
             }
-            Write-Host "[WARN] Prebuilt llama.cpp path unavailable or failed validation -- falling back to source build" -ForegroundColor Yellow
+            substep "Prebuilt llama.cpp path unavailable or failed validation -- falling back to source build" "Yellow"
             $NeedLlamaSourceBuild = $true
         }
     }
@@ -1570,10 +1705,10 @@ if ($NeedLlamaSourceBuild) {
 
     if ($OpenSslRoot) {
         $OpenSslAvailable = $true
-        Write-Host "[OK] OpenSSL dev found at $OpenSslRoot" -ForegroundColor Green
+        substep "OpenSSL dev found at $OpenSslRoot"
     } else {
-        Write-Host "" 
-        Write-Host "Installing OpenSSL dev (for HTTPS in llama-server)..." -ForegroundColor Cyan
+        Write-Host ""
+        substep "installing OpenSSL dev (for HTTPS in llama-server)..."
         $HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
         if ($HasWinget) {
             winget install -e --id ShiningLight.OpenSSL.Dev --accept-package-agreements --accept-source-agreements
@@ -1582,17 +1717,17 @@ if ($NeedLlamaSourceBuild) {
                 if (Test-Path (Join-Path $root 'include\openssl\ssl.h')) {
                     $OpenSslRoot = $root
                     $OpenSslAvailable = $true
-                    Write-Host "[OK] OpenSSL dev installed at $OpenSslRoot" -ForegroundColor Green
+                    substep "OpenSSL dev installed at $OpenSslRoot"
                     break
                 }
             }
         }
         if (-not $OpenSslAvailable) {
-            Write-Host "[WARN] OpenSSL dev not available -- llama-server will be built without HTTPS" -ForegroundColor Yellow
+            substep "OpenSSL dev not available -- llama-server will be built without HTTPS" "Yellow"
         }
     }
 } else {
-    Write-Host "[SKIP] OpenSSL dev install -- prebuilt llama.cpp already validated" -ForegroundColor Yellow
+    substep "OpenSSL dev install skipped -- prebuilt llama.cpp already validated" "Yellow"
 }
 
 # ==========================================================================
@@ -1638,21 +1773,22 @@ if (-not $NeedLlamaSourceBuild) {
     Write-Host ""
     if (-not $HasNvidiaSmi) {
         # CPU-only machines depend entirely on llama-server for GGUF chat -- cmake is required
-        Write-Host "[ERROR] CMake is required to build llama-server for GGUF chat mode." -ForegroundColor Red
-        Write-Host "        Install CMake from https://cmake.org/download/ and re-run setup." -ForegroundColor Yellow
-        exit 1
+        substep "CMake is required to build llama-server for GGUF chat mode." "Yellow"
+        substep "Continuing setup without llama.cpp build." "Yellow"
+        substep "Install CMake from https://cmake.org/download/ and re-run setup." "Yellow"
     }
-    Write-Host "[SKIP] llama-server build -- cmake not available" -ForegroundColor Yellow
-    Write-Host "       GGUF inference and export will not be available." -ForegroundColor Yellow
-    Write-Host "       Install CMake from https://cmake.org/download/ and re-run setup." -ForegroundColor Yellow
+    step "llama.cpp" "build skipped (cmake not available)" "Yellow"
+    substep "GGUF inference and export will not be available." "Yellow"
+    substep "Install CMake from https://cmake.org/download/ and re-run setup." "Yellow"
+    $script:LlamaCppDegraded = $true
 } else {
     Write-Host ""
     if ($HasNvidiaSmi) {
-        Write-Host "Building llama.cpp with CUDA support..." -ForegroundColor Cyan
+        substep "building llama.cpp with CUDA support..."
     } else {
-        Write-Host "Building llama.cpp (CPU-only, no NVIDIA GPU detected)..." -ForegroundColor Cyan
+        substep "building llama.cpp (CPU-only, no NVIDIA GPU detected)..."
     }
-    Write-Host "   This typically takes 5-10 minutes on first build." -ForegroundColor Gray
+    substep "This typically takes 5-10 minutes on first build."
     Write-Host ""
 
