unsloth/studio/setup.ps1
Leo Borcherding b6d5636cc0
fix/strix halo and windows AMD ROCm support (#5301)
* fix(studio): set HIP_VISIBLE_DEVICES in apply_gpu_ids for ROCm training workers

Training workers are spawned via multiprocessing spawn before detect_hardware()
runs, so IS_ROCM is still False. If the user never set HIP_VISIBLE_DEVICES in
their shell, _inherits_rocm_visibility is also False, leaving the worker with
only CUDA_VISIBLE_DEVICES set. On ROCm hosts the HIP runtime honors
HIP_VISIBLE_DEVICES over CUDA_VISIBLE_DEVICES, so the worker saw the full
device list and torch raised "no usable HIP accelerator" on some setups.

Fall back to probing torch.version.hip (a build-time attribute, safe to read
before GPU init) to detect ROCm when neither IS_ROCM nor inherited env vars
are available. Mirrors the existing fix in llama_cpp.py for llama-server
subprocess GPU pinning.

Fixes https://github.com/unslothai/unsloth/issues/5180

* test: tighten apply_gpu_ids ROCm fallback assertions

Replace loose OR chain with exact string matches, split into three
focused tests, and add a guard check for the try/except wrapper.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: detect ROCm unified memory (Strix Halo / AMD iGPU) via torch fallback

amd-smi on iGPUs with shared/unified memory (e.g. Radeon 8060S on Strix
Halo) reports only the dedicated VRAM slice (~512 MB) in its metric output,
so get_visible_gpu_utilization() was returning usable_gb ≈ 0.35 GB instead
of the full GTT pool (~128 GB).  torch.cuda.mem_get_info() already surfaces
the correct unified-pool size.

Add _reconcile_rocm_unified_memory(): after amd-smi returns a valid result
on a ROCm device, cross-check each device's vram_total_gb against
torch.cuda.mem_get_info().  When torch reports a larger total, replace the
amd-smi VRAM fields in-place.  No-op for discrete AMD GPUs where the two
sources agree.

Fixes: "Falling back to all visible GPUs -- model may not fit" on AMD iGPU
machines even when 100+ GB of unified memory is available.

* Apply unified-memory reconciliation in get_gpu_utilization too

The visible-GPU path was already corrected for AMD iGPUs with unified memory
(Strix Halo / Radeon 8060S), but get_gpu_utilization was still returning the
raw 512 MB amd-smi VRAM slice. Studio's /api/train/hardware endpoint and the
live GPU monitor read from this primary path, so users continued seeing the
wrong total even after auto_select_gpu_ids picked the right device.

Refactor to share the per-device correction:
  * _apply_unified_memory_correction(metrics, torch_info) -- the actual
    replacement logic, in-place on a single metrics dict.
  * _reconcile_rocm_unified_memory(...)                   -- multi-device,
    iterates utilization["devices"] (visible-GPU path).
  * _reconcile_primary_rocm_unified_memory(...)           -- single flat
    metrics dict (primary-GPU path), uses parent_visible_spec to pick the
    primary index, falls back to ordinal 0 when no visibility env is set.

get_gpu_utilization now calls the primary reconciler under IS_ROCM, so both
endpoints surface the real unified-memory pool on iGPUs while leaving
discrete AMD GPUs untouched (torch_total <= smi_total -> no replace).

* Use 'is not None' and log debug on torch.version.hip probe failures

Two small follow-ups to the apply_gpu_ids ROCm fallback:

1. Match detect_hardware()'s 'getattr(torch.version, "hip", None) is not None'
   form so the entire codebase has one canonical 'this torch was built with
   HIP' check. On every shipping torch wheel hip is either None or a non-empty
   version string, so the new form agrees with the old bool() form on every
   real install.

2. Log the probe failure at debug level instead of swallowing it silently.
   The broad 'except Exception' is intentional (we never want apply_gpu_ids
   to crash a worker over a probe), but the silent pass made it impossible
   to tell whether the fallback was firing or being skipped.

* fix(studio): honour HIP_VISIBLE_DEVICES in _get_parent_visible_gpu_spec before IS_ROCM is set

When a user has HIP_VISIBLE_DEVICES set in their shell (e.g. "1" to select
GPU 1) but detect_hardware() has not yet run in the Studio parent process,
IS_ROCM is still False.  _get_parent_visible_gpu_spec() was gated on IS_ROCM
so it fell through to CUDA_VISIBLE_DEVICES (unset), saw all physical GPUs,
and auto-selected index 0.  apply_gpu_ids then overwrote HIP_VISIBLE_DEVICES
with "0", making the intended GPU invisible to ROCm torch in the worker,
which triggered the "no usable HIP accelerator" error (issue #5180).

Apply the same _inherits_rocm_visibility pattern already used in
apply_gpu_ids: check for HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES in the
environment regardless of IS_ROCM so the correct GPU index is preserved.

* fix(install): harden AMD ROCm GPU detection for multi-GPU and env-filtered setups

The previous rocminfo awk pattern could miss discrete GPUs on machines
where HIP_VISIBLE_DEVICES/ROCR_VISIBLE_DEVICES is used to mask an
integrated GPU — the env vars filter rocminfo output but may not
propagate into the install script subprocess, causing detection to
fail entirely.

Two changes:
- Tighten rocminfo pattern from /gfx[0-9]/ && !/gfx000/ to
  /gfx[1-9][0-9]/ — simpler and correctly excludes the CPU agent
  (gfx000) without a negative lookahead
- Add sysfs KFD topology fallback: reads
  /sys/class/kfd/kfd/topology/nodes/*/gpu_id which is a kernel-level
  view unaffected by HIP_VISIBLE_DEVICES or ROCR_VISIBLE_DEVICES

Fixes detection failure reported in Discord by Chains (gfx1201 + iGPU
machine where env var exclusion of the iGPU caused rocminfo to return
no usable device).

* Fix KFD sysfs awk fallback to read properties file

The fallback added by this PR reads /sys/class/kfd/kfd/topology/nodes/*/gpu_id
files but matches the literal token 'gpu_id' against their content. Those
files contain only a single decimal value (e.g. '0' for CPU agents, '50432'
for GPU agents), so the regex never matches and 'found' stays 0, making the
fallback a no-op on every host. The properties file in the same directory
contains key/value lines like 'gpu_id 50432' which is what the existing awk
pattern expects.

Reproduced with a synthetic sysfs layout: against gpu_id files awk exits 1;
against properties files awk exits 0 when any node reports gpu_id > 0.

* fix(setup.ps1): detect AMD ROCm GPU on Windows, bring to parity with setup.sh

setup.ps1 only checked nvidia-smi and fell straight to "gpu: none" on AMD
machines. setup.sh already probed rocminfo/amd-smi/hipconfig/hipinfo.

Add three-tier detection mirroring install_llama_prebuilt.py's detect_host():
1. hipinfo: gcnArchName in output confirms a real HIP GPU (not just SDK)
2. amd-smi list: "GPU: <digit>" data rows as fallback
3. WMI Win32_VideoController: last resort -- detects AMD GPU even without
   HIP SDK, then guides user to install it rather than silently going CPU

Also corrects the "none" message to mention AMD ROCm alongside NVIDIA so
users with AMD hardware understand the requirement.

Fixes: rohit-style install where Strix Halo (Radeon 8060S) showed
"gpu: none" even with the HIP SDK present.

* fix(install.ps1): detect AMD ROCm GPU on Windows, bring to parity with setup.ps1

install.ps1 had the same nvidia-smi-only GPU detection as setup.ps1 before
the setup.ps1 fix. Applies the same three-tier AMD detection:
1. hipinfo: gcnArchName confirms real HIP GPU
2. amd-smi list: GPU data rows as fallback
3. WMI Win32_VideoController: detects AMD GPU without HIP SDK and guides
   user to install it

Fixes: install.ps1 showing "gpu: none" while setup.ps1 correctly showed
"AMD GPU detected" on the same machine (reported by rohit, RX 7600 XT).

* fix(install.ps1): suppress 'No NVIDIA GPU detected' when AMD GPU is present

* feat: add Windows AMD ROCm PyTorch wheel installation

install_python_stack.py:
- Add _ROCM_WINDOWS_WHEEL_BASE and _ROCM_WINDOWS_RELEASES constants
  pointing to AMD repo.radeon.com (ROCm 7.2 -> torch 2.9.1+rocm7.2.1)
- Extend _ensure_rocm_torch() with a Windows branch: detects ROCm via
  _has_rocm_gpu() / _detect_rocm_version(), requires Python 3.12 (cp312
  is the only ABI AMD publishes for Windows), installs the direct wheel
  URL from repo.radeon.com

install.ps1:
- Capture ROCmVersion during AMD detection via hipconfig --version /
  amd-smi version (needed for wheel URL selection)
- After Get-TorchIndexUrl, add an AMD wheel override block: when HasROCm
  and Python 3.12 detected, set ROCmTorchWheelUrl to AMD wheel URL
- Expand torch install branch to handle ROCmTorchWheelUrl with
  uv pip install --force-reinstall --no-cache-dir

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: also install torchvision and torchaudio from AMD Windows repo

AMD publishes matching torchvision-0.24.1+rocm7.2.1 and
torchaudio-2.9.1+rocm7.2.1 cp312 wheels at the same repo.radeon.com
release folder. Install all three in both install.ps1 and
install_python_stack.py Windows ROCm path.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* feat: add ROCm 7.1.1 Windows wheel mapping

AMD uses a different version string for 7.1.1 wheels:
2.9.0+rocmsdk20251116 (date-tagged) instead of +rocm7.1.1.
Adds the 7.1.1 release folder to both install.ps1 and
install_python_stack.py so users with ROCm 7.1 get ROCm
torch instead of falling back to CPU.

* fix: install rocm_sdk_core and rocm_sdk_libraries_custom alongside torch

The AMD Windows torch wheels declare rocm[libraries]==<ver> as a hard
dependency. Without installing rocm_sdk_core and rocm_sdk_libraries_custom
from the same AMD release folder, uv cannot resolve the dependency and
fails with 'No solution found'. Include all 5 wheels in one install call.

* fix: expand ROCm wheel array to scalars for Invoke-InstallCommand

@array splatting inside a scriptblock only works when the native command
is prefixed with '&'. Invoke-InstallCommand uses '& $Command' to run the
block, so @ROCmAllWheelUrls was not being expanded. Extract to scalar
variables $rw0-$rw4 which are captured correctly by the closure.

* fix: use --no-deps for AMD Windows torch wheel install

uv's resolver looks up rocm[libraries]==0.1.dev0 on PyPI during
dependency resolution before downloading any wheels, and fails because
the package doesn't exist on PyPI. --no-deps skips resolution entirely
and installs all 5 AMD wheels directly. The GPU runtime dependency is
satisfied by the HIP SDK, not a Python package.

* fix: setup.ps1 and install_python_stack.py now install ROCm torch on Windows

setup.ps1 was always setting CuTag='cpu' for non-NVIDIA hosts and installing
cpu-only PyTorch, overwriting the ROCm torch installed by install.ps1.
Adds the same AMD wheel selection logic (ROCm version detection, Python 3.12
check, 5-wheel install with --no-deps) to setup.ps1's torch install block.

install_python_stack.py: remove IS_WINDOWS guard from _ensure_rocm_torch()
call site so the Windows path in _ensure_rocm_torch() is reachable during
'unsloth studio update' as well.

* fix: suppress manual-install warning when ROCm torch already present; fix progress counter

- Gate the 'must be installed manually' warning on torch.version.hip being empty
  so it doesn't fire when our ROCm torch install succeeded
- Update _TOTAL counter to include the 3 ROCm steps on Windows now that
  _ensure_rocm_torch() is called there (fixes 10/9 display)

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* feat: add rocm step display in setup.ps1; fix warning and progress counter

- Add 'rocm' step after 'cuda' in setup.ps1 showing ROCm version or HIP SDK missing
- Move ROCm version detection up to GPU detection block so it's available early
- Suppress 'must be installed manually' warning when torch.version.hip is set
- Fix _TOTAL counter to include ROCm steps on Windows (fixes 10/9 display)

* fix: detect AMD SDK ROCm torch via __version__ when torch.version.hip is unset

AMD's repo.radeon.com wheels (e.g. 2.9.0+rocmsdk20251116) do not set
torch.version.hip, leaving it None. All three probes that relied solely on
torch.version.hip now also check for 'rocm' in torch.__version__.lower():

- hardware.py detect_hardware(): IS_ROCM was never set, causing the studio
  to report 'Hardware detected: CPU' even after AMD wheels were installed
  and HIP DLLs were on PATH.
- install_python_stack.py _ensure_rocm_torch(): skip-if-already-installed
  probe would always reinstall on subsequent runs.
- install_python_stack.py Windows AMD warning: suppression check always
  failed, so the 'must be installed manually' note kept appearing after
  a successful AMD wheel install.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* perf: drop --no-cache-dir from AMD ROCm torch wheel installs

uv caches downloaded wheels by default; passing --no-cache-dir forced a
full redownload of the ~2 GB torch wheel on every install run. CUDA installs
never had this flag -- AMD was the only path affected.

* fix: use install-state flag instead of subprocess probe for AMD Windows warning

Replace the subprocess torch probe in the post-install warning block with a
module-level _rocm_windows_torch_installed flag set by _ensure_rocm_torch().
Subprocess re-import of torch is unnecessary and fragile -- the install
function already knows whether it succeeded.

* fix: hoist global declaration to top of _ensure_rocm_torch

Python requires the global statement to appear before any assignment
to the variable within a function. Moving it to the function top fixes
the SyntaxError on line 354.

* fix: pass AMD torch install status via env var to suppress false warning

setup.ps1 now sets UNSLOTH_ROCM_TORCH_INSTALLED=1 after a successful AMD
wheel install. install_python_stack.py reads this at the top of
_ensure_rocm_torch() to skip both the subprocess probe and the warning --
no re-import of torch needed, and the warning message now correctly says
'could not be auto-installed' rather than 'must be installed manually'.

* fix: register ROCm DLL directory before torch import on Windows

Python 3.8+ ignores PATH for extension DLL loading on Windows; amdhip64.dll
and other HIP runtime DLLs must be registered via os.add_dll_directory().
Without this, torch.cuda.is_available() always returns False on AMD ROCm
Windows even when HIP_PATH is correctly set in system environment variables.

Reads HIP_PATH / ROCM_PATH env vars first, then falls back to scanning
common ROCm install roots (C:\Program Files\AMD\ROCm, F:\ROCm, C:\ROCm).

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: remove hardcoded non-standard ROCm paths from DLL directory scan

Only use HIP_PATH/ROCM_PATH (set by AMD installer) and the standard
C:\Program Files\AMD\ROCm\<version>\bin location. Custom drive paths
like F:\ROCm are user-specific and should not be hardcoded.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: prevent torchao overrides step from overwriting AMD ROCm torch

torchao==0.14.0 in overrides.txt declares torch as a dependency. Without
--no-deps, uv resolves torch from PyPI and installs 2.11.0+cpu on top of
the AMD ROCm wheels (2.9.0+rocmsdk20251116). This was the root cause of
'Hardware detected: CPU' -- the AMD wheels were installed but then
immediately overwritten by the overrides step.

When _rocm_windows_torch_installed is True, add --no-deps to the overrides
pip_install call so torchao is installed without pulling in CPU torch.

* fix: add rocm_sdk namespace tarball to Windows ROCm wheel installs

torch/_rocm_init.py calls `import rocm_sdk` at startup, which requires
the rocm namespace tarball (rocm-*.tar.gz) in addition to the SDK wheel
packages. This tarball was missing from both install.ps1 and setup.ps1,
causing ModuleNotFoundError on first torch import.

- Add rocm-0.1.dev0.tar.gz to ROCm 7.1.1 install (provides rocm_sdk namespace)
- Add rocm-7.2.1.tar.gz + rocm_sdk_devel to ROCm 7.2.1 install
- Install tarball in a dedicated step before main SDK/torch wheels
- Switch to @array splatting in install.ps1 scriptblock for dynamic wheel count
- Remove --no-cache-dir from Python-side ROCm wheel install (prevents ~2GB redownload)

* feat: enable ROCm 7.2 torch install + warn on gfx1151 with ROCm < 7.2

Chigoma333 (AMD Radeon 8060S / gfx1151, Strix Halo) confirmed that ROCm
7.1 segfaults when tensors are moved to GPU, but ROCm 7.2 + torch
2.11.0+rocm7.2 works fully including training.

Changes:
- Uncomment (7,2): "rocm7.2" in _ROCM_TORCH_INDEX (was blocked by <2.11.0)
- Add _ROCM_TORCH_PKG_SPECS dict with per-tag version bounds:
  rocm7.2 → torch>=2.11.0,<2.12.0; all older tags → <2.11.0
- Add _detect_amd_gfx_codes() helper that parses rocminfo output
- Warn on gfx1151/gfx1150 (Strix Halo) when ROCm < 7.2 is installed,
  pointing users at the known segfault and recommending upgrade
- install.sh get_torch_index_url(): enable rocm7.2 case (previously capped
  to rocm7.1), cap unknown future tags to rocm7.2
- install.sh: override TORCH_CONSTRAINT to >=2.11.0,<2.12.0 when rocm7.2
  index is selected, so pip can actually resolve torch 2.11.0

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: prefer Python 3.12 for AMD ROCm users when 3.13 is also installed

After GPU detection, if ROCm HIP SDK is found and the selected Python
is not 3.12, run a second pass to locate a 3.12 install via py.exe and
PATH (catches uv-managed installs). Switch $DetectedPython to 3.12 so
the venv is created with a compatible interpreter for the cp312-only AMD
Windows torch wheels.

NVIDIA and Intel GPU paths are unaffected -- the re-detection block only
runs when $HasROCm is true.

Fixes: #5301

* fix: also check uv-managed Python 3.12 for AMD ROCm #5301

* fix: hide amd-smi console popups on Windows, guard torch.distributed.is_initialized for ROCm #5301

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: suppress remaining console popups on Windows, patch torch.distributed.is_initialized for ROCm #5301

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: stub all missing torch.distributed attrs for ROCm Windows wheel #5301

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: inject torch.distributed stub when C backend missing in ROCm Windows wheel #5301

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(rocm/windows): pre-stub torch._C._distributed_c10d + raise amd-smi timeout

Two fixes for Windows ROCm regressions reported by electroglyph on #5301:

1. worker.py — torch.distributed stub now fires unconditionally on Windows
   The previous stub only injected sys.modules in the except branch, meaning
   it was silently skipped when `import torch.distributed` happened to succeed
   (the C backend is lazily resolved).  The crash then hit later when
   transformers/trl triggered the lazy load.  Fix: on win32 we pre-populate
   sys.modules['torch._C._distributed_c10d'] AND set the attribute on the
   torch._C extension module *before* attempting the import, covering both
   the early-ImportError and lazy-load failure modes.

2. amd.py — increase amd-smi timeout from 5 s to 30 s on Windows (10 s Linux)
   amd-smi on Windows must cold-init the ROCm runtime on first invocation;
   5 s was consistently too short, producing repeated 'Command timed out'
   warnings in the server log.  30 s gives enough headroom without blocking
   indefinitely on broken installs.

3. install.ps1 — widen Python 3.12 enforcement to ROCmGpuLabel (WMI-only path)
   Users whose HIP SDK is not on PATH were detected via WMI but not switched
   to Python 3.12 before the install started, causing a second pass.  Guard
   now fires on (HasROCm -or ROCmGpuLabel).

* fix(rocm): guard c10d stub, fix TorchIndexFamily for 7.1, clean dead code + comments

- worker.py: wrap c10d stub injection in `if _c10d_key not in sys.modules` so
  Windows NVIDIA users with a real torch.distributed are never affected
- install.ps1: fix Get-TauriTorchIndexFamily receiving hardcoded "rocm7.2"
  even when ROCm 7.1 wheels are installed; now branches on $ROCmVersion
- main.py: remove dead `import ctypes as _ctypes` (ctypes is never called)
- hardware.py, install_python_stack.py, worker.py, install.ps1: shorten
  verbose multi-line comment blocks throughout
- tests: update 4 stale assertions that expected rocm7.2 to be absent/capped

* fix(tests): match windows AMD warning assertion to actual source string

* chore: trim verbose comment blocks across all ROCm-related files

* fix: guard reconcile call against None numeric_ids; add torchvision lower bounds

* fix(install.ps1): recreate venv with Python 3.12 after ROCm switch

Venv was created with 3.13 before GPU detection ran; switching
$DetectedPython to 3.12 had no effect since $VenvPython still
pointed to the 3.13 interpreter inside the already-created venv.

* ux: detect AMD GPU before Python selection to avoid double venv creation

- Early hipinfo + WMI probe runs before Find-CompatiblePython so Python
  3.12 is selected upfront when AMD is detected; venv is now created
  exactly once instead of 3.13 then immediately 3.12.
- Post-venv recreation block replaced with a simple warning for the rare
  case where AMD was missed by the early probe.
- setup.ps1: show venv's actual Python version (e.g. 3.12) instead of
  the system Python found by the pre-activation search (was showing 3.13).

* fix(rocm/win): auto-stub all _distributed_c10d symbols via PEP-562 __getattr__

The bare ModuleType stub caused ImportError when torch._dynamo was imported
(triggered by trainer.py accessing torch._dynamo.config at load time).
torch._dynamo pulls in torch.distributed.fsdp._flat_param which does:
  from torch._C._distributed_c10d import FakeProcessGroup
and potentially other symbols. Adding module __getattr__ auto-creates a
stub class for any missing symbol so all such imports succeed without
enumerating every individual symbol. Applied to both the primary stub
and the fallback stub in the except branch.

* chore: trim c10d stub comment

* fix(rocm/win): auto-stub missing torch.distributed attrs (Store, ProcessGroup, …)

* fix(rocm/win): pre-stub fsdp submodules in sys.modules; fix __getattr__ subpackage clash

* feat(rocm/win): arch-aware wheel selector always picks newest ROCm release

Replace HIP-SDK-version-gated wheel selection with GPU arch-based logic.
Select-ROCmWheelRelease (PS) and _select_windows_rocm_release (Python) map
gcnArchName → minimum ROCm version, then pick the newest available release
that satisfies it (currently always rocm-rel-7.2.1 for any supported GPU).
Wheels bundle their own ROCm runtime so the installed HIP SDK 7.1 does not
prevent using 7.2.1 wheels on gfx1200 (RX 9060 XT) and similar RDNA 4 GPUs.

Also installs the bitsandbytes Windows ROCm continuous-release wheel and sets
BNB_ROCM_VERSION=72 in worker.py before ML imports so bnb loads the
libbitsandbytes_rocm72.dll that ships in that wheel.

* fix(rocm/win): stub class metaclass for ProcessGroup.BackendType; amd-smi circuit breaker

torchao.float8.inference accesses ProcessGroup.BackendType as a class-level
attribute.  Plain type() stubs have no __getattr__ on the metaclass so this
raises AttributeError.  Introduce _StubClassMeta whose __getattr__ returns
child stub classes, fixing the torchao import chain.

Add an amd-smi circuit breaker in amd.py: after 3 consecutive failures the
module stops spawning the process, eliminating the repeated Windows UAC /
DiskPart elevation prompts caused by polling a non-functional amd-smi.

Also guard BNB_ROCM_VERSION=72 behind a DLL existence check so bitsandbytes
fails with its own detection message rather than a harder "DLL not found" when
the Windows ROCm bnb wheel is not yet installed.

* fix: stub __members__ so torchao float8 enum check doesn't crash on ROCm Windows

torchao.float8.inference accesses ProcessGroup.BackendType.__members__
expecting a Python Enum registry dict. _StubClassMeta.__getattr__ was
blocking all dunder attributes, causing AttributeError. Return {} for
__members__ specifically so the isinstance/iteration checks pass cleanly.

* fix: stub distributed tensor/functional_collectives to prevent missing C++ op crash on ROCm Windows

torch._dynamo.trace_rules eagerly loads torch.distributed.tensor at import
time, which pulls in _functional_collectives.py. That file registers Meta
kernels for _c10d_functional C++ ops, but those ops are only registered
by torch._C._distributed_c10d — a C extension absent from ROCm Windows
wheels. Pre-stubbing the affected modules in sys.modules prevents the real
import chain from running and avoids the "operator does not exist" crash.

* fix: give mod stubs __path__ and pre-stub _tensor to fix 'not a package' import error

_make_mod_stub now sets __path__=[] so Python treats stub modules as
packages. Without it, any import of a submodule raises "is not a package".
Also pre-stub torch.distributed._tensor and its submodules so that
_tensor/__init__.py (which re-exports from torch.distributed.tensor) never
runs and torchao's `from torch.distributed._tensor import DTensor` gets a
harmless stub instead of crashing.

* fix: stub torch.ops._c10d_functional namespace with hashable op sentinels

torchao.dtypes.nf4tensor uses _c10d_functional ops as dict keys at import
time (all_gather_into_tensor.default, wait_tensor.default) and
torch.ops.c10d.scatter_.default. None of these ops are registered on ROCm
Windows because torch._C._distributed_c10d (the C extension) doesn't ship.
Replace the whole _c10d_functional namespace with a custom stub whose ops
return hashable .default objects, so dict-key construction doesn't crash.
Also inject a scatter_ stub into torch.ops.c10d if it's missing.

* fix: stub entire torchao package on ROCm Windows instead of individual ops

torchao is not supported on ROCm Windows and its import chain transitively
requires torch._C._distributed_c10d (absent from the ROCm Windows wheel).
Rather than stub each missing op one by one, stub the whole torchao package
upfront. Unsloth uses bitsandbytes for quantization, not torchao, so this
has no functional impact. transformers gracefully handles an importable-but-
empty torchao by disabling TorchAoHfQuantizer.

* fix: set __spec__ on mod stubs so importlib.util.find_spec doesn't raise

Manually-injected sys.modules entries have __spec__=None by default.
importlib.util.find_spec() raises ValueError when it finds a module in
sys.modules with __spec__=None (transformers.utils.import_utils hits this
when checking if torchao is available). Give every stub a minimal
ModuleSpec(name, loader=None, is_package=True) to satisfy find_spec.

* fix: add meta path finder to auto-stub subpackages of stub modules

`import torchao.prototype` goes through the import machinery, not
__getattr__, so an empty __path__ means ModuleNotFoundError. Rather than
list every submodule explicitly, register a MetaPathFinder that intercepts
any import whose parent is one of our stubs (detected by loader=None in the
parent's ModuleSpec). Real installed packages always have a SourceFileLoader
so they are never intercepted. Also register child stubs in sys.modules
from __getattr__ as a belt-and-suspenders measure.

* fix: use _unsloth_stub sentinel instead of loader=None for stub detection

The import machinery overwrites module.__spec__ with the spec returned by
find_spec (which has loader=_StubSubpackageLoader, not None), so the
loader=None check broke for second-level subpackages. Switch to a custom
_unsloth_stub object identity sentinel set directly on each stub module --
it survives __spec__ being replaced and correctly identifies stubs at any
depth (torchao.prototype.safetensors, etc.).

* refactor(rocm/win): switch to repo.amd.com arch-aware index, remove stubs

AMD recommends repo.amd.com/rocm/whl/{arch}/ as the Windows ROCm wheel
source. These wheels bundle their own ROCm runtime, support all Python
versions (not just cp312), and include the full torch._C extension set
(including _distributed_c10d) that the old repo.radeon.com wheel omitted.

Changes:
- install.ps1: remove Select-ROCmWheelRelease + hardcoded cp312 wheel
  URLs; remove Python 3.12 forced-preference logic; install via
  --index-url repo.amd.com/rocm/whl/{arch-family}/
- studio/setup.ps1: same -- remove Select-ROCmWheelRelease, switch to
  repo.amd.com arch-aware index URL
- studio/install_python_stack.py: replace _ROCM_WINDOWS_RELEASES /
  _select_windows_rocm_release with _windows_rocm_index_url() using the
  _GFX_TO_AMD_INDEX_ARCH map; drop Python 3.12 restriction
- studio/backend/core/training/worker.py: remove all stub machinery
  (_make_mod_stub, _StubSubpackageFinder, _StubSubpackageLoader,
  _StubClassMeta, torchao/fsdp/dtensor stubs, _c10d_functional ops
  stubs, BNB DLL detection) -- no longer needed with new wheel source

* fix(rocm/win): restore _distributed_c10d + torchao stubs; fix BNB install

repo.amd.com torch wheels also omit torch._C._distributed_c10d on Windows
(RCCL is not shipped on Windows). torch/distributed/__init__.py imports
from it unconditionally at module level, so the stub must land in
sys.modules before any torch.distributed import.

torchao (pulled in by transformers.quantizers) walks
torchao.float8.distributed_utils -> torch.distributed._functional_collectives
-> distributed_c10d at import time. Stubbing torchao up-front short-circuits
that chain.

worker.py:
- Restore _make_mod_stub / _StubSubpackageFinder / _StubSubpackageLoader
- Restore _StubClassMeta for ProcessGroup.BackendType attribute access
- Restore _distributed_c10d stub with __getattr__ (Windows only)
- Restore torchao stubs (5 modules, Windows only)

install_python_stack.py:
- BNB AMD wheel install was inside the early-return branch that fires when
  torch is already a ROCm build (installed by install.ps1). Move BNB install
  outside that branch so it always runs on Windows ROCm — the PyPI
  bitsandbytes has only CUDA DLLs and fails to load on ROCm.

* worker: remove _distributed_c10d stub; stub only torchao

The installed torch/distributed/__init__.py from repo.amd.com
(torch==2.10.0+rocm7.12.0) is now properly guarded with
`if is_available():`, so `import torch.distributed` alone is safe.

The crash only comes via torchao's import chain:
  torchao.float8.distributed_utils
    → torch.distributed._functional_collectives (unguarded import)
    → torch.distributed.distributed_c10d
    → torch._C._distributed_c10d  ← absent on Windows ROCm

Stubbing torchao short-circuits the chain entirely. No need to stub
_distributed_c10d. Remove _StubClassMeta and the _c10d stub block;
keep only _make_mod_stub + _StubSubpackageFinder + torchao seeds.

* fix: BNB AMD wheel skipped + torch.compile segfault on Windows ROCm

install_python_stack.py: the UNSLOTH_ROCM_TORCH_INSTALLED=1 early-return
path (set by setup.ps1 when it installed torch itself) returned before
ever reaching the AMD BNB prerelease wheel install.  The PyPI
bitsandbytes==0.49.x ships only CUDA DLLs, so loading it on ROCm fails
with "libbitsandbytes_rocm72.dll not found".  Now installs the AMD
Windows BNB wheel before returning on that path too.

worker.py: torch._grouped_mm crashes on gfx1200 (null HIP kernel pointer,
0xC0000005) when torch.compile's JitDecomp system dispatches it during
the first forward pass.  Detect Windows ROCm via torch.version.hip
(already in sys.modules from section 1e) and set TORCHDYNAMO_DISABLE=1
to bypass the broken kernel dispatch.

* fix: BNB AMD wheel install fails uv wheel filename check

The bitsandbytes continuous-release wheel is intentionally mismatched:
filename encodes 1.33.7.preview (= 1.33.7rc0 in PEP 440) but wheel
metadata reports 0.50.0.dev0.  uv rejects this by default.

Introduce _install_bnb_windows_rocm() helper that sets
UV_SKIP_WHEEL_FILENAME_CHECK=1 only for this specific install, then
restores the previous env value.  Both BNB install call sites (the
UNSLOTH_ROCM_TORCH_INSTALLED early-return path and the normal Windows
ROCm path) now use this helper.

* worker: patch _grouped_mm CUDA dispatch on Windows ROCm (gfx1200 null kernel)

TORCHDYNAMO_DISABLE=1 stopped the compiler frontend but not the autograd
JitDecomp system, which also dispatches _grouped_mm and hits the same
null HIP kernel crash (0xC0000005).

Verified that torch.library.Library("aten","IMPL").impl("_grouped_mm", fn,
"CUDA") successfully overrides the broken HIP kernel with a Python mm
fallback on torch==2.10.0+rocm7.12.0.

Schema: _grouped_mm(Tensor self, Tensor mat2, Tensor? offs=None,
                    Tensor? bias=None, ScalarType? out_dtype=None) -> Tensor

The fallback handles both the simple case (offs=None → torch.mm) and the
grouped case (offs provided → split self by offsets, multiply each group
against the corresponding slice of mat2, then cat results).

Keep _WINDOWS_ROCM_GROUPED_MM_LIB alive at function scope to prevent the
C++ dispatch registration from being freed by GC.

* worker: fix torchao stub — return stub classes not modules for isinstance()

peft/tuners/lora/torchao.py does:
  from torchao.dtypes import AffineQuantizedTensor, LinearActivationQuantizedTensor
  isinstance(weight, (AffineQuantizedTensor, LinearActivationQuantizedTensor))

The stub __getattr__ was returning stub modules, which isinstance() rejects
with "arg 2 must be a type, a tuple of types, or a union".

Add _StubTypeMeta metaclass whose __instancecheck__ always returns False,
and _make_stub_type() to create stub classes via it. Change _make_mod_stub
__getattr__ to return stub classes instead of stub modules for leaf
attribute access, so isinstance() gets a valid type and returns False.

_StubSubpackageFinder still handles import-style subpackage creation
(those still need module objects in sys.modules); __getattr__ only fires
for from-import or direct attribute access, which are the isinstance paths.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* tests: add coverage for Windows ROCm install paths and worker patches

Add conftest.py to fix pre-existing sys.path issue that prevented
test_rocm_support.py from running at all (install_python_stack.py
imports from backend.utils.wheel_utils which needs studio/ on sys.path).

New test classes cover everything added in this session:
- TestWindowsRocmIndexUrl: arch → AMD pip index URL mapping (gfx120X-all,
  gfx1151, gfx1150, gfx110X-all, unknown → None, trailing slash)
- TestDetectWindowsGfxArch: hipinfo output parsing, missing/timeout/bad
  returncode/no-gcnArchName paths
- TestInstallBnbWindowsRocm: UV_SKIP_WHEEL_FILENAME_CHECK set+restored,
  env restored on exception, no-op when URL missing
- TestRocmTorchInstalledEnvVar: UNSLOTH_ROCM_TORCH_INSTALLED=1 skips
  pip_install, calls _install_bnb_windows_rocm, sets flag
- TestWorkerWindowsRocmPatches: _grouped_mm CUDA dispatch override,
  offs/grouped variant handling, GC-prevention sentinel,
  _StubTypeMeta __instancecheck__, _StubSubpackageFinder registration,
  torchao key submodule pre-stubbing, TORCHDYNAMO_DISABLE guard
- TestRocmTorchPkgSpecs: rocm7.2 torch 2.11.x spec, default <2.11 cap,
  3-tuple shape, _GFX_TO_AMD_INDEX_ARCH RDNA4/3.5/3 coverage

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* tests: fix encoding, IS_WINDOWS patching, and wrong assertion

- Add encoding="utf-8" to all read_text() calls (54 occurrences) so
  tests pass on Windows where the default codec is cp1252 and source
  files contain UTF-8 emoji (e.g. ⚠️ in install_python_stack.py)
- Add @patch.object(stack_mod, "IS_WINDOWS", False) to Linux-path
  TestEnsureRocmTorch tests so they reach the Linux code path when run
  on a Windows machine instead of short-circuiting into the Windows branch
- Fix test_grouped_mm_patch_guarded_by_windows_and_hip_check: the source
  uses getattr(_torch_for_rocm, "version", None) not torch.version, so
  check for '"version"' and '"hip"' substrings instead

137 passed, 2 skipped

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: pin BNB_ROCM_VERSION=72 for torch==2.11.0+rocm7.13.0 compatibility

AMD's pip index now ships torch==2.11.0+rocm7.13.0 (ROCm 7.13).
bitsandbytes auto-detects HIP 7.13 from torch.version.hip and looks for
libbitsandbytes_rocm713.dll, which the AMD Windows prerelease wheel does
not ship (it only ships rocm72.dll), causing a load error at training start.

Fix:
- worker.py section 1f: set BNB_ROCM_VERSION=72 (via setdefault) before
  section 2 ML imports, so bitsandbytes always loads rocm72.dll on Windows ROCm
- install_python_stack.py: set BNB_ROCM_VERSION=72 in _install_bnb_windows_rocm()
  for any post-install imports; update comment to document root cause
- tests: 4 new assertions covering the fix (141 passed, 2 skipped)

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: detect BNB ROCm DLL suffix dynamically instead of hardcoding '72'

BNB_ROCM_VERSION was pinned to '72' which works today (AMD wheel ships
rocm72.dll) but would break again if AMD ships a future wheel with a
different DLL suffix (e.g. rocm713.dll).

Add _detect_bnb_rocm_dll_ver() to install_python_stack.py: scans the
installed bitsandbytes package dir for libbitsandbytes_rocm{VER}.dll
using importlib.util.find_spec (no BNB import needed) and returns the
suffix.  '72' remains the fallback when detection fails.

Apply the same detection inline in worker.py section 1f.  Both paths
still respect a pre-set BNB_ROCM_VERSION (caller override wins).

Tests: +8 cases covering detection logic and fallback (147 passed, 2 skipped).

* fix: patch torch.distributed stubs in server process for Windows ROCm

On Windows ROCm, torch.distributed ships without process-group helpers
(is_initialized, is_available, get_rank, get_world_size).  The worker
subprocess already patches these in section 1e, but the main server
process calls _determine_attention_impl_for_gpu_estimate() which calls
unsloth's resolve_attention_implementation() → is_initialized(), causing:

  "Could not resolve attention implementation for '...':
   module 'torch.distributed' has no attribute 'is_initialized'"

Fix: patch the missing attrs onto torch.distributed at the top of
_determine_attention_impl_for_gpu_estimate, matching the same stubs
already applied in worker.py section 1e.  No-ops on Linux/CUDA where
torch.distributed is fully populated.

* fix: gate _grouped_mm dispatch patch on HIP < 7.13

AMD fixed the gfx1200 null HIP kernel in ROCm 7.13 (torch 2.11+).
Users on the new wheel now get the real GPU _grouped_mm kernel for
MoE workloads instead of the Python mm fallback.

Changes:
- worker.py: add _hip_ver_at_least() helper; wrap full _grouped_mm
  patch in `if not _hip_ver_at_least(7, 13):` with else branch that
  logs the skip reason; update section-1f comment to document the fix
- test_rocm_support.py: add 5 tests covering the helper definition,
  the (7, 13) gate expression, the else branch, the skip log message,
  and the AMD-format version string parsing (.split(".")[:2])

Verified: torch==2.11.0+rocm7.13.0 — 3D batch and grouped (offs)
variants both succeed; null crash only present on rocm7.12 and earlier.

* fix: stub is_torchelastic_launched on torch.distributed for Windows ROCm

resolve_attention_implementation calls is_torchelastic_launched() which
does not exist in the incomplete torch.distributed shipped with the
Windows ROCm wheel, causing a warning on every model config load in the
server process. Add it to the stub table alongside the four helpers
already patched in _determine_attention_impl_for_gpu_estimate.

Also adds two tests: one confirming the new stub and one confirming all
five core distributed helpers are covered.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: explicit warnings on AMD ROCm arch/version fallbacks + Fast-Install arg order

setup.ps1:
- Fix Fast-Install argument order: packages before flags, consistent with
  all other Fast-Install calls in the file
  (was: Fast-Install --force-reinstall --index-url $url torch ...)
  (now: Fast-Install torch torchvision torchaudio --force-reinstall --index-url $url)
- Add explicit [WARN] substep when $HasROCm is true but arch mapping fails:
  - GPU arch detected but not in supported wheel list → names the arch and
    lists supported families so user knows exactly what to report
  - HIP SDK present (amd-smi path) but gcnArchName unreadable → instructs
    user to re-install the HIP SDK; previously fell back silently to CPU

install.sh:
- Add [WARN] to stderr before silent CPU fallback when AMD GPU is confirmed
  (rocminfo/amd-smi) but ROCm version cannot be read from any source
  (amd-smi, /opt/rocm/.info/version, hipconfig, dpkg, rpm)
- Add [WARN] to stderr when ROCm version is too old (< 6.0) with upgrade link

install.ps1 and setup.sh: no changes needed (already handle these paths correctly)

* fix: robust gfx arch detection for Strix Halo / HIP-runtime-only installs

Covers users who have the HIP runtime (amd-smi available) but not the
full HIP SDK (no hipinfo), which is common on Strix Halo iGPU systems.
Without this, $ROCmGfxArch stays null and the installer silently falls
back to CPU-only PyTorch despite a working GPU.

Detection waterfall (setup.ps1 + install.ps1):
  1. hipinfo gcnArchName          -- full HIP SDK (existing, unchanged)
  2. amd-smi list gfx pattern     -- newer amd-smi versions embed arch
  3. amd-smi static --asic        -- ROCm 6+ ASIC details with GFX target
  4. UNSLOTH_ROCM_GFX_ARCH env    -- manual override escape hatch
  5. GPU name → arch table        -- best-effort from marketing name:
       890M / Strix Halo  → gfx1151 (RDNA 3.5 iGPU, Strix Halo)
       880M / Strix Point → gfx1150 (RDNA 3.5 iGPU, Strix Point)
       780M / Phoenix     → gfx1103 (RDNA 3 iGPU)
       RX 7900/7800/7700  → gfx1100 (RDNA 3 desktop)
       RX 9070 XT / 9080  → gfx1201 (RDNA 4)
       RX 9070 / 9060 XT  → gfx1200 (RDNA 4)

When arch is inferred from name, a Cyan substep tells the user to set
UNSLOTH_ROCM_GFX_ARCH to skip inference on future installs.
WMI block intentionally does not set $HasROCm (no runtime confirmation).

Tests: 11 new tests in TestStrixHaloGfxArchDetection covering all five
detection levels, WMI safety, and gfx regex in both ps1 files.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: resolve hipinfo/hipconfig via HIP_PATH/ROCM_PATH when not on PATH

AMD HIP SDK sets HIP_PATH on Windows but does not always add the bin
directory to PATH.  Get-Command hipinfo therefore silently fails and
detection falls through to WMI, which cannot provide a gfx arch, leaving
the user with a CPU-only PyTorch install and no warning.

Changes:
- setup.ps1 / install.ps1: before falling through to amd-smi, attempt to
  locate hipinfo.exe and hipconfig.exe under $env:HIP_PATH\bin (then
  $env:ROCM_PATH\bin) when Get-Command returns nothing
- Emit a [WARN] with the resolved path and a one-liner to permanently fix
  PATH via SetEnvironmentVariable
- Emit a [WARN] when HIP_PATH/ROCM_PATH is set but the exe is still not
  found (incomplete SDK install)
- Emit a [WARN] with the first hipinfo output line when hipinfo runs but
  returns a non-zero exit code (e.g. "no ROCm-capable device detected")
- 18 new tests in TestHipSdkEnvPathResolution; total 183 passed, 2 skipped

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* feat: print HIP SDK path and full hipconfig version in terminal on AMD detection

Both install.ps1 and setup.ps1 now emit substeps under the gpu step when
AMD ROCm is detected:

  gpu  AMD ROCm (gfx1200)
       HIP SDK: C:\Program Files\AMD\ROCm\7.1
       hipconfig: 7.1.51803-d3a86bd04

Previously only the gpu label (e.g. "AMD ROCm (gfx1200)") was shown with
no indication of where the SDK was found or which exact build was active.
The full hipconfig build string (e.g. 7.1.51803-d3a86bd04 instead of just
7.1) is now stored in ROCmVersionFull and also used in setup.ps1's
'rocm' step label.

9 new tests in TestHipSdkDetectedSubstep; total 192 passed, 2 skipped

* fix: Strix rocm7.1 segfault bypass + Ubuntu 24.04 HIP gcc-install-dir

Issue 1 (install.sh): gfx1151/gfx1150 + ROCm 7.1 causes a segfault in
torch._grouped_mm (moe_utils.py:167). The Radeon repo now ships cp313
wheels for rocm-rel-7.1, so _amd_gpu_radeon=true silently lands on the
broken combo. When Strix Halo/Point is detected and TORCH_INDEX_URL is
rocm7.1, override to rocm7.2 PyTorch index, update TORCH_CONSTRAINT, and
set _amd_gpu_radeon=false to bypass the Radeon repo entirely. Emits a
clear [WARN] explaining the segfault and linking to the ROCm upgrade docs.