     # Start total build timer
@@ -1692,19 +1828,19 @@ if (-not $NeedLlamaSourceBuild) {
     if (Test-Path (Join-Path $LlamaCppDir ".git")) {
         Write-Host "   Syncing llama.cpp to $ResolvedLlamaTag..." -ForegroundColor Gray
         if ($UseConcreteRef) {
-            git -C $LlamaCppDir fetch --depth 1 origin $ResolvedLlamaTag 2>&1 | Out-Null
+            $gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin $ResolvedLlamaTag }
         } else {
-            git -C $LlamaCppDir fetch --depth 1 origin 2>&1 | Out-Null
+            $gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin }
         }
-        if ($LASTEXITCODE -ne 0) {
-            Write-Host "   [WARN] git fetch failed -- using existing source" -ForegroundColor Yellow
+        if ($gitFetchExit -ne 0) {
+            substep "git fetch failed -- using existing source" "Yellow"
         } else {
-            git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD 2>&1 | Out-Null
-            if ($LASTEXITCODE -ne 0) {
+            $gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD }
+            if ($gitCheckoutExit -ne 0) {
                 $BuildOk = $false
                 $FailedStep = "git checkout"
             } else {
-                git -C $LlamaCppDir clean -fdx 2>&1 | Out-Null
+                Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
             }
         }
     } else {
@@ -1716,8 +1852,8 @@ if (-not $NeedLlamaSourceBuild) {
             $cloneArgs += @("--branch", $ResolvedLlamaTag)
         }
         $cloneArgs += @("https://github.com/ggml-org/llama.cpp.git", $buildTmp)
-        git @cloneArgs 2>&1 | Out-Null
-        if ($LASTEXITCODE -ne 0) {
+        $cloneExit = Invoke-SetupCommand -AlwaysQuiet { git @cloneArgs }
+        if ($cloneExit -ne 0) {
             $BuildOk = $false
             $FailedStep = "git clone"
             if (Test-Path $buildTmp) { Remove-Item -Recurse -Force $buildTmp }
@@ -1775,8 +1911,8 @@ if (-not $NeedLlamaSourceBuild) {
                     $maxArch = Get-NvccMaxArch -NvccExe $NvccPath
                     if ($maxArch) {
                         $CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$maxArch"
-                        Write-Host "   [WARN] GPU is sm_$CudaArch but nvcc only supports up to sm_$maxArch" -ForegroundColor Yellow
-                        Write-Host "          Building with sm_$maxArch (PTX will JIT for your GPU at runtime)" -ForegroundColor Yellow
+                        substep "GPU is sm_$CudaArch but nvcc only supports up to sm_$maxArch" "Yellow"
+                        substep "Building with sm_$maxArch (PTX will JIT for your GPU at runtime)" "Yellow"
                     }
                     # else: omit flag entirely, let cmake pick defaults
                 }
@@ -1786,10 +1922,11 @@ if (-not $NeedLlamaSourceBuild) {
         }
 
         $cmakeOutput = cmake @CmakeArgs 2>&1 | Out-String
-        if ($LASTEXITCODE -ne 0) {
+        $cmakeConfigureExit = $LASTEXITCODE
+        if ($cmakeConfigureExit -ne 0) {
             $BuildOk = $false
             $FailedStep = "cmake configure"
-            Write-Host $cmakeOutput -ForegroundColor Red
+            Write-LlamaFailureLog -Output $cmakeOutput
             if ($cmakeOutput -match 'No CUDA toolset found|CUDA_TOOLKIT_ROOT_DIR|nvcc') {
                 Write-Host ""
                 Write-Host "   Hint: CUDA VS integration may be missing. Try running as admin:" -ForegroundColor Yellow
@@ -1812,10 +1949,11 @@ if (-not $NeedLlamaSourceBuild) {
         Write-Host ""
 