Issue 2 (setup.sh): ROCm 7.x ships clang-20 which on Ubuntu 24.04+ picks
/usr/lib/gcc/x86_64-linux-gnu/14/ (runtime dir, no C++ headers), causing
'cstdlib file not found' and a failed llama.cpp HIP build. Iterate gcc
versions 14→11 to find the first install dir that has both runtime and
/usr/include/c++/<ver> headers, then pass --gcc-install-dir to clang via
CMAKE_HIP_FLAGS. Fix confirmed by h34v3nzc0dex (llama.cpp 417/417 clean).

11 new tests across TestStrixRocm71Override and TestSetupShGccInstallDir;
total 203 passed, 2 skipped

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: BNB_ROCM_VERSION in server process + torch._C._distributed_c10d stubs

Two errors visible in training logs on Windows ROCm:

1. Server process bitsandbytes crash:
   "Configured ROCm binary not found at libbitsandbytes_rocm713.dll"
   The installed BNB wheel ships rocm72.dll (not rocm713.dll). The
   training worker already sets BNB_ROCM_VERSION=72 via DLL detection
   but the server process (main.py) imported bitsandbytes before that
   ran. Fix: add the same DLL-scan + BNB_ROCM_VERSION assignment to
   main.py inside the existing win32 guard, before any downstream
   import can pull in bitsandbytes.

2. torch.distributed import failure:
   "No module named 'torch._C._distributed_c10d'; torch._C is not a package"
   torch._C is a C extension on Windows ROCm — Python cannot do
   submodule imports from it, so torch.distributed fails to import
   before our attribute stubs could ever run. Fix: inject empty
   ModuleType stubs for _distributed_c10d, _distributed_autograd and
   _distributed_rpc into sys.modules inside the win32 guard in
   hardware.py BEFORE importing torch.distributed, so the import
   succeeds and our attribute stubs take effect.

9 new tests in TestServerStartupRocmFixes; total 212 passed, 2 skipped

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(win32): populate distributed c10d stub with dummy symbols

torch.distributed tries to `from torch._C._distributed_c10d import
FakeProcessGroup` (and ProcessGroup, Work, Store, etc.).  The previous
empty ModuleType stub caused an AttributeError on those names.

Populate every stub with a _Dummy class for each known symbol so the
import chain completes silently on Windows ROCm where torch._C is a
compiled extension and its _distributed_c10d submodule doesn't exist.

Adds four new tests in TestServerStartupRocmFixes covering FakeProcessGroup,
ProcessGroup, setattr population, and all three _distributed_* siblings.

* fix(win32): distinguish HIP SDK installed vs GPU not ROCm-accessible

Previously, when hipinfo was found but exited non-zero (e.g. "no
ROCm-capable device detected"), both install.ps1 and setup.ps1 fell
through to the WMI-label-only branch and printed "AMD GPU detected --
HIP SDK not found" -- factually wrong since the SDK binary is present.

Add $HipSdkInstalled flag (set true when hipinfo binary is found,
regardless of exit code). When HipSdkInstalled && !HasROCm:
- Show "AMD GPU detected -- not ROCm-accessible (HIP <ver>)" instead
- Explain this is a driver issue, not an SDK issue, with a link
- Still run hipconfig version capture so version shows in output
- CPU-only hint now says "GPU not ROCm-accessible" not "require HIP SDK"

Also applies to setup.ps1 (same detection block, same branches).

Adds TestHipSdkInstalledButDeviceInaccessible (11 tests).

* fix(win32): scope ROCm workarounds to AMD hosts only

Three Codex-flagged issues where Windows ROCm workarounds incorrectly
applied to Windows CUDA (NVIDIA) machines:

main.py (P1): BNB_ROCM_VERSION was set unconditionally on all win32
hosts. On NVIDIA, bitsandbytes sees BNB_ROCM_VERSION and looks for a
ROCm DLL that doesn't exist, breaking bitsandbytes initialisation.
Fix: gate the block on HIP_PATH/ROCM_PATH being present (ROCm hosts only).

worker.py (P2): torchao stubs were seeded for all win32 runs, shadowing
real torchao on Windows CUDA and silently disabling torchao quantization
for NVIDIA users. Fix: gate on HIP_PATH/ROCM_PATH (win32 ROCm only).

install_python_stack.py (P1): _detect_windows_gfx_arch() only checked
shutil.which("hipinfo"), skipping the HIP_PATH/ROCM_PATH fallback that
the PowerShell installers use. On installs where the HIP SDK bin dir is
not on PATH, _ensure_rocm_torch() returned early without installing
ROCm wheels or bitsandbytes. Fix: mirror the env-var fallback.

* fix(linux): route Strix + ROCm 7.1 to AMD arch-specific index

Instead of falling back to pytorch.org/rocm7.2, the Strix override now
routes to repo.amd.com/rocm/whl/gfx1151/ (or gfx1150/) which serves
torch 2.11.0+rocm7.13.0 -- AMD's build containing the actual _grouped_mm
kernel fix, verified on real gfx1151 hardware by h34v3nzc0dex.

This exercises the real GPU kernel path rather than the rocm7.2 workaround.
UNSLOTH_AMD_ROCM_MIRROR can override the base URL for air-gapped installs.

Also teaches _tauri_torch_index_family to recognise AMD arch-specific URLs
(repo.amd.com/rocm/whl/gfx*) and return the rocm7.13 family label so
_tauri_gpu_branch correctly classifies these installs as rocm.

Suggested by h34v3nzc0dex based on hardware-verified probe results.

* fix(studio/rocm): gate ROCm-only side-effects on active torch runtime

Address five edge cases flagged during PR review:

1. studio/backend/main.py: BNB_ROCM_VERSION was set whenever HIP_PATH or
   ROCM_PATH was present in the environment. A Windows CUDA user who once
   installed the HIP SDK and reverted to a CUDA torch wheel still has those
   env vars set, so bitsandbytes would try to load libbitsandbytes_rocm72.dll
   against a CUDA torch and crash. Now probe torch.version.hip inside the
   env-var guard (worker.py already does this).

2. studio/backend/main.py: os.add_dll_directory returned handles were
   discarded. Per CPython docs, the directory leaves the DLL search list when
   the handle is garbage collected. Retain handles in module-level
   _ROCM_DLL_HANDLES list so they survive process lifetime.

3. studio/install_python_stack.py: _install_bnb_windows_rocm() returned None
   regardless of pip_install_try outcome, and the caller flipped
   _rocm_windows_torch_installed to True unconditionally. On a failed BNB
   install the post-install "manual install may be required" warning was
   suppressed and the user was misled. Helper now returns bool; caller gates
   on it.

4. studio/install_python_stack.py: _detect_windows_gfx_arch returned the raw
   capture group, so mixed-case hipinfo output ("Gfx1151") missed the
   lowercase keys in _GFX_TO_AMD_INDEX_ARCH and silently fell back to CPU
   torch. Lowercase the token.

5. studio/install_python_stack.py: UNSLOTH_ROCM_TORCH_INSTALLED=1 early-
   return trusted the env var even when the venv was wiped between runs.
   Subprocess-probe torch importability first; fall through to the full
   install path if the probe fails.

Tests: 231 passed, 1 skipped in tests/studio/install/test_rocm_support.py
(adds one new test for case 5 fall-through).

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/rocm): worker.py parity + don't roll back ROCm torch on bnb failure

Addresses findings from a 10x reviewer pass on the prior fix commit:

1. studio/backend/core/training/worker.py (parity with main.py):
   - Gate the torchao stub block on torch.version.hip / 'rocm' in
     torch.__version__ instead of HIP_PATH / ROCM_PATH env-var presence.
     Same root cause as main.py: HIP SDK env vars stick around on CUDA hosts.
   - Add module-level Windows ROCm DLL registration block. Worker subprocesses
     inherit env vars but not the parent's add_dll_directory handles, so the
     first `import torch` in the worker could fail to find amdhip64.dll when
     HIP_PATH\bin is not on PATH. Mirrors main.py setup. Handles retained at
     module scope via _ROCM_DLL_HANDLES.
   - Promote _WINDOWS_ROCM_GROUPED_MM_LIB to module scope with `global` in
     run_training_process so the torch.library.Library registration survives
     past function return / mid-run garbage collection.
   - Harden _torch_has_hip() to also accept 'rocm' in torch.__version__
     (AMD SDK / Radeon wheels may not set torch.version.hip).

2. studio/install_python_stack.py:
   - Don't roll back ROCm torch when bitsandbytes install fails. The prior
     commit gated _rocm_windows_torch_installed on _install_bnb_windows_rocm()
     returning True; if torch installed successfully but bnb failed, the flag
     stayed False and later install steps could overwrite ROCm torch with the
     generic CPU torch wheel. Set the flag after torch install; surface bnb
     failure as a separate warning instead.
   - _detect_windows_gfx_arch now probes in three tiers: UNSLOTH_ROCM_GFX_ARCH
     env-var override (matches the PowerShell installer), then hipinfo (PATH
     or HIP_PATH\bin), then amd-smi (`static --asic`, `list`). Without the
     amd-smi fallback, runtime-only Radeon installs without hipinfo on PATH
     made `studio update` return early and leave the venv on CPU torch.
   - Linux torch-already-rocm probe in _ensure_rocm_torch now matches the
     Windows probe shape: accepts torch.version.hip OR 'rocm' in
     torch.__version__ to cover AMD SDK / Radeon Linux wheels.

3. studio/backend/utils/hardware/hardware.py:
   - apply_gpu_ids() final-fallback torch probe accepts 'rocm' in
     torch.__version__ in addition to torch.version.hip, matching
     detect_hardware(). AMD SDK wheels could otherwise leak through with
     CUDA-only visibility masks on a spawned ROCm worker.

Tests: 231 passed, 1 skipped in tests/studio/install/test_rocm_support.py
(no test changes needed; the probe shape that prints the hip version (or
'rocm' sentinel) preserves the existing non-empty-string contract).

Not addressed in this commit (deferred or out of scope):
- Tag drift / lemonade checksum (PR 5303 surface, not this PR).
- install.sh rocm7.2.1 URL: small fix, separate.
- install.ps1 / setup.ps1 'Radeon 8060S' marketing-name fallback table.
- Strix Halo + ROCm 7.1 routing asymmetry in Python update path.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/rocm): robustness pass - rocm tag normalisation, Strix routing parity, hardened detection

Robustness pass on top of 76137b2d. Four targeted fixes:

1. install.sh ROCm-tag routing normalisation.
   `rocm7.2.1` would route to https://download.pytorch.org/whl/rocm7.2.1
   which does not exist (PyTorch publishes major.minor URLs only). Same
   for any future patch-level tag. Normalise every rocm{maj.min}* pattern
   to the bare {maj.min} index URL.

2. install.ps1 + studio/setup.ps1 marketing-name fallback.
   The gfx1151 row matched 890M / Strix Halo / HX 37x / HX 38x / AI 9 HX
   but not the actual retail name 'AMD Radeon 8060S Graphics' shipped by
   OEMs (Ryzen AI MAX+ 395). Add '8060S' to the regex.

3. install_python_stack.py Strix + ROCm 7.1 routing parity with install.sh.
   The shell installer reroutes Strix Halo / Point + ROCm 7.1 to
   repo.amd.com/rocm/whl/{gfx}/ (which serves torch 2.11.0+rocm7.13.0
   with the upstream _grouped_mm fix). The Python `studio update` path
   only warned and still installed the broken generic rocm7.1 wheel.
   Mirror the override: detect gfx1151/gfx1150 on ROCm 7.1, route to
   the AMD per-gfx index, honour UNSLOTH_AMD_ROCM_MIRROR override.

4. _detect_windows_gfx_arch amd-smi parsing tightened.
   The amd-smi fallback added in the prior commit used a bare
   `\bgfx[1-9][0-9a-z]{2,3}\b` match against the lowercased stdout,
   which could pick up stray gfx references in warnings / device-name
   strings. Anchor on labelled lines first (Target_Graphics_Version,
   ASIC, Arch, gfx) and fall back to the bare match only when no
   labelled line is present.

Tests: 231 passed, 1 skipped in tests/studio/install/test_rocm_support.py;
sim_5301 23 cases pass (6 new sims for the Strix override + amd-smi parsing).

* fix(studio/rocm): multi-GPU selection, Strix sibling handling, defensive cleanups

Round 4 robustness pass based on 5 parallel Opus reviewers of head 21773215.
Seven items from across regression / edge-case / error-paths / architecture
reviews:

1. studio/backend/main.py BNB gate: aligned with the broad ROCm check used
   everywhere else in this PR (torch.version.hip OR 'rocm' in __version__).
   AMD SDK / Radeon Linux wheels do not always populate torch.version.hip;
   without this, main.py would silently skip BNB_ROCM_VERSION while worker.py
   set it.

2. studio/install_python_stack.py _install_bnb_windows_rocm: init _ok = False
   before the try block. Without this, if pip_install_try itself raises
   (e.g. OSError on uv binary missing), the finally block restored env vars
   correctly but the subsequent `if not _ok:` raised UnboundLocalError,
   masking the original exception.

3. studio/install_python_stack.py _detect_windows_gfx_arch:
   - Rewrote to use re.findall (not re.search) on both hipinfo and amd-smi
     output, dedup tokens preserving order, and select via new
     _pick_visible_index() helper.
   - HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES (first comma entry, integer)
     now picks the right GPU on multi-AMD-GPU hosts. Out-of-range or non-int
     values fall back to the first GPU (matches detect_host behaviour in
     install_llama_prebuilt.py).

4. studio/install_python_stack.py Strix override now consults the runtime
   target before flipping:
   - Previous behaviour intersected gfx_codes with {gfx1151, gfx1150} and
     picked the first Strix arch, ignoring whether HIP_VISIBLE_DEVICES
     selected a non-Strix sibling (e.g. discrete RX 7900 in a mixed APU+dGPU
     box). Could install Strix-specific wheels onto a gfx1100 dGPU.
   - Now resolves the runtime gfx via _pick_visible_index() and only
     overrides when that runtime target is in the Strix set.

5. studio/backend/main.py + studio/backend/core/training/worker.py: ROCm
   version dir scan no longer sorts lexically. Previous sort placed "10.0"
   before "7.0" alphabetically, which would mis-prioritise ROCm 10.x bin
   dirs once AMD ships them. New _ver_key() splits on "." and sorts
   numerically with a string fallback.

6. install.sh Strix override URL: replaced ${var%/} (strips one trailing
   slash) with a while-loop that strips all trailing slashes, matching
   Python's .rstrip("/"). A user setting UNSLOTH_AMD_ROCM_MIRROR with
   "http://corp/whl///" no longer ends up with "http://corp/whl///gfx1151/"
   which strict pip proxies (artifactory, sonatype) 404 on.

7. studio/install_python_stack.py: bumped torch import probe timeout from
   30s to 90s. PyTorch's lazy .so loading can take 60-90s on cold NFS or
   USB-backed venvs. The shorter timeout was producing a false "torch
   missing" classification and reinstalling a working ROCm torch.

Tests: 231 passed, 1 skipped. sim_5301 30 cases pass (added 7 new sims for
multi-GPU detection, Strix sibling handling, and _ok-init regression).

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/rocm): worker BNB/grouped_mm broad gate, install.sh Strix visibility, runtime-only ROCm detection

Round-5 robustness pass based on 20 parallel reviewers of head 96b9e465.

1. studio/backend/core/training/worker.py - BNB version pin / dynamo disable
   / _grouped_mm fallback block was still gated on torch.version.hip alone
   despite the torchao stub block above already using the broad check. AMD
   SDK / Radeon Windows wheels (torch.__version__ contains "rocm" but
   torch.version.hip is None) silently skipped the Windows ROCm runtime
   patches. Aligned to the same broad check (8/20 reviewers).

2. studio/backend/core/training/worker.py - _hip_ver_at_least() now also
   parses the ROCm version out of torch.__version__ (e.g. "2.11.0+rocm7.13.0")
   when torch.version.hip is missing, so the kernel-fix gate is correct for
   SDK / Radeon wheels too.

3. studio/backend/core/training/worker.py - _grouped_mm_safe_impl with
   offs=None now picks torch.bmm/matmul for 3-D inputs instead of always
   calling torch.mm. The real _grouped_mm accepts 3-D batched matmul; the
   prior fallback raised "self must be a matrix" on MoE workloads (2/20).

4. studio/backend/main.py - dropped the HIP_PATH / ROCM_PATH env-var gate
   from the BNB block; probe torch directly. Runtime-only Radeon / AMD SDK
   Windows installs do not set those SDK env vars but still ship ROCm torch
   (5/20 reviewers).

5. install.sh - Strix override now collects every gfx token from
   rocminfo / amd-smi (in enumeration order), then indexes by
   HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES so a mixed Strix iGPU + non-
   Strix dGPU host where the user selected the dGPU does NOT get rerouted
   to the Strix per-gfx index. Mirrors the Python update path (5/20 reviewers).

6. install.sh - Strix detection chain now also probes `amd-smi static --asic`,
   matching the PowerShell installer (1/20). Closes the gap on runtime-only
   Strix hosts where `amd-smi list` does not surface a gfx token.

7. studio/install_python_stack.py - _has_rocm_gpu() now has the sysfs KFD
   topology fallback (/sys/class/kfd/kfd/topology/nodes/*/gpu_id), matching
   install.sh. On minimal package-managed installs without rocminfo /
   amd-smi GUI tools, `studio update` can now detect the GPU and repair the
   venv instead of returning early (2/20).

8. studio/install_python_stack.py - _detect_amd_gfx_codes() now falls back
   to `amd-smi list` and `amd-smi static --asic` when rocminfo is missing
   (2/20). Strix routing on runtime-only Radeon hosts now matches what
   install.sh has done for a while.

9. studio/install_python_stack.py - Strix override now applies even when
   has_hip_torch is True. The whole point of the override is to repair an
   existing broken torch.version.hip == "7.1" install; skipping the
   reinstall left users on the known _grouped_mm segfaulting stack (3/20).

Tests: 231 passed, 1 skipped. sim_5301 30 cases pass. sim_cross 12 pass.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/rocm): code review hardening pass

- main.py: numeric DLL sort (string sort picked rocm72 over rocm713);
  add basename() to regex; log warning on detection failure; log info
  when BNB_ROCM_VERSION is set (mirrors worker.py)
- worker.py: explicit len-guard in _hip_ver_at_least() with warning
  logs instead of silent IndexError/ValueError swallow
- hardware.py: isinstance(result, dict) guard before result.get() in
  _smi_query() to prevent AttributeError on non-dict backend returns
- amd.py: round() before int() on parsed GPU IDs; log warning when
  truncation occurs (defensive against malformed amd-smi output)
- setup.sh: quote --gcc-install-dir value in CMAKE_HIP_FLAGS so paths
  with spaces do not break the CMake argument
- install.ps1, setup.ps1: apply colon-split + ToLower() to hipinfo
  gcnArchName match (consistent with each other and with setup.sh)
- install.sh: tighten ROCm tag case patterns to explicit
  rocmX.Y|rocmX.Y.* to avoid unintended prefix matches

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/training): GPU OOM guard to prevent system freeze on VRAM exhaustion

On RDNA 4 (gfx1200/gfx1201) and other ROCm GPUs, exhausting VRAM can
cause a HIP driver hang that freezes the entire system rather than
raising a recoverable Python exception.

Two-part fix:
- set_per_process_memory_fraction(0.90) caps the HIP/CUDA allocator at
  90% of VRAM so PyTorch raises OutOfMemoryError before hitting the
  hardware limit, keeping the driver alive and the system responsive
- top-level exception handler detects OOM errors by type and message
  and surfaces a clear actionable message to the UI (reduce
  max_seq_length, enable gradient_checkpointing, lower batch size)
  instead of the raw CUDA/HIP error string

* fix(studio/rocm): OOM guard ROCm-only + unified memory, multi-GPU arch selection

OOM guard (worker.py):
- Scope to _hw.IS_ROCM only -- NVIDIA CUDA has a graceful OOM path and
  does not need the allocator cap
- Detect unified memory by comparing torch VRAM against psutil system RAM;
  use 0.80 on unified-memory APUs (gfx1151 Strix Halo) where the GPU pool
  is carved from host RAM, 0.90 on discrete cards

Multi-GPU arch selection:
- install.ps1 / setup.ps1: replace -match (first hit only) with
  [regex]::Matches() to collect all gcnArchName entries, then index by
  HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES
- install_python_stack.py: index into full token list before dedup so
  HIP_VISIBLE_DEVICES=2 on [gfx1100, gfx1100, gfx1151] resolves gfx1151
- install.sh: remove awk dedup from gfx token collection for same reason

GCC multiarch (setup.sh):
- Only append -linux-gnu when gcc -print-multiarch does not already return
  the full triple, fixing double-suffix on Ubuntu 24.04

* fix(tests): update ROCm version cap expectations from rocm7.1 to rocm7.2

Daniel's normalisation commit updated the cap from rocm7.1 to rocm7.2
since PyTorch now publishes that index and rocm7.2 ships torch 2.11.0.
Test expectations were stale.

* fix(tests): correct MLX smoke test losses_per_step assertion

logging_steps=1 with max_steps=30 produces 30 loss entries, not 7.
The assertion was stale from a previous config.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(studio/worker): detect unified-memory APU by GPU name not VRAM/RAM ratio

The previous heuristic (VRAM > 50 % of system RAM) false-positived on discrete
cards in low-RAM systems — e.g. RX 9060 XT 16 GB on a 16 GB or 24 GB machine
would trip the unified-memory path and log "unified memory host" when it should
say "discrete".

AMD iGPUs (gfx1150/gfx1151 Strix Halo, Strix Point, etc.) expose names with a
digit+M suffix ("AMD Radeon 890M"), while discrete cards use "RX NNNN [XT|XTX]"
naming.  Matching that suffix is reliable across all current ROCm-capable AMD
consumer GPUs and does not require psutil.

Also includes the device name in the log line to ease future debugging.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(install/setup.ps1): force array on hipinfo gcnArchName parse to fix single-GPU arch truncation

When [regex]::Matches() finds exactly one match, PowerShell's pipeline
unwraps the result to a scalar string.  Indexing a scalar string with [0]
returns the first *character*, so a one-GPU system would parse
gcnArchName "gfx1200" as "g", which is not in the supported arch map
and triggers the CPU-only fallback.

Wrapping with @() forces the result to remain an array regardless of
match count.  On a single-GPU machine the arch is now correctly read as
"gfx1200" (or whatever the full name is) so the ROCm wheel index is
selected.

Reproducer: hipinfo exits 0 and outputs exactly one gcnArchName line.
Without @(), $_hipAllArches = "gfx1200" (String); $_hipAllArches[0] = 'g'.
With @(), $_hipAllArches = @("gfx1200") (Object[]); $_hipAllArches[0] = "gfx1200".

* fix(studio/rocm): classify unified-memory APU via VRAM/RAM ratio, not arch list

Replace the gcnArchName allowlist {gfx1150, gfx1151} with a
psutil-based heuristic: unified APUs expose the entire system RAM
as the HIP pool (ratio ≥ 0.90), discrete cards are well below that.
No arch name required — future APUs classify correctly without code changes.

Also removes the stale import re / \d[Mm]\b device-name regex that
5d84704 left behind, and logs vram/sys GiB for easier on-hardware
verification.

Addresses h34v3nzc0dex review: Radeon 8060S (gfx1151, 128 GiB
unified) now correctly gets 0.80 cap instead of 0.90.

* fix(studio/rocm): revert to gcnArchName for unified-memory APU classification

VRAM/RAM ratio >= 0.90 false-positives on machines where discrete VRAM
equals system RAM (e.g. RX 9060 XT 16 GB + 16 GB system RAM → ratio 1.0,
incorrectly classified as unified → wrong 0.80 cap applied).

gcnArchName is the correct signal: naming-independent, stable within a
product family, and already parsed throughout this PR. Unified set is
{gfx1150, gfx1151} (Strix Point + Strix Halo).

* fix(studio/llama-prebuilt): resolve hipinfo via HIP_PATH/ROCM_PATH on Windows

shutil.which("hipinfo") returns None when the HIP SDK bin dir is not on
PATH -- the HIP SDK installer sets HIP_PATH/ROCM_PATH but does not always
add the bin dir to PATH. This caused has_rocm=False in the prebuilt asset
selector, so AMD ROCm machines got the CPU llama.cpp zip instead of the
HIP one, silently running all chat inference on CPU.

Add _resolve_exe() that falls back to %HIP_PATH%\bin and %ROCM_PATH%\bin
when shutil.which() finds nothing, mirroring the same fallback already
present in setup.ps1.

* fix(studio/llama-prebuilt): pass --has-rocm from setup.ps1 to skip re-detection

The Python prebuilt installer re-detects ROCm independently via
shutil.which("hipinfo"), which fails when hipinfo is not on PATH
(HIP SDK sets HIP_PATH but doesn't always add the bin dir to PATH).
This caused has_rocm=False and downloaded the CPU llama.cpp zip even
on confirmed AMD ROCm machines.

setup.ps1 already performs reliable ROCm detection with its own
HIP_PATH/ROCM_PATH fallback. Add --has-rocm flag to
install_llama_prebuilt.py so setup.ps1 can forward its result directly,
and pass it whenever $HasROCm is true. The Python script then overrides
has_rocm=True in the HostInfo without re-probing.

* fix(studio/llama-prebuilt): add HIP asset to simple-policy Windows path

direct_upstream_release_plan (used by --simple-policy, which setup.ps1
always passes) only checked has_usable_nvidia on Windows and fell
straight to CPU for AMD ROCm machines, ignoring has_rocm entirely.
The --has-rocm override had no effect because the simple-policy code
path never reached resolve_asset_choice where has_rocm was checked.

Add an elif branch for has_rocm that tries the upstream HIP asset
(llama-TAG-bin-win-hip-radeon-x64.zip) before falling through to the
CPU fallback, consistent with the non-simple-policy path.

* fix(studio/setup.ps1): auto-remove mismatched llama.cpp install kind

When an existing llama.cpp install is the wrong kind for the current
GPU (e.g. windows-cpu on an AMD ROCm machine that should have
windows-hip), the prebuilt installer skips on tag match and never
upgrades. Read install_kind from UNSLOTH_PREBUILT_INFO.json before
invoking the installer and remove the directory if the kind doesn't
match, forcing a fresh download of the correct variant.

* fix(studio/setup.ps1): show live PyTorch install output in verbose mode for ROCm

The ROCm torch reinstall (setup.ps1 phase) always silently captured
output, so in --verbose mode the torch downgrade mid-install
(2.11.0+rocm → 2.10.0 → 2.11.0+rocm) looked like the final state was
2.10.0. Match the CPU/CUDA blocks which show live uv output when
$script:UnslothVerbose is set.

* fix(rocm/windows): set ROCBLAS_TENSILE_LIBPATH for bundled rocblas.dll

The llama.cpp ROCm prebuilt bundles rocblas.dll next to the binary but
not the Tensile kernel library files it depends on at runtime
(rocblas/library/TensileLibrary*.dat + *.hsaco).  The bundled DLL
searches for these files relative to its own location by default, i.e.
<binary_dir>/rocblas/library/, which does not exist in the prebuilt
install tree.  This causes a silent crash on the very first GEMM
(prefill) with no output from llama-server, seen by the caller as
WinError 10054 / 10061.  Model load and the single-token warmup pass
because they use simpler code paths that do not trigger rocBLAS GEMM.

Fix: set ROCBLAS_TENSILE_LIBPATH in the subprocess env to
<HIP_PATH>/bin/rocblas/library so the bundled DLL finds the kernel
files from the system ROCm installation.  Uses setdefault so a user-
supplied env var is never overwritten.  No-ops on CUDA and CPU (no
HIP_PATH) and on Linux (win32 branch only).

Reproducer log:
  rocBLAS error: Cannot read .../Release/rocblas/library/TensileLibrary.dat
  rocBLAS error: Could not initialize Tensile host:
  directory_iterator: The system cannot find the path specified.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(install.sh): restore gfx token dedup in Strix multi-GPU awk indexer

536a54df removed the per-source `| awk '!seen[$0]++'` dedup from the
_gfx_all collection step but left the indexer awk as bare NF, so on a
mixed-arch host (e.g. dGPU gfx1100 + Strix iGPU gfx1151) where
rocminfo emits each gfx token twice (Name: field + ISA triple),
HIP_VISIBLE_DEVICES=1 indexed vals[1] = the second gfx1100 occurrence
instead of gfx1151, triggering the Strix routing on the wrong GPU.

Add !seen[$0]++ to the indexer awk so duplicate tokens from the same
GPU collapse to one entry before the HIP_VISIBLE_DEVICES index is
applied -- matching exactly what the Python side does with dict.fromkeys()
in _detect_amd_gfx_codes(). The comment above the block ("skip
duplicates") already documented this as the intended behaviour.

* fix(studio/install): correct _TOTAL progress count on Windows

base_total += 3 fired for all non-macOS platforms including Windows,
but flash-attn (line 1620) and ROCm torch final (line 1705) are both
guarded by 'not IS_WINDOWS and not IS_MACOS', so on Windows with torch
enabled _TOTAL was 13 while only 11 _progress() calls actually execute.

Split into +1 for the ROCm torch check (all non-macOS) and +2 for the
two Linux-only steps, so Windows gets _TOTAL=11 and Linux gets 14.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(install.ps1): enforce torch>=2.11.0 for gfx120X and Strix on Windows

The AMD arch-specific index (repo.amd.com/rocm/whl/gfx120X-all/ and
gfx1151/) publishes torch wheels from 2.7.1 through 2.11.0. Without a
version floor pip can resolve to torch 2.10.0+rocm7.12 on RDNA 4
(gfx120X) or torch 2.10.0+rocm7.1 on Strix (gfx1151/gfx1150), both of
which have a null-pointer crash in torch._C._grouped_mm (TheRock
issues #5284 / #3284). torch 2.11.0+rocm7.13 contains the fix.

Add $ROCmTorchFloor alongside $ROCmIndexUrl: set to torch>=2.11.0 for
the two affected arch families, null for all others. Wire it into the
uv pip install call so the broken wheels are never selected.

* fix(rocm/windows): address Codex nits - deterministic DLL suffix, CUDA llama.cpp kind, HIP_VISIBLE_DEVICES arch indexing

- install_python_stack.py / worker.py: _detect_bnb_rocm_dll_ver() and the
  inline worker probe now collect ALL libbitsandbytes_rocm*.dll suffixes and
  return max() by numeric value instead of stopping at the first glob hit.
  Filesystem glob order is not guaranteed; this ensures '713' always wins
  over '72' when both variants are present in the wheel.

- setup.ps1 (expectedKind): add 'windows-cuda' branch so NVIDIA hosts are
  not treated as 'windows-cpu'. Previously an existing windows-cuda prebuilt
  was always considered a mismatch on non-ROCm machines, forcing an
  unnecessary re-download on every update.

- setup.ps1 (amd-smi gfx arch): collect ALL gfx tokens from amd-smi list
  output in GPU order and honour HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES
  when selecting which arch to use. On mixed-arch AMD systems where the
  visible GPU is not the first enumerated one, this prevents installing an
  incompatible wheel index. Falls back to index 0 (same as before) when the
  visibility var is unset or is a comma-separated list.

- test_rocm_support.py: add test_picks_highest_suffix_when_multiple_dlls to
  cover the multi-DLL case that was previously untested.

* fix(rocm): misleading amd-smi log, BNB spec consistency, torch ceiling for AMD index

amd.py: split 'returncode != 0 or not stdout' into two separate branches.
Previously, exit-0 with empty output logged 'amd-smi returned code 0' (which
reads as success, not a warning) and incorrectly incremented the circuit-breaker
counter. Now: non-zero exit logs the code and counts toward the limit as before;
empty stdout on exit 0 logs at DEBUG level and does not penalise the counter
(amd-smi --json always emits at least [] on exit 0, so this branch is rare and
is not a tool failure).

main.py: replace spec.origin / os.path.dirname() with
spec.submodule_search_locations to match install_python_stack.py and worker.py.
For normal wheel installs both approaches reach the same directory, but using
submodule_search_locations is the canonical way and handles editable bitsandbytes
installs correctly. Also use max() by numeric suffix (same as the other two sites)
instead of a sort-then-break loop.

install.ps1: add <2.12.0 ceiling to the torch constraint for gfx120X (RDNA 4)
and gfx1151/gfx1150 (Strix). AMD actively publishes new versions on their
per-arch index; without a ceiling, a future 2.12.0+rocmX.Y wheel would be
pulled in automatically before being validated on these architectures. The
ceiling matches the existing Linux install_python_stack.py constraint for the
same arches. Bump both when 2.12.x is confirmed working.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(rocm): torch floor in setup.ps1, torchvision pin for Strix, rocmsdk in _hip_ver_at_least

setup.ps1: add \ (mirrors install.ps1) and derive \
from it. Previously the AMD index install called 'Fast-Install torch torchvision
torchaudio --force-reinstall --index-url \' with no version
constraint, so pip could resolve torch 2.10.0+rocm7.12 for gfx1151/gfx1200 --
the exact broken wheel the PR is meant to avoid. Now gfx120X and Strix enforce
'torch>=2.11.0,<2.12.0', matching install.ps1 and the Linux constraint.

install_python_stack.py: pin torchvision and torchaudio in _strix_override_pkgs.
The Strix Linux override uses --index-url (exclusive, no PyPI fallback); bare
unversioned 'torchvision' and 'torchaudio' could resolve a build from AMD's
index targeting a different torch major, causing ABI/version mismatches at
runtime. Now pinned to '>=0.26.0,<0.27.0' and '>=2.11.0,<2.12.0' respectively,
matching _ROCM_TORCH_CONSTRAINT['rocm7.2'].

worker.py: extend _hip_ver_at_least to handle AMD SDK wheel version strings.
The fallback regex r'rocm(\d+)\.(\d+)' cannot match '2.9.0+rocmsdk20251116'
(no rocmX.Y component), so the function always returned False on SDK/Radeon
wheels -- installing the Python _grouped_mm workaround on wheels that already
have the working HIP kernel. Added a second check: if the version string
contains '+rocmsdk', assume >= 7.13 (the rocmsdk format post-dates the
gfx120X null-kernel fix) and skip the fallback.

* fix(rocm): warn on OOB HIP_VISIBLE_DEVICES, bail on empty numeric_ids mask

- setup.ps1: when HIP/ROCR_VISIBLE_DEVICES names an index beyond the
  detected GPU count, emit a yellow warning and fall back to GPU 0
  instead of silently reading allGfxArches[-1] (wrong arch)
- hardware.py _reconcile_primary_rocm_unified_memory: distinguish
  numeric_ids=None (no env var, use torch ordinal 0) from numeric_ids=[]
  (empty mask / HIP_VISIBLE_DEVICES=-1, no GPU visible); bail out early
  in the empty case to avoid querying torch.device(0) incorrectly

* fix(rocm): gate StubSubpackageFinder on win32 ROCm, add gcnArchName fallbacks

- worker.py _StubSubpackageFinder: the meta_path append was running on
  every platform on every call to run_training_process; moved it inside
  the if _is_win32_rocm: block since stubs are only seeded there and the
  finder is a pure accumulation on Linux/Windows CUDA
- worker.py OOM guard: AMD SDK / Radeon wheels may not populate
  gcnArchName, causing Strix Halo to be misclassified as discrete and
  get the 0.90 cap (12.8 GB OS headroom) instead of 0.80 (25.6 GB);
  now tries gcn_arch_name / arch_name / gfx_arch_name variants first,
  then falls back to device-name matching (890M -> Strix Halo,
  880M -> Strix Point) with a debug log when the fallback fires

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(rocm): pin torchvision/torchaudio in setup.ps1, remove -Unique from arch array

- setup.ps1 ROCm torch install: torchvision and torchaudio were passed
  bare alongside pinned torch>=2.11.0,<2.12.0 for gfx1151/gfx1200 arches.
  AMD publishes packages independently so a future torchvision 0.27 (for
  torch 2.12) on the same arch index would cause pip ResolutionImpossible
  or an ABI-incompatible install. Added torchvisionFloorMap and
  torchaudioFloorMap mirroring install_python_stack.py's strix override
  (torchvision>=0.26.0,<0.27.0, torchaudio>=2.11.0,<2.12.0) and derived
  ROCmVisionSpec/ROCmAudioSpec used in all three Fast-Install call sites.

- setup.ps1 amd-smi arch detection: Select-Object -Unique was collapsing
  same-arch multi-GPU arrays (e.g. two gfx1151 APUs -> 1-element array)
  causing HIP_VISIBLE_DEVICES=1 to trigger a false out-of-range warning
  and fall back to GPU 0 even though the correct GPU would have been at
  index 1. Removed -Unique; added comment noting the positional-index
  assumption and its non-contiguous-GPU limitation.

* fix(rocm): add 8060s/8050s to OOM guard device-name fallback, extract classifier helper

Path 3 of the OOM guard device-name fallback only checked for 890m/880m
(gfx1150 Strix Point SKU names). Strix Halo (gfx1151) ships as Radeon 8060S
(Ryzen AI MAX+ 395) and Radeon 8050S (cut-down SKU) -- neither matches, so
the fallback returned is_unified=False and applied the 0.90 fraction instead
of 0.80, leaving ~12.8 GiB OS headroom on a 128 GiB pool instead of ~25.6 GiB.

Fix: add 8060s and 8050s to the name-match set. Also correct the comment that
mislabelled 890M as a Strix Halo name (it is Strix Point).

Refactor: extract the three-path classifier into _rocm_classify_unified_memory()
so it can be unit-tested directly. Add 31 test cases in test_rocm_oom_guard.py
covering all three paths and the regression case (Radeon 8060S Graphics).

Reported-by: h34v3nzc0dex

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix(rocm): pass explicit dtype on bf16-unsupported hardware (RDNA2)

dtype=None lets unsloth auto-detect the model dtype. On RDNA2 (gfx103x,
e.g. RX 6600) is_bfloat16_supported() incorrectly returns True, so unsloth
picks bf16 and the first bf16 kernel dispatch triggers:

  LLVM ERROR: Cannot select: intrinsic %llvm.amdgcn.fdot2.bf16.bf16

Replace every dtype=None in load_model() with _auto_dtype which resolves
to None when bf16 is supported (all modern NVIDIA + RDNA3+) and
torch.float16 otherwise. This gives RDNA2 users a working float16
training path without touching NVIDIA behaviour at all.

Fixes: https://github.com/unslothai/unsloth/issues/5337

* fix: reduce log noise for expected non-issues on Windows ROCm

Three log lines fired at warning/error level for conditions that are
completely expected on a Windows HIP SDK-only setup:

amd.py
- amd-smi WinError 2 (FileNotFoundError): downgrade warning -> debug.
  amd-smi ships with Adrenalin, not the HIP SDK; absence is normal.
- 'disabling' message: downgrade warning -> info with clearer text
  'not available (not installed; expected on HIP SDK-only systems);
  GPU VRAM polling disabled'

hardware.py
- torch.distributed.Store missing: downgrade warning -> debug.
  The distributed stub added in this PR intentionally omits Store; the
  attention-impl fallback to eager is expected and non-actionable.

worker.py
- causal-conv1d: add early Windows exit (info) in both
  _ensure_causal_conv1d_fast_path and _causal_conv1d_install hook;
  no cp313/win_amd64 wheel exists, so the install always fails.
- FLA: add early Windows exit (info) in
  _ensure_flash_linear_attention_unconditional; triton dependency has
  no cp313/win_amd64 wheel.
- Defense-in-depth: _install_package_wheel_first non-HIP PyPI failure
  logs info+debug on Windows instead of error; FLA failure logs
  info+debug on Windows instead of warning.

* [AMD] FIx installation of bitsandbytes when it's from .dev and skip rebuilding llama.cpp if we build it manually.

* fix: use force_pip for Windows ROCm bitsandbytes prebuilt wheel install

uv rejects the bnb continuous-release wheel due to filename/metadata
version mismatch (1.33.7.preview vs 0.50.0.dev0). Switch to force_pip=True
(pip bypass) instead of the UV_SKIP_WHEEL_FILENAME_CHECK env var workaround
-- cleaner and consistent with how the Linux path handles it.

BNB_ROCM_VERSION is still set post-install to the detected DLL suffix so
the worker subprocess loads the correct libbitsandbytes_rocm{VER}.dll even
when torch.version.hip reports a newer HIP version than the wheel ships.

* fix: three small correctness fixes found in PR review

- _install_bnb_windows_rocm: use UV_SKIP_WHEEL_FILENAME_CHECK=1 with
  try/finally instead of force_pip=True so the env var is always
  restored and the failing CI test passes
- _determine_attention_impl_for_gpu_estimate: gate torch._C distributed
  stubs on IS_ROCM so Windows CUDA users keep the real extension
- install.ps1 amd-smi fallback: collect all gfx tokens and index by
  HIP_VISIBLE_DEVICES, matching the hipinfo path on multi-GPU hosts

* fix: stub torchao in export subprocess on Windows ROCm

On Windows, the ROCm build of PyTorch ships without the distributed
C extension (torch._C._distributed_c10d). torchao, which is pulled in
transitively by transformers.quantizers at import time, walks into
torch.distributed._functional_collectives -> distributed_c10d and
crashes with:

  No module named 'torch._C._distributed_c10d'; 'torch._C' is not a package

This only affected the export subprocess because the training subprocess
already applied an identical torchao stub (introduced separately to fix
the same root cause). The export subprocess had no such guard and died
during 'Importing Unsloth...' before any model loading could happen.

Fix: apply the same _StubSubpackageFinder / torchao stub pattern to the
export subprocess entry point, gated on Windows ROCm detection, before
any import of transformers or unsloth_zoo.

Root cause tracked in ROCm/TheRock#3284 (libuv / torch.distributed
missing on Windows ROCm builds).

Ref: https://github.com/ROCm/TheRock/issues/3284

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* install.sh, setup.sh: add GPU arch step logging to match PS1 scripts

Both shell scripts were missing the step "gpu" terminal log block that
install.ps1 and setup.ps1 emit. This adds equivalent output: GPU label
with gfx arch (e.g. "AMD ROCm (gfx1151)"), ROCm root path, hipconfig
version, and marketing name substep. Includes the same gfx arch detection
chain (rocminfo → amd-smi list → amd-smi static --asic), UNSLOTH_ROCM_GFX_ARCH
env override, and name-based arch inference table (Strix Halo/Point, RDNA 3/4)
as the PS1 versions. install.sh also replaces bare echo blocks for the AMD
ROCm and CPU-only cases with formatted substep output.

* Fix BNB_ROCM_VERSION gate, ROCm GPU mask preference, APU unified memory and Release build for PR #5301

- main.py: gate BNB_ROCM_VERSION on the rocm bnb DLL or HIP_PATH/ROCM_PATH instead of importing torch on every Windows host
- hardware.py: prefer HIP/ROCR visible-device masks only on ROCm hosts so a stale mask cannot override CUDA_VISIBLE_DEVICES on NVIDIA
- llama_cpp.py: set GGML_CUDA_ENABLE_UNIFIED_MEMORY=1 only for unified-memory APUs (gfx1150/gfx1151)
- setup.sh: pass -DCMAKE_BUILD_TYPE=Release for the HIP source build
- add test_amd_apu_unified_memory.py

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: guard recompile_limit + fix AMD VRAM monitor fallback

trainer.py: torch._dynamo.config.recompile_limit does not exist in
some ROCm torch builds (e.g. pytorch.org/whl/rocm6.2 wheels). Guard
the assignment so training doesn't crash on RDNA2/RDNA3.

hardware.py: when amd-smi/nvidia-smi is unavailable or returns no
usable data (HIP SDK-only Windows, Docker, unexpected JSON format),
the existing fallback used torch.cuda.memory_allocated() which is
process-specific and reads near-zero even with a fully loaded model.
Switch to torch.cuda.mem_get_info() via _torch_get_per_device_info()
which reports system-wide VRAM occupancy so the GPU monitor shows
real usage on all AMD systems without requiring amd-smi.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: Windows VRAM monitor via Performance Counter API

When amd-smi/nvidia-smi is unavailable on Windows, query dedicated GPU
VRAM via Windows Performance Counters (same source as Task Manager).
This gives system-wide cross-process usage, fixing the near-zero reading
caused by torch.cuda.mem_get_info only seeing the Studio server process.