         $output = cmake --build $BuildDir --config Release --target llama-server -j $NumCpu 2>&1 | Out-String
-        if ($LASTEXITCODE -ne 0) {
+        $cmakeBuildServerExit = $LASTEXITCODE
+        if ($cmakeBuildServerExit -ne 0) {
             $BuildOk = $false
             $FailedStep = "cmake build (llama-server)"
-            Write-Host $output -ForegroundColor Red
+            Write-LlamaFailureLog -Output $output
         }
     }
 
@@ -1824,9 +1962,10 @@ if (-not $NeedLlamaSourceBuild) {
         Write-Host ""
         Write-Host "--- cmake build (llama-quantize) ---" -ForegroundColor Cyan
         $output = cmake --build $BuildDir --config Release --target llama-quantize -j $NumCpu 2>&1 | Out-String
-        if ($LASTEXITCODE -ne 0) {
-            Write-Host "   [WARN] llama-quantize build failed (GGUF export may be unavailable)" -ForegroundColor Yellow
-            Write-Host $output -ForegroundColor Yellow
+        $cmakeBuildQuantizeExit = $LASTEXITCODE
+        if ($cmakeBuildQuantizeExit -ne 0) {
+            substep "llama-quantize build failed (GGUF export may be unavailable)" "Yellow"
+            Write-LlamaFailureLog -Output $output
         }
     }
 
@@ -1867,9 +2006,9 @@ if (-not $NeedLlamaSourceBuild) {
             step "llama.cpp" "built"
             step "build time" "${totalMin}m ${totalSec}s" "DarkGray"
         } else {
-            step "llama.cpp" "build failed at: $FailedStep (${totalMin}m ${totalSec}s)" "Red"
+            step "llama.cpp" "build failed at: $FailedStep (${totalMin}m ${totalSec}s); continuing" "Yellow"
             substep "To retry: delete $LlamaCppDir and re-run setup." "Yellow"
-            exit 1
+            $script:LlamaCppDegraded = $true
         }
     }
 }
@@ -1877,14 +2016,31 @@ if (-not $NeedLlamaSourceBuild) {
 # ─────────────────────────────────────────────
 # Footer
 # ─────────────────────────────────────────────
+$DoneLabel = if ($env:SKIP_STUDIO_BASE -eq "1") { "Unsloth Studio Setup Complete" } else { "Unsloth Studio Updated" }
 if ($script:StudioVtOk -and -not $env:NO_COLOR) {
     Write-Host ("  {0}{1}{2}" -f (Get-StudioAnsi Dim), $Rule, (Get-StudioAnsi Reset))
-    Write-Host ("  " + (Get-StudioAnsi Title) + "Unsloth Studio Installed" + (Get-StudioAnsi Reset))
+    if ($script:LlamaCppDegraded) {
+        Write-Host ("  " + (Get-StudioAnsi Warn) + "$DoneLabel (limited: llama.cpp unavailable)" + (Get-StudioAnsi Reset))
+    } else {
+        Write-Host ("  " + (Get-StudioAnsi Title) + $DoneLabel + (Get-StudioAnsi Reset))
+    }
     Write-Host ("  {0}{1}{2}" -f (Get-StudioAnsi Dim), $Rule, (Get-StudioAnsi Reset))
 } else {
     Write-Host "  $Rule" -ForegroundColor DarkGray
-    Write-Host "  Unsloth Studio Installed" -ForegroundColor Green
+    if ($script:LlamaCppDegraded) {
+        Write-Host "  $DoneLabel (limited: llama.cpp unavailable)" -ForegroundColor Yellow
+    } else {
+        Write-Host "  $DoneLabel" -ForegroundColor Green
+    }
     Write-Host "  $Rule" -ForegroundColor DarkGray
 }
 step "launch" "unsloth studio -H 0.0.0.0 -p 8888"
 Write-Host ""
+
+# Match studio/setup.sh: exit non-zero for degraded llama.cpp when called
+# from install.ps1 (SKIP_STUDIO_BASE=1) so the installer can detect the
+# failure. Direct 'unsloth studio update' does not set SKIP_STUDIO_BASE,
+# so it keeps degraded installs successful.
+if ($script:LlamaCppDegraded -and $env:SKIP_STUDIO_BASE -eq "1") {
+    exit 1
+}