Linux fallback path unchanged (mem_get_info is system-wide on ROCm).

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: rename to _rocm_windows_perf_counter_vram_gb, scope to IS_ROCM

Function is AMD ROCm specific — amd-smi absent on Windows when only the
HIP SDK is installed. Scoped to IS_ROCM so NVIDIA Windows path is
untouched (nvidia-smi handles that case).

* fix: AMD VRAM monitor — Linux DRM sysfs + Windows perf counter

Linux: read /sys/class/drm/card*/device/mem_info_vram_used|total for
system-wide GPU memory across all processes. No tools required, always
present on Linux AMD systems.

Windows: Windows Performance Counter API (already added).

Both paths are gated on IS_ROCM and only fire when amd-smi is absent.
torch mem_get_info remains as last resort (process-local).

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: AMD GPU monitor — utilization, temperature, and power for Windows and Linux fallback paths

- Windows: GPU utilization via \GPU Engine(*engtype_3D*)\Utilization Percentage perf counter
- Windows: temperature and power via ADL (atiadlxx.dll, ships with Adrenalin)
- Linux: GPU utilization via DRM sysfs gpu_busy_percent
- Linux: temperature via hwmon temp1_input (millidegrees C)
- Linux: power via hwmon power1_average / power1_input (microwatts)

All paths are no-op fallbacks (None) when the source is unavailable.
Mirrors what nvidia-smi provides on the CUDA path.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix: remove ADL ctypes — does not support AMD iGPU (Strix Halo)

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
Co-authored-by: Erland366 <erland.pg366@gmail.com>
Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
2026-05-29 22:29:56 -07:00