studio/setup.sh

@@ -28,12 +28,43 @@ fi
 step()    { printf "  ${C_DIM}%-15.15s${C_RST}${3:-$C_OK}%s${C_RST}\n" "$1" "$2"; }
 substep() { printf "  ${C_DIM}%-15s%s${C_RST}\n" "" "$1"; }
 
+_is_verbose() {
+    [ "${UNSLOTH_VERBOSE:-0}" = "1" ]
+}
+
+verbose_substep() {
+    if _is_verbose; then
+        substep "$1"
+    fi
+    return 0
+}
+
+run_maybe_quiet() {
+    if _is_verbose; then
+        "$@"
+    else
+        "$@" > /dev/null 2>&1
+    fi
+}
+
 # ── Helper: run command quietly, show output only on failure ──
 _run_quiet() {
     local on_fail=$1
     local label=$2
     shift 2
 
+    if _is_verbose; then
+        local exit_code
+        "$@" && return 0
+        exit_code=$?
+        step "error" "$label failed (exit code $exit_code)" "$C_ERR" >&2
+        if [ "$on_fail" = "exit" ]; then
+            exit "$exit_code"
+        else
+            return "$exit_code"
+        fi
+    fi
+
     local tmplog
     tmplog=$(mktemp) || {
         step "error" "Failed to create temporary file" "$C_ERR" >&2
@@ -65,11 +96,18 @@ run_quiet_no_exit() {
     _run_quiet return "$@"
 }
 
+print_llama_error_log() {
+    local log_file=$1
+    [ -s "$log_file" ] || return 0
+    substep "llama.cpp diagnostics (last 120 lines):"
+    tail -n 120 "$log_file" | sed 's/^/   | /' >&2
+}
+
 # ── Banner ──
 echo ""
 printf "  ${C_TITLE}%s${C_RST}\n" "🦥 Unsloth Studio Setup"
 printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
-
+verbose_substep "verbose diagnostics enabled"
 # ── Clean up stale caches ──
 rm -rf "$REPO_ROOT/unsloth_compiled_cache"
 rm -rf "$SCRIPT_DIR/backend/unsloth_compiled_cache"
@@ -97,6 +135,7 @@ fi
 
 if [ "$_NEED_FRONTEND_BUILD" = false ]; then
     step "frontend" "up to date"
+    verbose_substep "frontend dist is newer than source inputs"
 else
 
 # ── Node ──
@@ -117,7 +156,7 @@ if command -v node &>/dev/null && command -v npm &>/dev/null; then
             # In Colab, just upgrade npm directly - nvm doesn't work well
             if [ "$NPM_MAJOR" -lt 11 ]; then
                 substep "upgrading npm..."
-                npm install -g npm@latest > /dev/null 2>&1
+                run_maybe_quiet npm install -g npm@latest
             fi
             NEED_NODE=false
         fi
@@ -127,7 +166,11 @@ fi
 if [ "$NEED_NODE" = true ]; then
     substep "installing nvm..."
     export NODE_OPTIONS=--dns-result-order=ipv4first
-    curl -so- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash > /dev/null 2>&1
+    if _is_verbose; then
+        curl -so- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash
+    else
+        curl -so- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash > /dev/null 2>&1
+    fi
 
     export NVM_DIR="$HOME/.nvm"
     set +u
@@ -141,7 +184,11 @@ if [ "$NEED_NODE" = true ]; then
 
     substep "installing Node LTS..."
     run_quiet "nvm install" nvm install --lts
-    nvm use --lts > /dev/null 2>&1
+    if _is_verbose; then
+        nvm use --lts
+    else
+        nvm use --lts > /dev/null 2>&1
+    fi
     set -u
 