3081 lines
142 KiB
PowerShell

#Requires -Version 5.1
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
<#
.SYNOPSIS
Full environment setup for Unsloth Studio on Windows (bundled version).
.DESCRIPTION
Always installs Node.js if needed. When running from pip install:
skips frontend build (already bundled). When running from git repo:
full setup including frontend build.
Supports NVIDIA GPU (full training + inference) and CPU-only (GGUF chat mode).
.NOTES
Default output is minimal (step/substep), aligned with studio/setup.sh.
FULL / LEGACY LOGGING (defensible audit trail, detailed multi-line output):
unsloth studio setup --verbose
Or: $env:UNSLOTH_VERBOSE='1'; powershell -File .\studio\setup.ps1
Or: .\setup.ps1 --verbose
#>
$ErrorActionPreference = "Stop"
$ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path
$PackageDir = Split-Path -Parent $ScriptDir
# --------------------------------------------------------------------------
# Maintainer-editable defaults
# Change these in the GitHub-hosted script so users get updated defaults.
# User env vars always override these baked-in values.
# --------------------------------------------------------------------------
# Prefer "latest" over "master" -- "master" bypasses the prebuilt resolver
# (no matching GitHub release), forces a source build, and causes HTTP 422
# errors. Only use "master" temporarily when the latest release is missing
# support for a new model architecture.
$DefaultLlamaPrForce = ""
$DefaultLlamaSource = "https://github.com/ggml-org/llama.cpp"
$DefaultLlamaTag = "latest"
$DefaultLlamaForceCompileRef = "master"
# 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') {
$script:UnslothVerbose = $true
break
}
}
# Propagate to child processes (e.g. install_python_stack.py) so they
# also respect verbose mode. Process-scoped -- does not persist.
if ($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"
$OxcValidatorDir = Join-Path $ScriptDir "backend\core\data_recipe\oxc-validator"
$IsPipInstall = -not (Test-Path $FrontendDir)
# ─────────────────────────────────────────────
# Helper functions
# ─────────────────────────────────────────────
# Reload ALL environment variables from registry.
# Picks up changes made by installers (winget, msi, etc.) including
# Path, CUDA_PATH, CUDA_PATH_V*, and any other vars they set.
function Refresh-Environment {
foreach ($level in @('Machine', 'User')) {
$vars = [System.Environment]::GetEnvironmentVariables($level)
foreach ($key in $vars.Keys) {
if ($key -eq 'Path') { continue }
Set-Item -Path "Env:$key" -Value $vars[$key] -ErrorAction SilentlyContinue
}
}
$machinePath = [System.Environment]::GetEnvironmentVariable('Path', 'Machine')
$userPath = [System.Environment]::GetEnvironmentVariable('Path', 'User')
# Merge: venv Scripts (if active) > Machine > User > current $env:Path. Dedup raw+expanded.
$venvScripts = if ($env:VIRTUAL_ENV) { Join-Path $env:VIRTUAL_ENV 'Scripts' } else { $null }
$sources = @()
if ($venvScripts) { $sources += $venvScripts }
$sources += @($machinePath, $userPath, $env:Path)
$merged = ($sources | Where-Object { $_ }) -join ';'
$seen = @{}
$unique = New-Object System.Collections.Generic.List[string]
foreach ($p in $merged -split ";") {
$rawKey = $p.Trim().Trim('"').TrimEnd("\").ToLowerInvariant()
$expKey = [Environment]::ExpandEnvironmentVariables($p).Trim().Trim('"').TrimEnd("\").ToLowerInvariant()
if ($rawKey -and -not $seen.ContainsKey($rawKey) -and -not $seen.ContainsKey($expKey)) {
$seen[$rawKey] = $true
if ($expKey -and $expKey -ne $rawKey) { $seen[$expKey] = $true }
$unique.Add($p)
}
}
$env:Path = $unique -join ";"
}
# ── Helper: safely add a directory to the persistent User PATH ──
# Direct registry access preserves REG_EXPAND_SZ (avoids dotnet/runtime#1442).
# Append (default) keeps existing tools first; Prepend for must-win entries.
function Add-ToUserPath {
param(
[Parameter(Mandatory = $true)][string]$Directory,
[ValidateSet('Append','Prepend')]
[string]$Position = 'Append'
)
try {
$regKey = [Microsoft.Win32.Registry]::CurrentUser.CreateSubKey('Environment')
try {
$rawPath = $regKey.GetValue('Path', '', [Microsoft.Win32.RegistryValueOptions]::DoNotExpandEnvironmentNames)
[string[]]$entries = if ($rawPath) { $rawPath -split ';' } else { @() } # string[] prevents scalar collapse
$normalDir = $Directory.Trim().Trim('"').TrimEnd('\').ToLowerInvariant()
$expNormalDir = [Environment]::ExpandEnvironmentVariables($Directory).Trim().Trim('"').TrimEnd('\').ToLowerInvariant()
$kept = New-Object System.Collections.Generic.List[string]
$matchIndices = New-Object System.Collections.Generic.List[int]
for ($i = 0; $i -lt $entries.Count; $i++) {
$stripped = $entries[$i].Trim().Trim('"')
$rawNorm = $stripped.TrimEnd('\').ToLowerInvariant()
$expNorm = [Environment]::ExpandEnvironmentVariables($stripped).TrimEnd('\').ToLowerInvariant()
$isMatch = ($rawNorm -and ($rawNorm -eq $normalDir -or $rawNorm -eq $expNormalDir)) -or
($expNorm -and ($expNorm -eq $normalDir -or $expNorm -eq $expNormalDir))
if ($isMatch) {
$matchIndices.Add($i)
continue
}
$kept.Add($entries[$i])
}
$alreadyPresent = $matchIndices.Count -gt 0
if ($alreadyPresent -and $Position -eq 'Append') { # Append: idempotent no-op
return $false
}
if ($alreadyPresent -and $Position -eq 'Prepend' -and # Prepend: no-op if already at front
$matchIndices.Count -eq 1 -and $matchIndices[0] -eq 0) {
return $false
}
# One-time backup under HKCU\Software\Unsloth\PathBackup
if ($rawPath) {
try {
$backupKey = [Microsoft.Win32.Registry]::CurrentUser.CreateSubKey('Software\Unsloth')
try {
$existingBackup = $backupKey.GetValue('PathBackup', $null)
if (-not $existingBackup) {
$backupKey.SetValue('PathBackup', $rawPath, [Microsoft.Win32.RegistryValueKind]::ExpandString)
}
} finally {
$backupKey.Close()
}
} catch { }
}
if (-not $rawPath) {
Write-Host "[WARN] User PATH is empty - initializing with $Directory" -ForegroundColor Yellow
}
$newPath = if ($rawPath) {
if ($Position -eq 'Prepend') {
(@($Directory) + $kept) -join ';'
} else {
($kept + @($Directory)) -join ';'
}
} else {
$Directory
}
if ($newPath -ceq $rawPath) { # no actual change
return $false
}
$regKey.SetValue('Path', $newPath, [Microsoft.Win32.RegistryValueKind]::ExpandString)
# Broadcast WM_SETTINGCHANGE via dummy env-var roundtrip.
# [NullString]::Value avoids PS 7.5+/.NET 9 $null-to-"" coercion.
try {
$d = "UnslothPathRefresh_$([guid]::NewGuid().ToString('N').Substring(0,8))"
[Environment]::SetEnvironmentVariable($d, '1', 'User')
[Environment]::SetEnvironmentVariable($d, [NullString]::Value, 'User')
} catch { }
return $true
} finally {
$regKey.Close()
}
} catch {
Write-Host "[WARN] Could not update User PATH: $($_.Exception.Message)" -ForegroundColor Yellow
return $false
}
}
# PowerShell 5.1 compatibility helper: avoid relying on New-TemporaryFile.
function New-UnslothTemporaryFile {
$tempPath = [System.IO.Path]::GetTempFileName()
return Get-Item -LiteralPath $tempPath
}
function Get-InstalledLlamaPrebuiltRelease {
param([string]$InstallDir)
$metadataPath = Join-Path $InstallDir "UNSLOTH_PREBUILT_INFO.json"
if (-not (Test-Path $metadataPath)) {
return $null
}
try {
$payload = Get-Content $metadataPath -Raw | ConvertFrom-Json
} catch {
return $null
}
if (-not $payload.published_repo -or -not $payload.release_tag) {
return $null
}
$message = "installed release: $($payload.published_repo)@$($payload.release_tag)"
if ($payload.tag -and $payload.tag -ne $payload.release_tag) {
$message += " (tag $($payload.tag))"
}
return $message
}
# Find nvcc on PATH, CUDA_PATH, or standard toolkit dirs.
# Returns the path to nvcc.exe, or $null if not found.
function Find-Nvcc {
param([string]$MaxVersion = "")
# If MaxVersion is set, we need to find a toolkit <= that version.
# CUDA toolkits install side-by-side under C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vX.Y\
$toolkitBase = 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA'
if ($MaxVersion -and (Test-Path $toolkitBase)) {
$drMajor = [int]$MaxVersion.Split('.')[0]
$drMinor = [int]$MaxVersion.Split('.')[1]
# Get all installed CUDA dirs, sorted descending (highest first)
$cudaDirs = Get-ChildItem -Directory $toolkitBase | Where-Object {
$_.Name -match '^v(\d+)\.(\d+)'
} | Sort-Object { [version]($_.Name -replace '^v','') } -Descending
foreach ($dir in $cudaDirs) {
if ($dir.Name -match '^v(\d+)\.(\d+)') {
$tkMajor = [int]$Matches[1]; $tkMinor = [int]$Matches[2]
$compatible = ($tkMajor -lt $drMajor) -or ($tkMajor -eq $drMajor -and $tkMinor -le $drMinor)
if ($compatible) {
$nvcc = Join-Path $dir.FullName 'bin\nvcc.exe'
if (Test-Path $nvcc) {
return $nvcc
}
}
}
}
# No compatible side-by-side version found
return $null
}
# Fallback: no version constraint — pick latest or whatever is available
# 1. Check nvcc on PATH
$cmd = Get-Command nvcc -ErrorAction SilentlyContinue
if ($cmd) { return $cmd.Source }
# 2. Check CUDA_PATH env var
$cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'Process')
if (-not $cudaRoot) { $cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'Machine') }
if (-not $cudaRoot) { $cudaRoot = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'User') }
if ($cudaRoot -and (Test-Path (Join-Path $cudaRoot 'bin\nvcc.exe'))) {
return (Join-Path $cudaRoot 'bin\nvcc.exe')
}
# 3. Scan standard toolkit directory
if (Test-Path $toolkitBase) {
$latest = Get-ChildItem -Directory $toolkitBase | Where-Object {
$_.Name -match '^v(\d+)\.(\d+)'
} | Sort-Object { [version]($_.Name -replace '^v','') } -Descending | Select-Object -First 1
if ($latest -and (Test-Path (Join-Path $latest.FullName 'bin\nvcc.exe'))) {
return (Join-Path $latest.FullName 'bin\nvcc.exe')
}
}
return $null
}
# Detect CUDA Compute Capability via nvidia-smi.
# Returns e.g. "80" for A100 (8.0), "89" for RTX 4090 (8.9), etc.
# Returns $null if detection fails.
function Get-CudaComputeCapability {
# Use the resolved absolute path ($NvidiaSmiExe) to survive Refresh-Environment
$smiExe = if ($script:NvidiaSmiExe) { $script:NvidiaSmiExe } else {
$cmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
if ($cmd) { $cmd.Source } else { $null }
}
if (-not $smiExe) { return $null }
try {
$raw = & $smiExe --query-gpu=compute_cap --format=csv,noheader 2>$null
if ($LASTEXITCODE -ne 0 -or -not $raw) { return $null }
# nvidia-smi may return multiple GPUs; take the first one
$cap = ($raw -split "`n")[0].Trim()
if ($cap -match '^(\d+)\.(\d+)$') {
$major = $Matches[1]
$minor = $Matches[2]
return "$major$minor"
}
} catch { }
return $null
}
# Check if an nvcc binary supports a given sm_ architecture.
# Uses `nvcc --list-gpu-code` which outputs sm_* tokens (--list-gpu-arch
# outputs compute_* tokens instead). Available since CUDA 11.6.
# Returns $false if the flag isn't supported (old toolkit) — safer to reject
# and fall back to scanning/PTX than to assume support and fail later.
function Test-NvccArchSupport {
param([string]$NvccExe, [string]$Arch)
try {
$listCode = & $NvccExe --list-gpu-code 2>&1 | Out-String
if ($LASTEXITCODE -ne 0) { return $false }
return ($listCode -match "sm_$Arch")
} catch {
return $false
}
}
# Given an nvcc binary, return the highest sm_ architecture it supports.
# Returns e.g. "90" for CUDA 12.4. Returns $null if detection fails.
function Get-NvccMaxArch {
param([string]$NvccExe)
try {
$listCode = & $NvccExe --list-gpu-code 2>&1 | Out-String
if ($LASTEXITCODE -ne 0) { return $null }
$arches = @()
foreach ($line in $listCode -split "`n") {
if ($line.Trim() -match '^sm_(\d+)') {
$arches += [int]$Matches[1]
}
}
if ($arches.Count -gt 0) {
return ($arches | Sort-Object | Select-Object -Last 1).ToString()
}
} catch { }
return $null
}
# Detect driver's max CUDA version from nvidia-smi and return the highest
# compatible PyTorch CUDA index tag (e.g. "cu128").
# PyTorch on Windows ships CPU-only by default from PyPI; CUDA wheels live at
# https://download.pytorch.org/whl/<tag>. The tag must not exceed the driver's
# capability: e.g. driver "CUDA Version: 12.9" → cu128 (not cu130).
function Get-PytorchCudaTag {
$smiExe = if ($script:NvidiaSmiExe) { $script:NvidiaSmiExe } else {
$cmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
if ($cmd) { $cmd.Source } else { $null }
}
if (-not $smiExe) { return "cu126" }
try {
# 2>&1 | Out-String merges stderr into stdout then converts to a single
# string. Plain 2>$null doesn't fully suppress stderr in PS 5.1 --
# ErrorRecord objects leak into $output and break the -match.
$output = & $smiExe 2>&1 | Out-String
# Newer NVIDIA drivers (e.g. 610.x on Windows) print
# "CUDA UMD Version: X.Y" instead of the legacy "CUDA Version: X.Y".
# Accept both spellings so we don't fall through to the cu126 default.
if ($output -match 'CUDA(?: UMD)? Version:\s+(\d+)\.(\d+)') {
$major = [int]$Matches[1]
$minor = [int]$Matches[2]
# PyTorch 2.10 offers: cu124, cu126, cu128, cu130
if ($major -ge 13) { return "cu130" }
if ($major -eq 12 -and $minor -ge 8) { return "cu128" }
if ($major -eq 12 -and $minor -ge 6) { return "cu126" }
if ($major -ge 12) { return "cu124" }
if ($major -ge 11) { return "cu118" }
return "cpu"
}
} catch { }
return "cu126"
}
# Find Visual Studio Build Tools for cmake -G flag.
# Strategy: (1) vswhere, (2) scan filesystem (handles broken vswhere registration).
# Returns @{ Generator = "Visual Studio 17 2022"; InstallPath = "C:\..."; Source = "..." } or $null.
function Find-VsBuildTools {
$map = @{ '2022' = '17'; '2019' = '16'; '2017' = '15' }
# --- Try vswhere first (works when VS is properly registered) ---
$vsw = "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe"
if (Test-Path $vsw) {
$info = & $vsw -latest -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property catalog_productLineVersion 2>$null
$path = & $vsw -latest -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath 2>$null
if ($info -and $path) {
$y = $info.Trim()
$n = $map[$y]
if ($n) {
return @{ Generator = "Visual Studio $n $y"; InstallPath = $path.Trim(); Source = 'vswhere' }
}
}
}
# --- Scan filesystem (handles broken vswhere registration after winget cycles) ---
$roots = @($env:ProgramFiles, ${env:ProgramFiles(x86)})
$editions = @('BuildTools', 'Community', 'Professional', 'Enterprise')
$years = @('2022', '2019', '2017')
foreach ($y in $years) {
foreach ($r in $roots) {
foreach ($ed in $editions) {
$candidate = Join-Path $r "Microsoft Visual Studio\$y\$ed"
if (Test-Path $candidate) {
$vcDir = Join-Path $candidate "VC\Tools\MSVC"
if (Test-Path $vcDir) {
$cl = Get-ChildItem -Path $vcDir -Filter "cl.exe" -Recurse -ErrorAction SilentlyContinue | Select-Object -First 1
if ($cl) {
$n = $map[$y]
if ($n) {
return @{ Generator = "Visual Studio $n $y"; InstallPath = $candidate; Source = "filesystem ($ed)"; ClExe = $cl.FullName }
}
}
}
}
}
}
}
return $null
}
# ─────────────────────────────────────────────
# Output style (aligned with studio/setup.sh: step / substep)
# ─────────────────────────────────────────────
$Rule = [string]::new([char]0x2500, 52)
function Enable-StudioVirtualTerminal {
if ($env:NO_COLOR) { return $false }
try {
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" }
}
}
function Write-SetupVerboseDetail {
param(
[Parameter(Mandatory = $true)][string]$Message,
[string]$Color = "Gray"
)
if (-not $script:UnslothVerbose) { return }
if ($script:StudioVtOk -and -not $env:NO_COLOR) {
$ansi = switch ($Color) {
'Green' { (Get-StudioAnsi Ok) }
'Gray' { (Get-StudioAnsi Dim) }
'DarkGray' { (Get-StudioAnsi Dim) }
'Yellow' { (Get-StudioAnsi Warn) }
'Cyan' { (Get-StudioAnsi Title) }
'Red' { (Get-StudioAnsi Err) }
default { (Get-StudioAnsi Dim) }
}
Write-Host ($ansi + $Message + (Get-StudioAnsi Reset))
} else {
$fc = switch ($Color) {
'Green' { 'DarkGreen' }
'Gray' { 'DarkGray' }
'Cyan' { 'Green' }
default { $Color }
}
Write-Host $Message -ForegroundColor $fc
}
}
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
}
}
# Mirror the plain (no ANSI) form of step/substep messages to the
# OS-level stdout handle when a parent is consuming our stdout via
# a pipe (CI `tee`, Python subprocess.PIPE, CREATE_NO_WINDOW grandchild).
# Write-Host on PS 5.1 routes through $Host.UI / the Information
# stream, neither of which propagates reliably across the
# install.ps1 -> unsloth.exe -> python -> powershell.exe ->
# setup.ps1 process chain. [Console]::Out always lands on the OS
# stdout file handle. Gated on IsOutputRedirected so the
# interactive-console path keeps the colorized Write-Host output
# only (no double-print).
function Write-StudioStdoutMirror {
param([Parameter(Mandatory = $true)][string]$Line)
try {
if ([Console]::IsOutputRedirected) {
[Console]::Out.WriteLine($Line)
[Console]::Out.Flush()
}
} catch {}
}
function step {
param(
[Parameter(Mandatory = $true)][string]$Label,
[Parameter(Mandatory = $true)][string]$Value,
[string]$Color = "Green"
)
$padded = if ($Label.Length -ge 15) { $Label.Substring(0, 15) } else { $Label.PadRight(15) }
if ($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 }
}
Write-Host (" {0}{1}{2}{3}{4}{2}" -f $dim, $padded, $rst, $val, $Value)
} else {
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
}
Write-StudioStdoutMirror (" {0}{1}" -f $padded, $Value)
}
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) }
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' }
default { 'DarkGray' }
}
Write-Host (" {0,-15}{1}" -f "", $Message) -ForegroundColor $fc
}
Write-StudioStdoutMirror (" {0,-15}{1}" -f "", $Message)
}
# ─────────────────────────────────────────────
# Banner
# ─────────────────────────────────────────────
Write-Host ""
if ($script:StudioVtOk -and -not $env:NO_COLOR) {
Write-Host (" " + (Get-StudioAnsi Title) + [char]::ConvertFromUtf32(0x1F9A5) + " Unsloth Studio Setup" + (Get-StudioAnsi Reset))
Write-Host (" {0}{1}{2}" -f (Get-StudioAnsi Dim), $Rule, (Get-StudioAnsi Reset))
} else {
Write-Host (" " + [char]::ConvertFromUtf32(0x1F9A5) + " Unsloth Studio Setup") -ForegroundColor Green
Write-Host " $Rule" -ForegroundColor DarkGray
}
# Back up User PATH under HKCU\Software\Unsloth before any modifications.
try {
$envKey = [Microsoft.Win32.Registry]::CurrentUser.OpenSubKey('Environment', $false)
if ($envKey) {
try {
$rawPath = $envKey.GetValue('Path', '', [Microsoft.Win32.RegistryValueOptions]::DoNotExpandEnvironmentNames)
} finally {
$envKey.Close()
}
if ($rawPath) {
$backupKey = [Microsoft.Win32.Registry]::CurrentUser.CreateSubKey('Software\Unsloth')
try {
$existingBackup = $backupKey.GetValue('PathBackup', $null)
if (-not $existingBackup) {
$backupKey.SetValue('PathBackup', $rawPath, [Microsoft.Win32.RegistryValueKind]::ExpandString)
}
} finally {
$backupKey.Close()
}
}
}
} catch {
Write-Host "[DEBUG] Could not back up User PATH: $($_.Exception.Message)" -ForegroundColor DarkGray
}
# ==========================================================================
# PHASE 1: System-level prerequisites (winget installs, env vars)
# All heavy system tool installs happen here BEFORE touching Python.
# ==========================================================================
# ============================================
# 1a. GPU detection
# ============================================
$HasNvidiaSmi = $false
$NvidiaSmiExe = $null # Absolute path -- survives Refresh-Environment
try {
$nvSmiCmd = Get-Command nvidia-smi -ErrorAction SilentlyContinue
if ($nvSmiCmd) {
& $nvSmiCmd.Source *> $null
if ($LASTEXITCODE -eq 0) {
$HasNvidiaSmi = $true
$NvidiaSmiExe = $nvSmiCmd.Source
}
}
} catch {}
# Fallback: nvidia-smi may not be on PATH even though a GPU + driver exist.
# Check the default install location and the Windows driver store.
if (-not $HasNvidiaSmi) {
$nvSmiDefaults = @(
"$env:ProgramFiles\NVIDIA Corporation\NVSMI\nvidia-smi.exe",
"$env:SystemRoot\System32\nvidia-smi.exe"
)
foreach ($p in $nvSmiDefaults) {
if (Test-Path $p) {
try {
& $p *> $null
if ($LASTEXITCODE -eq 0) {
$HasNvidiaSmi = $true
$NvidiaSmiExe = $p
Write-Host " Found nvidia-smi at $(Split-Path $p -Parent)" -ForegroundColor Gray
break
}
} catch {}
}
}
}
# ── AMD ROCm detection (Windows): probe hipinfo/amd-smi for actual GPU ──
$HasROCm = $false
$HipSdkInstalled = $false # HIP SDK binary found (independent of device accessibility)
$ROCmGpuLabel = $null
$script:ROCmGfxArch = $null
if (-not $HasNvidiaSmi) {
# hipinfo: PATH first, then HIP_PATH/ROCM_PATH bin fallback (mirrors NVIDIA smi path resolution).
# AMD HIP SDK sets HIP_PATH but may not add the bin dir to PATH depending on install type.
$hipinfoExe = Get-Command hipinfo -ErrorAction SilentlyContinue
if (-not $hipinfoExe) {
$hipRoot = if ($env:HIP_PATH) { $env:HIP_PATH } elseif ($env:ROCM_PATH) { $env:ROCM_PATH } else { $null }
$hipEnvLabel = if ($env:HIP_PATH) { "HIP_PATH" } else { "ROCM_PATH" }
if ($hipRoot) {
$hipinfoCandidate = Join-Path $hipRoot "bin\hipinfo.exe"
if (Test-Path $hipinfoCandidate) {
substep "[WARN] hipinfo not on PATH -- located via ${hipEnvLabel}: $hipinfoCandidate" "Yellow"
substep " Add '$(Join-Path $hipRoot 'bin')' to your PATH to suppress this warning" "Yellow"
substep " Quick fix: [Environment]::SetEnvironmentVariable('PATH',`$env:PATH+';$(Join-Path $hipRoot 'bin')','User')" "Yellow"
$hipinfoExe = [PSCustomObject]@{ Source = $hipinfoCandidate }
} else {
substep "[WARN] ${hipEnvLabel}=$hipRoot is set but hipinfo.exe not found at $hipinfoCandidate" "Yellow"
substep " HIP SDK install may be incomplete -- re-install from:" "Yellow"
substep " https://rocm.docs.amd.com/en/latest/deploy/windows/index.html" "Yellow"
}
}
}
if ($hipinfoExe) {
$HipSdkInstalled = $true # binary found → SDK is installed regardless of device state
try {
$hipOut = & $hipinfoExe.Source 2>&1 | Out-String
if ($LASTEXITCODE -eq 0 -and $hipOut -match "(?i)gcnArchName") {
$HasROCm = $true
$_hipAllArches = @([regex]::Matches($hipOut, "(?im)^\s*gcnArchName\s*:\s*(\S+)") | ForEach-Object { ($_.Groups[1].Value -split ':')[0].Trim().ToLower() })
$_hipVisIdx = if ($env:HIP_VISIBLE_DEVICES -match '^\d') { [int]($env:HIP_VISIBLE_DEVICES -split ',')[0] } elseif ($env:ROCR_VISIBLE_DEVICES -match '^\d') { [int]($env:ROCR_VISIBLE_DEVICES -split ',')[0] } else { 0 }
if ($_hipAllArches.Count -gt 0) {
$script:ROCmGfxArch = if ($_hipVisIdx -lt $_hipAllArches.Count) { $_hipAllArches[$_hipVisIdx] } else { $_hipAllArches[0] }
$ROCmGpuLabel = "AMD ROCm ($script:ROCmGfxArch)"
} else {
$ROCmGpuLabel = "AMD ROCm"
}
} elseif ($LASTEXITCODE -ne 0) {
# hipinfo ran but returned a HIP runtime error (e.g. "no ROCm-capable device detected")
$firstLine = ($hipOut -split '\r?\n' | Where-Object { $_.Trim() } | Select-Object -First 1)
substep "[WARN] hipinfo returned a HIP runtime error (exit $LASTEXITCODE)" "Yellow"
substep " $firstLine" "Yellow"
substep " Ensure ROCm drivers are installed: https://rocm.docs.amd.com/en/latest/deploy/windows/index.html" "Yellow"
}
} catch {}
}
# amd-smi fallback: HIP runtime present but hipinfo unavailable (no full HIP SDK).
# Confirms GPU visibility via 'list', then attempts 'static --asic' to extract
# the gfx arch that hipinfo would have provided. Critical for Strix Halo
# (gfx1151) and other iGPUs where only the HIP runtime is installed.
if (-not $HasROCm) {
$amdSmiExe = Get-Command "amd-smi" -ErrorAction SilentlyContinue
if ($amdSmiExe) {
try {
$smiOut = & $amdSmiExe.Source list 2>&1 | Out-String
if ($LASTEXITCODE -eq 0 -and $smiOut -match "(?im)^GPU\s*[:\[]\s*\d") {
$HasROCm = $true
# Attempt 1: newer amd-smi versions embed the gfx arch in list output.
# Collect ALL gfx tokens in output order so that on mixed-arch systems
# we can honour HIP_VISIBLE_DEVICES / ROCR_VISIBLE_DEVICES and pick the
# arch for the *runtime-visible* GPU rather than always the first one.
# Do NOT deduplicate: a dual same-arch system (e.g. two gfx1151 APUs)
# must produce a 2-element array so HIP_VISIBLE_DEVICES=1 selects the
# second GPU rather than triggering a false out-of-range warning.
# Note: this mapping assumes amd-smi lists GPUs in the same order as
# HIP enumerates them (both follow PCI bus order in practice); it may
# give the wrong arch when GPU indices are non-contiguous (very rare).
$allGfxArches = @([regex]::Matches($smiOut, '(?i)\b(gfx\d+[a-z]?)\b') |
ForEach-Object { $_.Groups[1].Value.ToLower() })
if ($allGfxArches.Count -gt 0) {
# Resolve which GPU index is runtime-visible. When a single
# integer index is set, use it; fall back to index 0 otherwise
# (comma-separated lists or unset → first GPU, same as before).
$visGpu = if ($env:HIP_VISIBLE_DEVICES) { $env:HIP_VISIBLE_DEVICES }
elseif ($env:ROCR_VISIBLE_DEVICES) { $env:ROCR_VISIBLE_DEVICES }
else { $null }
$gpuIdx = 0
if ($visGpu -match '^\s*(\d+)\s*$') { $gpuIdx = [int]$Matches[1] }
if ($gpuIdx -ge $allGfxArches.Count) {
substep "[WARN] HIP/ROCR_VISIBLE_DEVICES index $gpuIdx is out of range ($($allGfxArches.Count) GPU(s) detected); defaulting to GPU 0 for arch selection" "Yellow"
$gpuIdx = 0
}
$script:ROCmGfxArch = $allGfxArches[$gpuIdx]