     NODE_MAJOR=$(node -v | sed 's/v//' | cut -d. -f1)
@@ -158,13 +205,14 @@ if [ "$NEED_NODE" = true ]; then
 fi
 
 step "node" "$(node -v) | npm $(npm -v)"
+verbose_substep "node check: NEED_NODE=$NEED_NODE NODE_OK=${NODE_OK:-unknown} NPM_MAJOR=${NPM_MAJOR:-unknown}"
 
 # ── Install bun (optional, faster package installs) ──
 # Uses npm to install bun globally -- Node is already guaranteed above,
 # avoids platform-specific installers, PATH issues, and admin requirements.
 if ! command -v bun &>/dev/null; then
     substep "installing bun..."
-    if npm install -g bun > /dev/null 2>&1 && command -v bun &>/dev/null; then
+    if run_maybe_quiet npm install -g bun && command -v bun &>/dev/null; then
         substep "bun installed ($(bun --version))"
     else
         substep "bun install skipped (npm will be used instead)"
@@ -209,7 +257,10 @@ _try_bun_install() {
     _log=$(mktemp)
     bun install >"$_log" 2>&1 || _exit_code=$?
 
-    if [ "$_exit_code" -eq 0 ] && [ -x node_modules/.bin/tsc ] && [ -x node_modules/.bin/vite ]; then
+    # bun may create .exe shims on Windows (Git Bash / MSYS2) instead of plain scripts
+    if [ "$_exit_code" -eq 0 ] \
+        && { [ -x node_modules/.bin/tsc ] || [ -f node_modules/.bin/tsc.exe ] || [ -f node_modules/.bin/tsc.bunx ]; } \
+        && { [ -x node_modules/.bin/vite ] || [ -f node_modules/.bin/vite.exe ] || [ -f node_modules/.bin/vite.bunx ]; }; then
         rm -f "$_log"
         return 0
     fi
@@ -228,21 +279,25 @@ _try_bun_install() {
 
 _bun_install_ok=false
 if command -v bun &>/dev/null; then
-    echo "   Using bun for package install (faster)"
+    substep "using bun for package install (faster)"
     if _try_bun_install; then
         _bun_install_ok=true
     else
         # First attempt failed, likely due to corrupt cache entries.
         # Clear the cache and retry once.
         echo "   Clearing bun cache and retrying..."
-        bun pm cache rm > /dev/null 2>&1 || true
+        run_maybe_quiet bun pm cache rm || true
         if _try_bun_install; then
             _bun_install_ok=true
         fi
     fi
 fi
 if [ "$_bun_install_ok" = false ]; then
-    run_quiet "npm install" npm install
+    run_quiet_no_exit "npm install" npm install --no-fund --no-audit --loglevel=error
+    _npm_install_rc=$?
+    if [ "$_npm_install_rc" -ne 0 ]; then
+        exit "$_npm_install_rc"
+    fi
 fi
 run_quiet "npm run build" npm run build
 
@@ -265,7 +320,11 @@ fi  # end frontend build check
 # ── oxc-validator runtime ──
 if [ -d "$SCRIPT_DIR/backend/core/data_recipe/oxc-validator" ] && command -v npm &>/dev/null; then
     cd "$SCRIPT_DIR/backend/core/data_recipe/oxc-validator"
-    run_quiet "npm install (oxc validator runtime)" npm install
+    run_quiet_no_exit "npm install (oxc validator runtime)" npm install --no-fund --no-audit --loglevel=error
+    _oxc_install_rc=$?
+    if [ "$_oxc_install_rc" -ne 0 ]; then
+        exit "$_oxc_install_rc"
+    fi
     cd "$SCRIPT_DIR"
 fi
 