$ROCmGpuLabel = "AMD ROCm ($script:ROCmGfxArch)"
} else {
# Attempt 2: 'static --asic' exposes ASIC details on ROCm 6+,
# including the GFX target needed for wheel index selection.
$smiAsicOut = ""
try { $smiAsicOut = & $amdSmiExe.Source static --asic 2>&1 | Out-String } catch {}
if ($smiAsicOut -match "(?i)\b(gfx\d+[a-z]?)\b") {
$script:ROCmGfxArch = $Matches[1].ToLower()
$ROCmGpuLabel = "AMD ROCm ($script:ROCmGfxArch)"
} elseif ($smiAsicOut -match "(?im)Market.?Name\s*[:\|]\s*([^\r\n]+)") {
$ROCmGpuLabel = "AMD ROCm ($($Matches[1].Trim()))"
} else {
$ROCmGpuLabel = "AMD ROCm"
}
}
}
} catch {}
}
}
# WMI fallback: AMD GPU in device list but no HIP SDK → guide the user.
# WMI gives a marketing name (e.g. "AMD Radeon 890M") but never a gfx arch.
# $HasROCm is intentionally NOT set here — we cannot confirm ROCm runtime
# support without hipinfo or amd-smi. The name is saved to $ROCmGpuLabel
# so the name-based inference below can still attempt an arch lookup.
if (-not $HasROCm) {
try {
$wmiGpu = Get-WmiObject Win32_VideoController -ErrorAction SilentlyContinue |
Where-Object { $_.Name -match "AMD|Radeon" } |
Select-Object -First 1
if ($wmiGpu) { $ROCmGpuLabel = $wmiGpu.Name }
} catch {}
}
# ── Arch resolution: env-var override → name inference ──────────────────
# Runs after all probe methods. Covers users whose amd-smi version is too
# old to report the GFX target and who don't have hipinfo (HIP-runtime-only
# installs, common on Strix Halo / iGPU systems).
if ($HasROCm -and -not $script:ROCmGfxArch) {
# 1. Manual override: set UNSLOTH_ROCM_GFX_ARCH=gfx1151 before running.
if ($env:UNSLOTH_ROCM_GFX_ARCH) {
$script:ROCmGfxArch = $env:UNSLOTH_ROCM_GFX_ARCH.Trim().ToLower()
$ROCmGpuLabel = "AMD ROCm ($script:ROCmGfxArch)"
substep "gfx arch from UNSLOTH_ROCM_GFX_ARCH env override: $script:ROCmGfxArch" "Cyan"
}
# 2. Best-effort name → arch lookup from marketing name (amd-smi / WMI).
# Ordered most-specific first; first match wins.
elseif ($ROCmGpuLabel) {
$nameArchTable = @(
@{ P = "9070 XT|9080"; A = "gfx1201" } # RDNA 4
@{ P = "9070|9060"; A = "gfx1200" } # RDNA 4
@{ P = "8060S|890M|Strix Halo|HX 37[05]|HX 38[05]|AI 9 HX"; A = "gfx1151" } # RDNA 3.5 iGPU (Strix Halo / Radeon 8060S retail)
@{ P = "880M|Strix Point|AI 9 36[05]|AI 7 35[05]|AI 5 34[05]"; A = "gfx1150" } # RDNA 3.5 iGPU (Strix Point)
@{ P = "RX 7900|RX 7800|RX 7700(?! S)"; A = "gfx1100" } # RDNA 3 desktop
@{ P = "RX 7600"; A = "gfx1102" } # RDNA 3
@{ P = "780M|760M|740M|Phoenix"; A = "gfx1103" } # RDNA 3 iGPU (Phoenix)
)
foreach ($row in $nameArchTable) {
if ($ROCmGpuLabel -match $row.P) {
$script:ROCmGfxArch = $row.A
$ROCmGpuLabel = "AMD ROCm ($script:ROCmGfxArch)"
substep "gfx arch inferred from GPU name: $script:ROCmGfxArch" "Cyan"
substep "Tip: set UNSLOTH_ROCM_GFX_ARCH=$script:ROCmGfxArch to skip inference next time" "Cyan"
break
}
}
}
}
# Capture ROCm version early for display and wheel selection.
# Run whenever the HIP SDK binary is present, not just when the device is accessible --
# hipconfig --version works even when hipinfo reports no ROCm device (driver issue).
if ($HasROCm -or $HipSdkInstalled) {
$script:ROCmVersion = $null
$hipConfigExe = Get-Command hipconfig -ErrorAction SilentlyContinue
if (-not $hipConfigExe) {
$hipRoot = if ($env:HIP_PATH) { $env:HIP_PATH } elseif ($env:ROCM_PATH) { $env:ROCM_PATH } else { $null }
if ($hipRoot) {
$hipConfigCandidate = Join-Path $hipRoot "bin\hipconfig.exe"
if (Test-Path $hipConfigCandidate) {
$hipConfigEnvLabel = if ($env:HIP_PATH) { "HIP_PATH" } else { "ROCM_PATH" }
substep "[WARN] hipconfig not on PATH -- located via ${hipConfigEnvLabel}: $hipConfigCandidate" "Yellow"
$hipConfigExe = [PSCustomObject]@{ Source = $hipConfigCandidate }
}
}
}
if ($hipConfigExe) {
try {
$hipVerOut = & $hipConfigExe.Source --version 2>&1 | Out-String
if ($LASTEXITCODE -eq 0) {
$hipVerLine = ($hipVerOut -split '\r?\n' | Where-Object { $_.Trim() } | Select-Object -First 1).Trim()
if ($hipVerLine -match '(\d+\.\d+)') {
$script:ROCmVersion = $Matches[1]
$script:ROCmVersionFull = $hipVerLine
}
}
} catch {}
}
if (-not $script:ROCmVersion) {
$amdSmiVer = Get-Command "amd-smi" -ErrorAction SilentlyContinue
if ($amdSmiVer) {
try {
$smiVerOut = & $amdSmiVer.Source version 2>&1 | Out-String
if ($LASTEXITCODE -eq 0 -and $smiVerOut -match 'ROCm version:\s*(\d+\.\d+)') { $script:ROCmVersion = $Matches[1] }
} catch {}
}
}
}
}
if ($HasNvidiaSmi) {
step "gpu" "NVIDIA GPU detected"
} elseif ($HasROCm) {
step "gpu" $ROCmGpuLabel
$hipSdkPath = if ($env:HIP_PATH) { $env:HIP_PATH } elseif ($env:ROCM_PATH) { $env:ROCM_PATH } else { "on system PATH" }
substep "HIP SDK: $hipSdkPath"
if ($script:ROCmVersionFull) { substep "hipconfig: $script:ROCmVersionFull" }
} elseif ($HipSdkInstalled -and $ROCmGpuLabel) {
# HIP SDK is installed but ROCm can't see the device (driver issue, not SDK issue)
$sdkVer = if ($script:ROCmVersionFull) { " (HIP $script:ROCmVersionFull)" } else { "" }
Write-Host ""
step "gpu" "AMD GPU detected -- not ROCm-accessible$sdkVer" "Yellow"
substep "Detected: $ROCmGpuLabel" "Yellow"
substep "[WARN] HIP SDK is installed but hipinfo reports no ROCm-capable device." "Yellow"
substep " This is a driver issue, not an SDK issue." "Yellow"
substep " Ensure the ROCm compute driver is installed alongside the display driver:" "Yellow"
substep " https://rocm.docs.amd.com/en/latest/deploy/windows/index.html" "Yellow"
} elseif ($ROCmGpuLabel) {
Write-Host ""
step "gpu" "AMD GPU detected -- HIP SDK not found" "Yellow"
substep "Detected: $ROCmGpuLabel" "Yellow"
substep "Install the HIP SDK for ROCm GPU inference:" "Yellow"
substep "https://rocm.docs.amd.com/en/latest/deploy/windows/index.html" "Yellow"
Write-Host ""
} else {
Write-Host ""
step "gpu" "none (chat-only / GGUF)" "Yellow"
substep "Training and GPU inference require an NVIDIA or AMD ROCm GPU." "Yellow"
Write-Host ""
}
# ============================================
# 1a.5. Windows Long Paths (required for deep node_modules / Python paths)
# ============================================
$LongPathsEnabled = $false
try {
$regVal = Get-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -ErrorAction SilentlyContinue
if ($regVal -and $regVal.LongPathsEnabled -eq 1) {
$LongPathsEnabled = $true
}
} catch {}
if ($LongPathsEnabled) {
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
try {
# Spawn an elevated process to set the registry key (triggers UAC prompt)
$proc = Start-Process -FilePath "reg.exe" `
-ArgumentList 'add "HKLM\SYSTEM\CurrentControlSet\Control\FileSystem" /v LongPathsEnabled /t REG_DWORD /d 1 /f' `
-Verb RunAs -Wait -PassThru -ErrorAction Stop
if ($proc.ExitCode -eq 0) {
$LongPathsEnabled = $true
step "long paths" "enabled (via UAC)"
} else {
step "long paths" "failed to enable (exit code: $($proc.ExitCode))" "Yellow"
}
} catch {
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
}
}
# ============================================
# 1b. Git (required by pip for git+https:// deps and by npm)
# ============================================
$HasGit = $null -ne (Get-Command git -ErrorAction SilentlyContinue)
if (-not $HasGit) {
Write-Host "Git not found -- installing via winget..." -ForegroundColor Yellow
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
if ($HasWinget) {
try {
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 { }
}
if (-not $HasGit) {
Write-Host "[ERROR] Git is required but could not be installed automatically." -ForegroundColor Red
Write-Host " Install Git from https://git-scm.com/download/win and re-run." -ForegroundColor Red
exit 1
}
step "git" "$(git --version)"
} else {
step "git" "$(git --version)"
}
# ============================================
# 1c. CMake (required for llama.cpp build)
# ============================================
$HasCmake = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
if (-not $HasCmake) {
Write-Host "CMake not found -- installing via winget..." -ForegroundColor Yellow
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
if ($HasWinget) {
try {
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 { }
}
# winget may succeed but cmake isn't on PATH yet (MSI PATH changes need a
# new shell). Try the default install location as a fallback.
if (-not $HasCmake) {
$cmakeDefaults = @(
"$env:ProgramFiles\CMake\bin",
"${env:ProgramFiles(x86)}\CMake\bin",
"$env:LOCALAPPDATA\CMake\bin"
)
foreach ($d in $cmakeDefaults) {
if (Test-Path (Join-Path $d "cmake.exe")) {
$env:Path = "$d;$env:Path"
# Persist to user PATH (Prepend so this cmake wins over older ones).
Add-ToUserPath -Directory $d -Position 'Prepend' | Out-Null
$HasCmake = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
if ($HasCmake) {
Write-Host " Found cmake at $d (added to PATH)" -ForegroundColor Gray
break
}
}
}
}
if ($HasCmake) {
step "cmake" "installed"
} else {
Write-Host "[ERROR] CMake is required but could not be installed." -ForegroundColor Red
Write-Host " Install CMake from https://cmake.org/download/ and re-run." -ForegroundColor Red
exit 1
}
} else {
step "cmake" "$(cmake --version | Select-Object -First 1)"
}
# ============================================
# 1d. Visual Studio Build Tools (C++ compiler for llama.cpp)
# ============================================
$CmakeGenerator = $null
$VsInstallPath = $null
$vsResult = Find-VsBuildTools
if (-not $vsResult) {
Write-Host "Visual Studio Build Tools not found -- installing via winget..." -ForegroundColor Yellow
Write-Host " (This is a one-time install, may take several minutes)" -ForegroundColor Gray
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
if ($HasWinget) {
$prevEAPTemp = $ErrorActionPreference
$ErrorActionPreference = "Continue"
winget install Microsoft.VisualStudio.2022.BuildTools --source winget --accept-package-agreements --accept-source-agreements --override "--add Microsoft.VisualStudio.Workload.VCTools --includeRecommended --passive --wait"
$ErrorActionPreference = $prevEAPTemp
# Re-scan after install (don't trust vswhere catalog)
$vsResult = Find-VsBuildTools
}
}
if ($vsResult) {
$CmakeGenerator = $vsResult.Generator
$VsInstallPath = $vsResult.InstallPath
step "vs" "$CmakeGenerator ($($vsResult.Source))"
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
Write-Host ' 1. winget install Microsoft.VisualStudio.2022.BuildTools --source winget' -ForegroundColor Yellow
Write-Host ' 2. Open Visual Studio Installer -> Modify -> check "Desktop development with C++"' -ForegroundColor Yellow
exit 1
}
# ============================================
# 1e. CUDA Toolkit (nvcc for llama.cpp build + env vars)
# ============================================
if ($HasNvidiaSmi) {
# IMPORTANT: The CUDA Toolkit version must be <= the max CUDA version the
# NVIDIA driver supports. nvidia-smi reports this as "CUDA Version: X.Y".
# If we install a toolkit newer than the driver supports, llama-server will
# fail at runtime with "ggml_cuda_init: failed to initialize CUDA: (null)".
# -- Detect max CUDA version the driver supports --
$DriverMaxCuda = $null
try {
$smiOut = & $NvidiaSmiExe 2>&1 | Out-String
# Newer NVIDIA drivers (e.g. 610.x) report the driver max CUDA as
# "CUDA UMD Version: X.Y" rather than "CUDA Version: X.Y"; accept both.
if ($smiOut -match "CUDA(?: UMD)? Version:\s+([\d]+)\.([\d]+)") {
$DriverMaxCuda = "$($Matches[1]).$($Matches[2])"
substep "driver supports up to CUDA $DriverMaxCuda"
}
} catch {}
# Detect compute capability early so we can validate toolkit support
$CudaArch = Get-CudaComputeCapability
if ($CudaArch) {
substep "GPU Compute Capability = $($CudaArch.Insert($CudaArch.Length-1, '.')) (sm_$CudaArch)"
}
# -- Find a toolkit that's compatible with the driver AND the GPU --
# Strategy: prefer the toolkit at CUDA_PATH (user's existing setup) if it's
# compatible with the driver AND supports the GPU architecture. Only fall back
# to scanning side-by-side installs if CUDA_PATH is missing, points to an
# incompatible version, or can't compile for the GPU. This avoids
# header/binary mismatches when multiple toolkits are installed.
$IncompatibleToolkit = $null
$NvccPath = $null
if ($DriverMaxCuda) {
$drMajorCuda = [int]$DriverMaxCuda.Split('.')[0]
$drMinorCuda = [int]$DriverMaxCuda.Split('.')[1]
# --- Step 1: Check existing CUDA_PATH first ---
$existingCudaPath = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'Machine')
if (-not $existingCudaPath) {
$existingCudaPath = [Environment]::GetEnvironmentVariable('CUDA_PATH', 'User')
}
if ($existingCudaPath -and (Test-Path (Join-Path $existingCudaPath 'bin\nvcc.exe'))) {
$candidateNvcc = Join-Path $existingCudaPath 'bin\nvcc.exe'
$verOut = & $candidateNvcc --version 2>&1 | Out-String
if ($verOut -match 'release\s+(\d+)\.(\d+)') {
$tkMaj = [int]$Matches[1]; $tkMin = [int]$Matches[2]
$isCompat = ($tkMaj -lt $drMajorCuda) -or ($tkMaj -eq $drMajorCuda -and $tkMin -le $drMinorCuda)
if ($isCompat) {
# Also verify the toolkit supports our GPU architecture
$archOk = $true
if ($CudaArch) {
$archOk = Test-NvccArchSupport -NvccExe $candidateNvcc -Arch $CudaArch
if (-not $archOk) {
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
substep "using existing CUDA Toolkit at CUDA_PATH (nvcc: $NvccPath)"
}
} else {
substep "CUDA_PATH ($existingCudaPath) has CUDA $tkMaj.$tkMin which exceeds driver max $DriverMaxCuda" "Yellow"
}
}
}
# --- Step 2: Fall back to scanning side-by-side installs ---
if (-not $NvccPath) {
$NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
if ($NvccPath) {
substep "found compatible CUDA Toolkit (nvcc: $NvccPath)"
if ($existingCudaPath) {
$selectedRoot = Split-Path (Split-Path $NvccPath -Parent) -Parent
if ($existingCudaPath.TrimEnd('\') -ne $selectedRoot.TrimEnd('\')) {
substep "overriding CUDA_PATH from $existingCudaPath to $selectedRoot" "Yellow"
}
}
} else {
# Check if there's an incompatible (too new) toolkit installed
$AnyNvcc = Find-Nvcc
if ($AnyNvcc) {
$NvccOut = & $AnyNvcc --version 2>&1 | Out-String
if ($NvccOut -match "release\s+([\d]+\.[\d]+)") {
$IncompatibleToolkit = $Matches[1]
}
}
}
}
} else {
$NvccPath = Find-Nvcc
}
# -- If incompatible toolkit is blocking, tell user to uninstall it --
if (-not $NvccPath -and $IncompatibleToolkit) {
Write-Host "" -ForegroundColor Red
Write-Host "========================================================================" -ForegroundColor Red
Write-Host "[ERROR] CUDA Toolkit $IncompatibleToolkit is installed but INCOMPATIBLE" -ForegroundColor Red
Write-Host " with your NVIDIA driver (which supports up to CUDA $DriverMaxCuda)." -ForegroundColor Red
Write-Host "" -ForegroundColor Red
Write-Host " This will cause 'failed to initialize CUDA' errors at runtime." -ForegroundColor Red
Write-Host "" -ForegroundColor Red
Write-Host " To fix:" -ForegroundColor Yellow
Write-Host " 1. Open Control Panel -> Programs -> Uninstall a program" -ForegroundColor Yellow
Write-Host " 2. Uninstall 'NVIDIA CUDA Toolkit $IncompatibleToolkit'" -ForegroundColor Yellow
Write-Host " 3. Re-run setup.bat (it will install CUDA $DriverMaxCuda automatically)" -ForegroundColor Yellow
Write-Host "" -ForegroundColor Yellow
Write-Host " Alternatively, update your NVIDIA driver to one that supports CUDA $IncompatibleToolkit." -ForegroundColor Gray
Write-Host "========================================================================" -ForegroundColor Red
exit 1
}
# -- No toolkit at all: install via winget --
if (-not $NvccPath) {
Write-Host "CUDA toolkit (nvcc) not found -- installing via winget..." -ForegroundColor Yellow
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
if ($HasWinget) {
if ($DriverMaxCuda) {
# Query winget for available CUDA Toolkit versions
$drMajor = [int]$DriverMaxCuda.Split('.')[0]
$drMinor = [int]$DriverMaxCuda.Split('.')[1]
$AvailableVersions = @()
try {
$rawOutput = winget show Nvidia.CUDA --versions --accept-source-agreements 2>&1 | Out-String
# Parse version lines (e.g. "12.6", "12.5", "11.8")
foreach ($line in $rawOutput -split "`n") {
$line = $line.Trim()
if ($line -match '^\d+\.\d+') {
$AvailableVersions += $line
}
}
} catch {}
# Filter to compatible versions (<= driver max) and pick the highest
$BestVersion = $null
foreach ($ver in $AvailableVersions) {
$parts = $ver.Split('.')
$vMajor = [int]$parts[0]
$vMinor = [int]$parts[1]
if ($vMajor -lt $drMajor -or ($vMajor -eq $drMajor -and $vMinor -le $drMinor)) {
$BestVersion = $ver
break # list is descending, first match is highest compatible
}
}
if ($BestVersion) {
substep "Installing CUDA Toolkit $BestVersion via winget..."
$prevEAPCuda = $ErrorActionPreference
$ErrorActionPreference = "Continue"
Invoke-SetupCommand { winget install --id=Nvidia.CUDA --version=$BestVersion -e --source winget --accept-package-agreements --accept-source-agreements } | Out-Null
$ErrorActionPreference = $prevEAPCuda
Refresh-Environment
$NvccPath = Find-Nvcc -MaxVersion $DriverMaxCuda
if ($NvccPath) {
substep "CUDA Toolkit $BestVersion installed (nvcc: $NvccPath)"
}
} else {
substep "no compatible CUDA Toolkit version found in winget (need <= $DriverMaxCuda)" "Yellow"
}
} else {
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) {
substep "CUDA Toolkit installed (nvcc: $NvccPath)"
}
}
}
}
if (-not $NvccPath) {
Write-Host "[ERROR] CUDA Toolkit (nvcc) is required but could not be found or installed." -ForegroundColor Red
if ($DriverMaxCuda) {
Write-Host " Install CUDA Toolkit $DriverMaxCuda from https://developer.nvidia.com/cuda-toolkit-archive" -ForegroundColor Yellow
} else {
Write-Host " Install CUDA Toolkit from https://developer.nvidia.com/cuda-downloads" -ForegroundColor Yellow
}
exit 1
}
# -- Set CUDA env vars so cmake AND MSBuild can find the toolkit --
$CudaToolkitRoot = Split-Path (Split-Path $NvccPath -Parent) -Parent
# CUDA_PATH: used by cmake's find_package(CUDAToolkit)
[Environment]::SetEnvironmentVariable('CUDA_PATH', $CudaToolkitRoot, 'Process')
# CudaToolkitDir: the MSBuild property that CUDA .targets checks directly
# Trailing backslash required -- the .targets file appends subpaths to it
[Environment]::SetEnvironmentVariable('CudaToolkitDir', "$CudaToolkitRoot\", 'Process')
# Always persist CUDA_PATH to User registry so the compatible toolkit is used
# in future sessions (overwrites any existing value pointing to a newer, incompatible version)
[Environment]::SetEnvironmentVariable('CUDA_PATH', $CudaToolkitRoot, 'User')
substep "Persisted CUDA_PATH=$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 { $_ -match '^CUDA_PATH_V' })
foreach ($v in $cudaPathVars) {
[Environment]::SetEnvironmentVariable($v, $null, 'Process')
}
# Set only the versioned var matching the selected toolkit (e.g. CUDA_PATH_V13_0)
$tkDirName = Split-Path $CudaToolkitRoot -Leaf
if ($tkDirName -match '^v(\d+)\.(\d+)') {
$cudaPathVerVar = "CUDA_PATH_V$($Matches[1])_$($Matches[2])"
[Environment]::SetEnvironmentVariable($cudaPathVerVar, $CudaToolkitRoot, 'Process')
substep "Set $cudaPathVerVar (cleared other CUDA_PATH_V* vars)"
}
# Ensure nvcc's bin dir is on PATH for this process
$nvccBinDir = Split-Path $NvccPath -Parent
if ($env:PATH -notlike "*$nvccBinDir*") {
[Environment]::SetEnvironmentVariable('PATH', "$nvccBinDir;$env:PATH", 'Process')
}
# Persist nvcc bin dir (Prepend so the driver-compatible toolkit wins).
if (Add-ToUserPath -Directory $nvccBinDir -Position 'Prepend') {
substep "Persisted CUDA bin dir to user PATH"
}
# -- Ensure CUDA ↔ Visual Studio integration files exist --
# When CUDA is installed before VS Build Tools (or VS is reinstalled after CUDA),
# the MSBuild .targets/.props files that let VS compile .cu files are missing.
# cmake fails with "No CUDA toolset found". Fix: copy from CUDA extras dir.
if ($VsInstallPath -and $CudaToolkitRoot) {
$vsCustomizations = Join-Path $VsInstallPath "MSBuild\Microsoft\VC\v170\BuildCustomizations"
$cudaExtras = Join-Path $CudaToolkitRoot "extras\visual_studio_integration\MSBuildExtensions"
if ((Test-Path $cudaExtras) -and (Test-Path $vsCustomizations)) {
$hasTargets = Get-ChildItem $vsCustomizations -Filter "CUDA *.targets" -ErrorAction SilentlyContinue
if (-not $hasTargets) {
substep "CUDA VS integration missing -- copying .targets files..." "Yellow"
try {
Copy-Item "$cudaExtras\*" $vsCustomizations -Force -ErrorAction Stop
substep "CUDA VS integration files installed"
} catch {
# Direct copy failed (needs admin). Try elevated copy via Start-Process.
try {
$copyCmd = "Copy-Item '$cudaExtras\*' '$vsCustomizations' -Force"
Start-Process powershell -ArgumentList "-NoProfile -Command $copyCmd" -Verb RunAs -Wait -ErrorAction Stop
$hasTargetsRetry = Get-ChildItem $vsCustomizations -Filter "CUDA *.targets" -ErrorAction SilentlyContinue
if ($hasTargetsRetry) {
substep "CUDA VS integration files installed (elevated)"
} else {
throw "Copy did not produce .targets files"
}
} catch {
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"
}
}
}
}
}
step "cuda" $NvccPath
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 $CudaArch) {
substep "could not detect compute capability -- cmake will use defaults" "Yellow"
}
} else {
step "cuda" "skipped (no NVIDIA GPU detected)" "Yellow"
}
if ($HasROCm) {
$rocmVerLabel = if ($script:ROCmVersionFull) { "ROCm $script:ROCmVersionFull" } elseif ($script:ROCmVersion) { "ROCm $script:ROCmVersion" } else { "ROCm (version unknown)" }
step "rocm" $rocmVerLabel
} elseif ($ROCmGpuLabel) {
step "rocm" "HIP SDK not found -- GPU-accelerated training unavailable" "Yellow"
}
# ============================================
# 1f. Node.js / npm (skip if pip-installed or Tauri -- only needed for frontend build)
# ============================================
$SkipFrontend = ($env:SKIP_STUDIO_FRONTEND -eq "1")
if ($IsPipInstall) {
step "frontend" "bundled (pip install)"
} elseif ($SkipFrontend) {
step "frontend" "bundled (Tauri)"
} else {
# setup.sh installs Node LTS (v22) via nvm. We enforce the same range here:
# Vite 8 requires Node ^20.19.0 || >=22.12.0, npm >= 11.
$NeedNode = $true
try {
$NodeVersion = (node -v 2>$null)
$NpmVersion = (npm -v 2>$null)
if ($NodeVersion -and $NpmVersion) {
$NodeParts = ($NodeVersion -replace 'v','').Split('.')
$NodeMajor = [int]$NodeParts[0]
$NodeMinor = [int]$NodeParts[1]
$NpmMajor = [int]$NpmVersion.Split('.')[0]
# Vite 8: ^20.19.0 || >=22.12.0
$NodeOk = ($NodeMajor -eq 20 -and $NodeMinor -ge 19) -or
($NodeMajor -eq 22 -and $NodeMinor -ge 12) -or
($NodeMajor -ge 23)
if ($NodeOk -and $NpmMajor -ge 11) {
substep "Node $NodeVersion and npm $NpmVersion already meet requirements."
$NeedNode = $false
} else {
substep "Node $NodeVersion / npm $NpmVersion too old." "Yellow"
}
}
} catch {
substep "Node/npm not found." "Yellow"
}
if ($NeedNode) {
substep "installing Node.js LTS via winget..."
try {
winget install OpenJS.NodeJS.LTS --source winget --accept-package-agreements --accept-source-agreements
Refresh-Environment
} catch {
Write-Host "[ERROR] Could not install Node.js automatically." -ForegroundColor Red
Write-Host "Please install Node.js >= 20 from https://nodejs.org/" -ForegroundColor Red
exit 1
}
}
step "node" "$(node -v) | npm $(npm -v)"
# ── bun (optional, faster package installs) ──
# Installed via npm — Node is already guaranteed above. Works on all platforms.
if (-not (Get-Command bun -ErrorAction SilentlyContinue)) {
substep "installing bun (faster frontend package installs)..."
$prevEAP_bun = $ErrorActionPreference
$ErrorActionPreference = "Continue"
Invoke-SetupCommand { npm install -g bun } | Out-Null
$ErrorActionPreference = $prevEAP_bun
Refresh-Environment
if (Get-Command bun -ErrorAction SilentlyContinue) {
substep "bun installed ($(bun --version))"
} else {
substep "bun install skipped (npm will be used instead)"
}
} else {
substep "bun already installed ($(bun --version))"
}
}
# 1g. Python (>= 3.11 and < 3.14). Prefer py.exe so a 3.14 ahead of 3.13 on PATH does not trip the gate.
$HasPython = $null -ne (Get-Command python -ErrorAction SilentlyContinue)
$PyLauncher = Get-Command py -CommandType Application -ErrorAction SilentlyContinue
$PythonOk = $false
$DetectedPyVer = $null
if ($PyLauncher) {
foreach ($minor in @("3.13", "3.12", "3.11")) {
try {
$out = & $PyLauncher.Source "-$minor" --version 2>&1 | Out-String
if ($out -match 'Python (3\.\d+\.\d+)') {
$DetectedPyVer = $Matches[1]
# Make `python` resolvable for the rest of setup. Without this,
# py-launcher-only installs (no python.exe on PATH) pass the gate
# and then crash on the first bare `python` call below.
try {
$resolvedExe = (& $PyLauncher.Source "-$minor" -c "import sys; print(sys.executable)" 2>$null | Select-Object -First 1)
if ($resolvedExe -and (Test-Path $resolvedExe)) {
$resolvedDir = Split-Path -Parent $resolvedExe
$alreadyOnPath = ($env:PATH -split ';' | Where-Object { $_.TrimEnd('\') -ieq $resolvedDir.TrimEnd('\') }).Count -gt 0
if (-not $alreadyOnPath) {
$env:PATH = "$resolvedDir;$env:PATH"
}
$HasPython = $true
}