@@ -287,9 +346,19 @@ if [ ! -x "$VENV_DIR/bin/python" ]; then
         # packages (huggingface-hub, datasets, transformers) and only pulls
         # in genuinely missing ones (structlog, fastapi, etc.).
         substep "Colab detected, installing Studio backend dependencies..."
+        _COLAB_REQS_TMP="$(mktemp)"
         sed 's/[><=!~;].*//' "$SCRIPT_DIR/backend/requirements/studio.txt" \
-            | grep -v '^#' | grep -v '^$' \
-            | pip install -q -r /dev/stdin 2>/dev/null || true
+            | grep -v '^#' | grep -v '^$' > "$_COLAB_REQS_TMP"
+        if [ -s "$_COLAB_REQS_TMP" ]; then
+            if ! run_quiet_no_exit "install Colab backend deps" pip install -q -r "$_COLAB_REQS_TMP"; then
+                rm -f "$_COLAB_REQS_TMP"
+                step "python" "Colab backend dependency install failed" "$C_ERR"
+                exit 1
+            fi
+        else
+            step "python" "no Colab backend dependencies resolved from requirements file" "$C_WARN"
+        fi
+        rm -f "$_COLAB_REQS_TMP"
         _COLAB_NO_VENV=true
     else
         step "python" "venv not found at $VENV_DIR" "$C_ERR"
@@ -308,7 +377,13 @@ install_python_stack() {
 USE_UV=false
 if command -v uv &>/dev/null; then
     USE_UV=true
-elif curl -LsSf https://astral.sh/uv/install.sh | sh > /dev/null 2>&1; then
+elif {
+    if _is_verbose; then
+        curl -LsSf https://astral.sh/uv/install.sh | sh
+    else
+        curl -LsSf https://astral.sh/uv/install.sh | sh > /dev/null 2>&1
+    fi
+}; then
     export PATH="$HOME/.local/bin:$PATH"
     command -v uv &>/dev/null && USE_UV=true
 fi
@@ -325,7 +400,8 @@ cd "$SCRIPT_DIR"
 # On Colab without a venv, skip venv-dependent Python deps sections but
 # continue to llama.cpp install so GGUF inference is available.
 if [ "$_COLAB_NO_VENV" = true ]; then
-    echo "✅ Studio backend dependencies installed into system Python"
+    step "python" "backend deps installed into system Python"
+    substep "continuing to llama.cpp install for GGUF inference support"
 fi
 
 # ── Check if Python deps need updating ──
@@ -375,6 +451,7 @@ if [ "$_SKIP_PYTHON_DEPS" = false ]; then
     step "transformers" "5.x pre-installed"
 else
     step "python" "dependencies up to date"
+    verbose_substep "python deps check: installed=$_PKG_NAME@${INSTALLED_VER:-unknown} latest=${LATEST_VER:-unknown}"
 fi
 
 # ── 7. Prefer prebuilt llama.cpp bundles before any source build path ──
@@ -383,6 +460,7 @@ mkdir -p "$UNSLOTH_HOME"
 LLAMA_CPP_DIR="$UNSLOTH_HOME/llama.cpp"
 LLAMA_SERVER_BIN="$LLAMA_CPP_DIR/build/bin/llama-server"
 _NEED_LLAMA_SOURCE_BUILD=false
+_LLAMA_CPP_DEGRADED=false
 _LLAMA_FORCE_COMPILE="${UNSLOTH_LLAMA_FORCE_COMPILE:-0}"
 _REQUESTED_LLAMA_TAG="${UNSLOTH_LLAMA_TAG:-latest}"
 _HELPER_RELEASE_REPO="${UNSLOTH_LLAMA_RELEASE_REPO:-unslothai/llama.cpp}"
@@ -400,7 +478,7 @@ else
 fi
 if [ -z "$_RESOLVED_LLAMA_TAG" ]; then
     step "llama.cpp" "failed to resolve prebuilt tag via $_HELPER_RELEASE_REPO" "$C_WARN"
-    cat "$_RESOLVE_LLAMA_LOG" >&2 || true
+    print_llama_error_log "$_RESOLVE_LLAMA_LOG"
     set +e
     # Resolve the llama.cpp tag for source-build fallback. Pass --published-repo
     # so the resolver prefers Unsloth's tested tag (e.g. b8508) over the upstream
@@ -426,6 +504,7 @@ fi
 rm -f "$_RESOLVE_LLAMA_LOG"
 
 substep "resolved llama.cpp tag: $_RESOLVED_LLAMA_TAG"
+verbose_substep "requested llama.cpp tag: $_REQUESTED_LLAMA_TAG (repo: $_HELPER_RELEASE_REPO)"
 
 if [ "$_LLAMA_FORCE_COMPILE" = "1" ]; then
     step "llama.cpp" "UNSLOTH_LLAMA_FORCE_COMPILE=1 -- skipping prebuilt" "$C_WARN"
@@ -447,14 +526,25 @@ else
         if [ -n "${UNSLOTH_LLAMA_RELEASE_TAG:-}" ]; then
             _PREBUILT_CMD+=(--published-release-tag "$UNSLOTH_LLAMA_RELEASE_TAG")
         fi
+        _PREBUILT_LOG="$(mktemp)"
         set +e
-        "${_PREBUILT_CMD[@]}"
-        _PREBUILT_STATUS=$?
+        if _is_verbose; then
+            "${_PREBUILT_CMD[@]}" 2>&1 | tee "$_PREBUILT_LOG"
+            _PREBUILT_STATUS=${PIPESTATUS[0]}
+        else
+            "${_PREBUILT_CMD[@]}" >"$_PREBUILT_LOG" 2>&1
+            _PREBUILT_STATUS=$?
+        fi
         set -e
 
         if [ "$_PREBUILT_STATUS" -eq 0 ]; then
             step "llama.cpp" "prebuilt installed and validated"
+            verbose_substep "llama.cpp install dir: $LLAMA_CPP_DIR"
+            rm -f "$_PREBUILT_LOG"
         else
+            step "llama.cpp" "prebuilt install failed (continuing)" "$C_WARN"
+            print_llama_error_log "$_PREBUILT_LOG"
+            rm -f "$_PREBUILT_LOG"
             if [ -d "$LLAMA_CPP_DIR" ]; then
                 substep "prebuilt update failed; existing install restored"
             fi
@@ -523,12 +613,15 @@ if [ "$_NEED_LLAMA_SOURCE_BUILD" = false ]; then
     :
 elif [ "${_SKIP_GGUF_BUILD:-}" = true ]; then
     step "llama.cpp" "skipped (missing build deps)" "$C_WARN"
+    [ -f "$LLAMA_SERVER_BIN" ] || _LLAMA_CPP_DEGRADED=true
 else
 {
     if ! command -v cmake &>/dev/null; then
         step "llama.cpp" "skipped (cmake not found)" "$C_WARN"
+        [ -f "$LLAMA_SERVER_BIN" ] || _LLAMA_CPP_DEGRADED=true
     elif ! command -v git &>/dev/null; then
         step "llama.cpp" "skipped (git not found)" "$C_WARN"
+        [ -f "$LLAMA_SERVER_BIN" ] || _LLAMA_CPP_DEGRADED=true
     else
         BUILD_OK=true
         _CLONE_BRANCH_ARGS=()
@@ -691,8 +784,10 @@ else
             [ -f "$LLAMA_CPP_DIR/llama-quantize" ] && step "llama-quantize" "built"
         elif [ "$BUILD_OK" = true ]; then
             step "llama.cpp" "binary not found after build" "$C_WARN"
+            _LLAMA_CPP_DEGRADED=true
         else
             step "llama.cpp" "build failed" "$C_ERR"
+            [ -f "$LLAMA_SERVER_BIN" ] || _LLAMA_CPP_DEGRADED=true
         fi
     fi
 }
@@ -702,14 +797,35 @@ fi  # end _SKIP_GGUF_BUILD check
 if [ "$IS_COLAB" = true ]; then
     echo ""
     printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
-    printf "  ${C_TITLE}%s${C_RST}\n" "Unsloth Studio Setup Complete"
+    if [ "$_LLAMA_CPP_DEGRADED" = true ]; then
+        printf "  ${C_WARN}%s${C_RST}\n" "Unsloth Studio Setup Complete (limited: llama.cpp unavailable)"
+    else
+        printf "  ${C_TITLE}%s${C_RST}\n" "Unsloth Studio Setup Complete"
+    fi
     printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
     substep "from colab import start"
     substep "start()"
 else
     printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
-    printf "  ${C_TITLE}%s${C_RST}\n" "Unsloth Studio Installed"
+    if [ "$_LLAMA_CPP_DEGRADED" = true ]; then
+        printf "  ${C_WARN}%s${C_RST}\n" "Unsloth Studio Installed (limited: llama.cpp unavailable)"
+    else
+        printf "  ${C_TITLE}%s${C_RST}\n" "Unsloth Studio Installed"
+    fi
     printf "  ${C_DIM}%s${C_RST}\n" "$RULE"
-    printf "  ${C_DIM}%-15s${C_OK}%s${C_RST}\n" "launch" "unsloth studio -H 0.0.0.0 -p 8888"
+    if [ "$_LLAMA_CPP_DEGRADED" = true ]; then
+        printf "  ${C_DIM}%-15s${C_WARN}%s${C_RST}\n" "launch" "unsloth studio -H 0.0.0.0 -p 8888"
+    else
+        printf "  ${C_DIM}%-15s${C_OK}%s${C_RST}\n" "launch" "unsloth studio -H 0.0.0.0 -p 8888"
+    fi
 fi
 echo ""
+
+# When called from install.sh (SKIP_STUDIO_BASE=1), exit non-zero so the
+# installer can report the GGUF failure after finishing PATH/shortcut setup.
+# When called directly via 'unsloth studio update', keep the install
+# successful -- the footer above already reports the limitation and Studio
+# is still usable for non-GGUF workflows.
+if [ "$_LLAMA_CPP_DEGRADED" = true ] && [ "${SKIP_STUDIO_BASE:-0}" = "1" ]; then
+    exit 1
+fi

unsloth_cli/commands/studio.py

@@ -267,10 +267,7 @@ def setup(
         help = "Full pip/build output during setup for troubleshooting.",
     ),
 ):
-    """Deprecated: use 'unsloth studio update' or re-run install.sh."""
-    typer.echo(
-        "Note: 'unsloth studio setup' is deprecated. Use 'unsloth studio update' or re-run install.sh."
-    )
+    """Run Studio setup (called by install.ps1 / install.sh)."""
     _run_setup_script(verbose = verbose)
 
 
@@ -290,13 +287,18 @@ def update(
     ),
 ):
     """Update Unsloth Studio dependencies and rebuild."""
-    os.environ["STUDIO_LOCAL_INSTALL"] = "1" if local else "0"
+    # Ensure SKIP_STUDIO_BASE is not inherited from a parent install.ps1 session
+    os.environ.pop("SKIP_STUDIO_BASE", None)
     os.environ["STUDIO_PACKAGE_NAME"] = package
     if local:
+        os.environ["STUDIO_LOCAL_INSTALL"] = "1"
         # Pass the repo root explicitly so install_python_stack.py doesn't
         # have to guess from SCRIPT_DIR (which may be inside site-packages).
         repo_root = Path(__file__).resolve().parents[2]
         os.environ["STUDIO_LOCAL_REPO"] = str(repo_root)
+    else:
+        os.environ["STUDIO_LOCAL_INSTALL"] = "0"
+        os.environ.pop("STUDIO_LOCAL_REPO", None)
     _run_setup_script(verbose = verbose)