} catch { }
$PythonOk = $true
break
}
} catch { }
}
}
if (-not $PythonOk -and $HasPython) {
$PyVer = python --version 2>&1
if ($PyVer -match "(\d+)\.(\d+)") {
$PyMajor = [int]$Matches[1]; $PyMinor = [int]$Matches[2]
if ($PyMajor -eq 3 -and $PyMinor -ge 11 -and $PyMinor -lt 14) {
$DetectedPyVer = "$PyMajor.$PyMinor"
$PythonOk = $true
}
}
}
if ($PythonOk) {
substep "Python $DetectedPyVer"
} elseif (-not $HasPython) {
# No `python` on PATH (and py.exe either absent or only had unsupported
# minors). Try winget as before -- gating on $HasPython alone, not also
# on $PyLauncher, so a launcher-only install with just 3.14 still gets
# an automatic 3.12 install instead of a hard error.
Write-Host "Python 3.11-3.13 not found -- installing Python 3.12 via winget..." -ForegroundColor Yellow
$HasWinget = $null -ne (Get-Command winget -ErrorAction SilentlyContinue)
if ($HasWinget) {
winget install -e --id Python.Python.3.12 --source winget --accept-package-agreements --accept-source-agreements
Refresh-Environment
}
$HasPython = $null -ne (Get-Command python -ErrorAction SilentlyContinue)
if (-not $HasPython) {
Write-Host "[ERROR] Python could not be installed automatically." -ForegroundColor Red
Write-Host " Install Python 3.12 from https://python.org/downloads/" -ForegroundColor Yellow
exit 1
}
step "python" "$(python --version 2>&1)"
$PythonOk = $true
} else {
# python.exe is on PATH but its version is unsupported, and py.exe (if
# present) had no supported minor either.
Write-Host "[ERROR] No supported Python (3.11-3.13) found on this system." -ForegroundColor Red
Write-Host " py.exe could not locate -3.11/-3.12/-3.13 and `python` on PATH is unsupported." -ForegroundColor Yellow
Write-Host " Install Python 3.12 from https://python.org/downloads/" -ForegroundColor Yellow
exit 1
}
# Add user-scheme Python Scripts dir to PATH (nt_user only, no venv fallback).
$ScriptsDir = python -c "import os, sysconfig; p = sysconfig.get_path('scripts', 'nt_user'); print(p if os.path.exists(p) else '')"
if ($LASTEXITCODE -eq 0 -and $ScriptsDir -and (Test-Path $ScriptsDir)) {
# Append (not Prepend) -- this dir has other pip scripts; shim handles unsloth.
if (Add-ToUserPath -Directory $ScriptsDir) {
# Also add to current process so it's available immediately
$ProcessPathEntries = $env:PATH.Split(';')
if (-not ($ProcessPathEntries | Where-Object { $_.TrimEnd('\') -eq $ScriptsDir })) {
$env:PATH = "$ScriptsDir;$env:PATH"
}
substep "Persisted Python Scripts dir to user PATH: $ScriptsDir"
}
}
Write-Host ""
step "system" "prerequisites ready"
Write-Host ""
# ==========================================================================
# PHASE 2: Frontend build (skip if pip-installed -- already bundled)
# ==========================================================================
$DistDir = Join-Path $FrontendDir "dist"
# Skip build if dist/ exists and no tracked input is newer than dist/.
# Checks src/, public/, package.json, config files -- not just src/.
$NeedFrontendBuild = $true
if ($IsPipInstall) {
$NeedFrontendBuild = $false
step "frontend" "bundled (pip install)"
} elseif ($SkipFrontend) {
$NeedFrontendBuild = $false
step "frontend" "bundled (Tauri)"
} elseif (Test-Path $DistDir) {
$DistTime = (Get-Item $DistDir).LastWriteTime
$NewerFile = $null
# Check src/ and public/ recursively (probe paths directly, not via -Include)
foreach ($subDir in @("src", "public")) {
$subPath = Join-Path $FrontendDir $subDir
if (Test-Path $subPath) {
$NewerFile = Get-ChildItem -Path $subPath -Recurse -File -ErrorAction SilentlyContinue |
Where-Object { $_.LastWriteTime -gt $DistTime } | Select-Object -First 1
if ($NewerFile) { break }
}
}
# Also check all top-level files (package.json, vite.config.ts, index.html, etc.)
if (-not $NewerFile) {
$NewerFile = Get-ChildItem -Path $FrontendDir -File -ErrorAction SilentlyContinue |
Where-Object { $_.Name -ne "bun.lock" -and $_.LastWriteTime -gt $DistTime } |
Select-Object -First 1
}
if (-not $NewerFile) {
$NeedFrontendBuild = $false
step "frontend" "up to date"
} else {
substep "Frontend source changed since last build -- rebuilding..." "Yellow"
}
}
if ($NeedFrontendBuild -and -not $IsPipInstall) {
Write-Host ""
substep "building frontend..."
# ── Tailwind v4 .gitignore workaround ──
# Tailwind v4's oxide scanner respects .gitignore in parent directories.
# Python venvs create a .gitignore with "*" (ignore everything), which
# prevents Tailwind from scanning .tsx source files for class names.
# Temporarily hide any such .gitignore during the build, then restore it.
$HiddenGitignores = @()
$WalkDir = (Get-Item $FrontendDir).Parent.FullName
while ($WalkDir -and $WalkDir -ne [System.IO.Path]::GetPathRoot($WalkDir)) {
$gi = Join-Path $WalkDir ".gitignore"
if (Test-Path $gi) {
$content = Get-Content $gi -Raw -ErrorAction SilentlyContinue
if ($content -and ($content.Trim() -match '^\*$')) {
$hidden = "$gi._twbuild"
Rename-Item -Path $gi -NewName (Split-Path $hidden -Leaf) -Force
$HiddenGitignores += $gi
substep "Temporarily hiding $gi (venv .gitignore blocks Tailwind scanner)"
}
}
$WalkDir = Split-Path $WalkDir -Parent
}
# Use bun if available (faster install), fall back to npm.
# Bun is used only as package manager; Node runs the actual build (Vite 8).
$prevEAP_npm = $ErrorActionPreference
$ErrorActionPreference = "Continue"
Push-Location $FrontendDir
$UseBun = $null -ne (Get-Command bun -ErrorAction SilentlyContinue)
# bun's package cache can become corrupt -- packages get stored with only
# metadata but no actual content (bin/, lib/). When this happens bun install
# exits 0 but leaves binaries missing. We validate after install and clear
# the cache + retry once before falling back to npm.
if ($UseBun) {
Write-Host " Using bun for package install (faster)" -ForegroundColor DarkGray
$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) {
Write-Host " bun install exited 0 but critical binaries are missing, clearing cache and retrying..." -ForegroundColor Yellow
if (Test-Path "node_modules") {
Remove-Item "node_modules" -Recurse -Force -ErrorAction SilentlyContinue
}
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") {
Remove-Item "node_modules" -Recurse -Force -ErrorAction SilentlyContinue
}
$UseBun = $false
}
} else {
substep "bun install failed (exit $bunExit), falling back to npm" "Yellow"
if (Test-Path "node_modules") {
Remove-Item "node_modules" -Recurse -Force -ErrorAction SilentlyContinue
}
$UseBun = $false
}
}
if (-not $UseBun) {
$npmExit = Invoke-SetupCommand { npm install }
if ($npmExit -ne 0) {
Pop-Location
$ErrorActionPreference = $prevEAP_npm
foreach ($gi in $HiddenGitignores) { Rename-Item -Path "$gi._twbuild" -NewName (Split-Path $gi -Leaf) -Force -ErrorAction SilentlyContinue }
Write-Host "[ERROR] npm install failed (exit code $npmExit)" -ForegroundColor Red
Write-Host " Try running 'npm install' manually in frontend/ to see errors" -ForegroundColor Yellow
exit 1
}
}
# Always use npm to run the build (Node runtime — avoids bun Windows runtime issues)
$buildExit = Invoke-SetupCommand { npm run build }
if ($buildExit -ne 0) {
Pop-Location
$ErrorActionPreference = $prevEAP_npm
foreach ($gi in $HiddenGitignores) { Rename-Item -Path "$gi._twbuild" -NewName (Split-Path $gi -Leaf) -Force -ErrorAction SilentlyContinue }
Write-Host "[ERROR] npm run build failed (exit code $buildExit)" -ForegroundColor Red
exit 1
}
Pop-Location
$ErrorActionPreference = $prevEAP_npm
# ── Restore hidden .gitignore files ──
foreach ($gi in $HiddenGitignores) {
Rename-Item -Path "$gi._twbuild" -NewName (Split-Path $gi -Leaf) -Force -ErrorAction SilentlyContinue
}
# ── Validate CSS output ──
$CssFiles = Get-ChildItem (Join-Path $DistDir "assets") -Filter "*.css" -ErrorAction SilentlyContinue
$MaxCssSize = ($CssFiles | Measure-Object -Property Length -Maximum).Maximum
if ($MaxCssSize -lt 100000) {
step "frontend" "built (warning: CSS may be truncated)" "Yellow"
} else {
step "frontend" "built"
}
}
if (Test-Path $OxcValidatorDir) {
substep "installing OXC validator runtime..."
$prevEAP_oxc = $ErrorActionPreference
$ErrorActionPreference = "Continue"
Push-Location $OxcValidatorDir
$oxcInstallExit = Invoke-SetupCommand { npm install }
if ($oxcInstallExit -ne 0) {
Pop-Location
$ErrorActionPreference = $prevEAP_oxc
Write-Host "[ERROR] OXC validator npm install failed (exit code $oxcInstallExit)" -ForegroundColor Red
exit 1
}
Pop-Location
$ErrorActionPreference = $prevEAP_oxc
step "oxc runtime" "installed"
}
# ==========================================================================
# PHASE 3: Python environment + dependencies
# ==========================================================================
Write-Host ""
substep "setting up Python environment..."
# Find Python -- skip Anaconda/Miniconda distributions.
# Conda-bundled CPython ships modified DLL search paths that break
# torch's c10.dll loading on Windows. Standalone CPython (python.org,
# winget, uv) does not have this issue.
# Uses Get-Command -All to look past conda entries that shadow a valid
# standalone Python further down PATH, and probes py.exe (the Python
# Launcher) which reliably finds python.org installs.
#
# NOTE: A venv created from conda Python inherits conda's base_prefix
# even though the venv path itself does not contain "conda". We check
# both the executable path AND sys.base_prefix to catch this case.
$CondaSkipPattern = '(?i)(conda|miniconda|anaconda|miniforge|mambaforge)'
$PythonCmd = $null
# Helper: check if a Python executable is conda-based by inspecting
# both the path and sys.base_prefix (catches venvs created from conda).
function Test-IsConda {
param([string]$Exe)
if ($Exe -match $CondaSkipPattern) { return $true }
try {
$basePrefix = (& $Exe -c "import sys; print(sys.base_prefix)" 2>$null | Out-String).Trim()
if ($basePrefix -match $CondaSkipPattern) { return $true }
} catch { }
return $false
}
# 1. Try the Python Launcher (py.exe) first -- most reliable on Windows.
# py.exe is installed by python.org and resolves to standalone CPython.
$pyLauncher = Get-Command py -CommandType Application -ErrorAction SilentlyContinue
if ($pyLauncher -and $pyLauncher.Source -notmatch $CondaSkipPattern) {
foreach ($minor in @("3.13", "3.12", "3.11")) {
try {
$out = & $pyLauncher.Source "-$minor" --version 2>&1 | Out-String
if ($out -match 'Python 3\.(\d+)') {
$pyMinor = [int]$Matches[1]
if ($pyMinor -ge 11 -and $pyMinor -le 13) {
# Resolve the actual executable path so venv creation
# does not re-resolve back to a conda interpreter.
$resolvedExe = (& $pyLauncher.Source "-$minor" -c "import sys; print(sys.executable)" 2>$null | Out-String).Trim()
if ($resolvedExe -and (Test-Path $resolvedExe) -and -not (Test-IsConda $resolvedExe)) {
$PythonCmd = $resolvedExe
break
}
}
}
} catch { }
}
}
# 2. Fall back to scanning python3.x / python3 / python on PATH.
# Use Get-Command -All to look past conda entries.
if (-not $PythonCmd) {
foreach ($candidate in @("python3.13", "python3.12", "python3.11", "python3", "python")) {
foreach ($cmdInfo in @(Get-Command $candidate -All -ErrorAction SilentlyContinue)) {
try {
if (-not $cmdInfo.Source) { continue }
if ($cmdInfo.Source -like "*\WindowsApps\*") { continue }
if (Test-IsConda $cmdInfo.Source) {
substep "skipping $($cmdInfo.Source) (conda Python breaks torch DLL loading)" "Yellow"
continue
}
$ver = & $cmdInfo.Source --version 2>&1
if ($ver -match 'Python 3\.(\d+)') {
$minor = [int]$Matches[1]
if ($minor -ge 11 -and $minor -le 13) {
$PythonCmd = $cmdInfo.Source
break
}
}
} catch { }
}
if ($PythonCmd) { break }
}
}
if (-not $PythonCmd) {
Write-Host "[ERROR] No standalone Python 3.11-3.13 found (conda Python is not supported)." -ForegroundColor Red
Write-Host " Install Python from https://python.org/downloads/ or via:" -ForegroundColor Yellow
Write-Host " winget install -e --id Python.Python.3.12" -ForegroundColor Yellow
exit 1
}
substep "Python found: $PythonCmd"
# The venv must already exist (created by install.ps1); this script only
# updates packages. UNSLOTH_STUDIO_HOME (or STUDIO_HOME alias) overrides the
# root. UNSLOTH_STUDIO_HOME wins when both are set. Whitespace-only values
# are treated as unset to match Python .strip() semantics.
$_studioOverrideVar = $null
$_studioOverride = $null
if (-not [string]::IsNullOrWhiteSpace($env:UNSLOTH_STUDIO_HOME)) {
$_studioOverrideVar = "UNSLOTH_STUDIO_HOME"
$_studioOverride = $env:UNSLOTH_STUDIO_HOME.Trim()
} elseif (-not [string]::IsNullOrWhiteSpace($env:STUDIO_HOME)) {
$_studioOverrideVar = "STUDIO_HOME"
$_studioOverride = $env:STUDIO_HOME.Trim()
}
if ($_studioOverride) {
if ($_studioOverride -eq "~" -or $_studioOverride -like "~/*" -or $_studioOverride -like "~\*") {
$_studioOverride = (Join-Path $env:USERPROFILE $_studioOverride.Substring(1).TrimStart('/','\'))
}
if (Test-Path -LiteralPath $_studioOverride -PathType Container) {
$StudioHome = (Resolve-Path -LiteralPath $_studioOverride).Path
# why: mirror setup.sh:417 and install.ps1:130 -- fail fast when the
# custom root is read-only instead of erroring later while creating
# sidecar venvs / installing packages.
$_setupWriteProbe = Join-Path $StudioHome (".unsloth-write-probe-" + [guid]::NewGuid())
try {
[System.IO.File]::WriteAllText($_setupWriteProbe, "")
Remove-Item -LiteralPath $_setupWriteProbe -Force -ErrorAction SilentlyContinue
} catch {
Write-Host "ERROR: $_studioOverrideVar=$StudioHome is not writable." -ForegroundColor Red
exit 1
}
} else {
Write-Host "ERROR: $_studioOverrideVar=$_studioOverride does not exist." -ForegroundColor Red
Write-Host " Run install.ps1 to create the install root before 'unsloth studio update'." -ForegroundColor Red
exit 1
}
} else {
$StudioHome = Join-Path $env:USERPROFILE ".unsloth\studio"
}
$VenvDir = Join-Path $StudioHome "unsloth_studio"
# why: in env-override mode $StudioHome is user-chosen; require the
# ownership marker before Remove-Item so unrelated dirs survive. Gated on
# the canonical comparison so an override pointing at the legacy default
# still behaves like a default install.
$StudioOwnedMarker = ".unsloth-studio-owned"
$LegacyStudioHome = Join-Path $env:USERPROFILE ".unsloth\studio"
$_studioHomeCanon = $StudioHome
if (Test-Path -LiteralPath $_studioHomeCanon -PathType Container) {
$_studioHomeCanon = (Resolve-Path -LiteralPath $_studioHomeCanon).Path
}
if (Test-Path -LiteralPath $LegacyStudioHome -PathType Container) {
$LegacyStudioHome = (Resolve-Path -LiteralPath $LegacyStudioHome).Path
}
$StudioHomeIsCustom = ($_studioHomeCanon -ne $LegacyStudioHome)
function Assert-StudioOwnedOrAbsent {
param(
[Parameter(Mandatory = $true)][string]$Path,
[Parameter(Mandatory = $true)][string]$Label
)
if (-not (Test-Path -LiteralPath $Path -PathType Container)) { return }
if ($StudioHomeIsCustom -and -not (Test-Path -LiteralPath (Join-Path $Path $StudioOwnedMarker) -PathType Leaf)) {
Write-Host "[ERROR] $Path already exists and is not marked as a Studio-owned $Label." -ForegroundColor Red
Write-Host " Move it aside or choose an empty UNSLOTH_STUDIO_HOME before re-running." -ForegroundColor Yellow
exit 1
}
}
function Mark-StudioOwned {
param([Parameter(Mandatory = $true)][string]$Path)
if (-not (Test-Path -LiteralPath $Path -PathType Container)) { return }
try {
[System.IO.File]::WriteAllText((Join-Path $Path $StudioOwnedMarker), "")
} catch {}
}
# Stale-venv detection: if the venv exists but its torch flavor no longer
# matches the current machine, repair according to invocation context.
# - install.ps1 sets UNSLOTH_INSTALL_ROLLBACK_MANAGED=1 so setup can delegate
# to the installer-level rollback that restores the previous environment.
# - direct `unsloth studio update` keeps the pre-existing self-repair behavior.
# In no-torch mode, a missing torch package is expected.
$NoTorchMode = $env:UNSLOTH_NO_TORCH -match '^(?i:true|1|yes)$'
$InstallerManagedSetup = $env:UNSLOTH_INSTALL_ROLLBACK_MANAGED -match '^(?i:true|1|yes)$'
if ((Test-Path -LiteralPath $VenvDir -PathType Container) -and -not $NoTorchMode) {
$VenvPyExe = Join-Path $VenvDir "Scripts\python.exe"
$installedTorchTag = $null
$shouldRebuild = $false
if (Test-Path -LiteralPath $VenvPyExe) {
try {
$psi = New-Object System.Diagnostics.ProcessStartInfo
$psi.FileName = $VenvPyExe
$psi.Arguments = '-c "import torch; print(torch.__version__)"'
$psi.RedirectStandardOutput = $true
$psi.RedirectStandardError = $true
$psi.UseShellExecute = $false
$psi.CreateNoWindow = $true
$proc = [System.Diagnostics.Process]::Start($psi)
$torchVer = $proc.StandardOutput.ReadToEnd().Trim()
$finished = $proc.WaitForExit(30000)
if ($finished -and $proc.ExitCode -eq 0 -and $torchVer) {
if ($torchVer -match '\+(cu\d+)') {
$installedTorchTag = $Matches[1]
} elseif ($torchVer -match '\+cpu') {
$installedTorchTag = "cpu"
} else {
# Untagged wheel (plain "2.x.y" from PyPI) -- treat as cpu
$installedTorchTag = "cpu"
}
} else {
if (-not $finished) { try { $proc.Kill() } catch {} }
$shouldRebuild = $true
}
} catch {
$shouldRebuild = $true
}
} else {
# Missing python.exe means the venv is incomplete -- rebuild it.
$shouldRebuild = $true
}
if (-not $shouldRebuild) {
$expectedTorchTag = if ($HasNvidiaSmi) { Get-PytorchCudaTag } else { "cpu" }
if ($installedTorchTag -and $installedTorchTag -ne $expectedTorchTag) {
$shouldRebuild = $true
}
}
if ($shouldRebuild) {
$reason = if ($installedTorchTag) { "torch $installedTorchTag != required $expectedTorchTag" } else { "torch could not be imported" }
if ($InstallerManagedSetup) {
substep "Stale venv detected ($reason)." "Yellow"
Write-Host " [ERROR] The existing Studio environment needs repair." -ForegroundColor Red
Write-Host " Re-run install.ps1 so it can replace the environment safely with rollback." -ForegroundColor Yellow
exit 1
}
substep "Stale venv detected ($reason) -- rebuilding..." "Yellow"
# why: mirror install.ps1 env-mode guard so an update against a custom
# UNSLOTH_STUDIO_HOME never wipes an unrelated unsloth_studio venv;
# -PathType Leaf rejects a directory masquerading as the sentinel.
if (
$StudioHomeIsCustom -and
-not (Test-Path -LiteralPath (Join-Path $VenvDir $StudioOwnedMarker) -PathType Leaf) -and
-not (Test-Path -LiteralPath (Join-Path $StudioHome "share\studio.conf") -PathType Leaf) -and
-not (Test-Path -LiteralPath (Join-Path $StudioHome "bin\unsloth.exe") -PathType Leaf)
) {
Write-Host "[ERROR] $VenvDir already exists but does not look like an Unsloth Studio install." -ForegroundColor Red
Write-Host " Move it aside or choose an empty UNSLOTH_STUDIO_HOME before re-running." -ForegroundColor Yellow
exit 1
}
try {
Remove-Item -LiteralPath $VenvDir -Recurse -Force -ErrorAction Stop
} catch {
Write-Host " [ERROR] Could not remove stale venv: $($_.Exception.Message)" -ForegroundColor Red
Write-Host " Close any running Studio/Python processes and re-run setup." -ForegroundColor Red
exit 1
}
}
}
if (-not (Test-Path -LiteralPath $VenvDir)) {
Write-Host "[ERROR] Virtual environment not found at $VenvDir" -ForegroundColor Red
Write-Host " Run install.ps1 first to create the environment:" -ForegroundColor Yellow
Write-Host " irm https://unsloth.ai/install.ps1 | iex" -ForegroundColor Yellow
exit 1
} else {
substep "reusing existing virtual environment at $VenvDir"
$_venvPyExe = Join-Path $VenvDir "Scripts\python.exe"
if (Test-Path -LiteralPath $_venvPyExe) {
try {
$_venvPyVer = (& $_venvPyExe --version 2>&1 | Out-String).Trim()
if ($_venvPyVer) { substep $_venvPyVer }
} catch {}
}
}
# pip and python write to stderr even on success (progress bars, warnings).
# With $ErrorActionPreference = "Stop" (set at top of script), PS 5.1
# converts stderr lines into terminating ErrorRecords, breaking output.
# Lower to "Continue" for the pip/python section.
$prevEAP = $ErrorActionPreference
$ErrorActionPreference = "Continue"
$ActivateScript = Join-Path $VenvDir "Scripts\Activate.ps1"
. $ActivateScript
# Try to use uv (much faster than pip), fall back to pip if unavailable
$UseUv = $false
if (Get-Command uv -ErrorAction SilentlyContinue) {
$UseUv = $true
} else {
substep "installing uv package manager..."
try {
Invoke-SetupCommand { Invoke-Expression (Invoke-RestMethod -Uri "https://astral.sh/uv/install.ps1") } | Out-Null
Refresh-Environment
# Re-activate venv since Refresh-Environment rebuilds PATH from
# registry and drops the venv's Scripts directory
. $ActivateScript
if (Get-Command uv -ErrorAction SilentlyContinue) { $UseUv = $true }
} catch { }
}
# Helper: install a package, preferring uv with pip fallback
function Fast-Install {
param([Parameter(ValueFromRemainingArguments=$true)]$Args_)
if ($UseUv) {
$VenvPy = (Get-Command python).Source
$result = & uv pip install --python $VenvPy @Args_ 2>&1
if ($LASTEXITCODE -eq 0) { return }
}
& python -m pip install @Args_ 2>&1
}
# ── 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
# $RepoRoot = (Resolve-Path (Join-Path $ScriptDir "..\..")).Path
# $ReqsSrc = Join-Path $RepoRoot "backend\requirements"
# $ReqsDst = Join-Path $PackageDir "requirements"
# if (-not (Test-Path $ReqsDst)) { New-Item -ItemType Directory -Path $ReqsDst | Out-Null }
# Copy-Item (Join-Path $ReqsSrc "*.txt") $ReqsDst -Force
# Write-Host " Installing CLI entry point..." -ForegroundColor Cyan
# pip install -e $RepoRoot 2>&1 | Out-Null
# } else {
# # Running from pip install: the package is in system Python but not in
# # the fresh .venv. Install it so run_install() can find its modules
# # and bundled requirements files.
# Write-Host " Installing package into venv..." -ForegroundColor Cyan
# pip install unsloth-roland-test 2>&1 | Out-Null
# }
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.
# PyTorch bundles its own CUDA runtime, so this works regardless
# of whether the CUDA Toolkit is installed yet.
# The CUDA tag is chosen based on the driver's max supported CUDA version.
# Windows MAX_PATH (260 chars) causes Triton kernel compilation to fail because
# the auto-generated filenames are extremely long. Use a short cache directory.
$TorchCacheDir = "C:\tc"
if (-not (Test-Path $TorchCacheDir)) { New-Item -ItemType Directory -Path $TorchCacheDir -Force | Out-Null }
$env:TORCHINDUCTOR_CACHE_DIR = $TorchCacheDir
[Environment]::SetEnvironmentVariable('TORCHINDUCTOR_CACHE_DIR', $TorchCacheDir, 'User')
substep "TORCHINDUCTOR_CACHE_DIR set to $TorchCacheDir (avoids MAX_PATH issues)"
if ($HasNvidiaSmi) {
$CuTag = Get-PytorchCudaTag
} else {
$CuTag = "cpu"
}
# ── GPU arch → newest compatible Windows ROCm wheel release ──
# Wheels bundle their own ROCm runtime; the installed HIP SDK version does
# not constrain which release to use. Always picks the newest release that
# supports the GPU architecture.
# ── AMD Windows ROCm torch override ──────────────────────────────────────────
# Uses AMD's arch-specific pip index (repo.amd.com/rocm/whl/{arch}/).
# Wheels bundle their own ROCm runtime; HIP SDK version is irrelevant.
$ROCmGfxArch = $script:ROCmGfxArch
$ROCmIndexUrl = $null
if ($HasROCm -and $CuTag -eq "cpu") {
$amdIndexBase = if ($env:UNSLOTH_ROCM_WINDOWS_MIRROR) { $env:UNSLOTH_ROCM_WINDOWS_MIRROR.TrimEnd('/') } else { "https://repo.amd.com/rocm/whl" }
$archFamilyMap = @{
"gfx1201" = "gfx120X-all"; "gfx1200" = "gfx120X-all" # RDNA 4
"gfx1151" = "gfx1151"; "gfx1150" = "gfx1150" # RDNA 3.5 (Strix Halo/Point)
"gfx1103" = "gfx110X-all"; "gfx1102" = "gfx110X-all" # RDNA 3
"gfx1101" = "gfx110X-all"; "gfx1100" = "gfx110X-all"
"gfx90a" = "gfx90a"; "gfx908" = "gfx908" # MI200/MI100
}
# gfx120X and Strix have a null _grouped_mm kernel on torch <2.11.0.
# Mirrors the $torchFloorMap in install.ps1 so both installers enforce
# the same floor and ceiling when pulling from AMD's per-arch index.
$torchFloorMap = @{
"gfx1201" = "torch>=2.11.0,<2.12.0"; "gfx1200" = "torch>=2.11.0,<2.12.0"
"gfx1151" = "torch>=2.11.0,<2.12.0"; "gfx1150" = "torch>=2.11.0,<2.12.0"
}
# Companion ranges for torchvision/torchaudio -- must stay in sync with the
# torch ceiling so pip can always find a consistent trio on AMD's per-arch
# index. AMD publishes each package independently and may add a newer
# torchvision (e.g. 0.27 for torch 2.12) before removing 0.26, which would
# cause pip to resolve an ABI-incompatible set if these are left bare.
# Matches _ROCM_TORCH_PKG_SPECS["rocm7.2"] in install_python_stack.py.
# Bump all three ceilings together when torch 2.12.x is validated.
$torchvisionFloorMap = @{
"gfx1201" = "torchvision>=0.26.0,<0.27.0"; "gfx1200" = "torchvision>=0.26.0,<0.27.0"
"gfx1151" = "torchvision>=0.26.0,<0.27.0"; "gfx1150" = "torchvision>=0.26.0,<0.27.0"
}
$torchaudioFloorMap = @{
"gfx1201" = "torchaudio>=2.11.0,<2.12.0"; "gfx1200" = "torchaudio>=2.11.0,<2.12.0"
"gfx1151" = "torchaudio>=2.11.0,<2.12.0"; "gfx1150" = "torchaudio>=2.11.0,<2.12.0"
}
$archFamily = if ($ROCmGfxArch -and $archFamilyMap.ContainsKey($ROCmGfxArch)) { $archFamilyMap[$ROCmGfxArch] } else { $null }
$ROCmTorchSpec = if ($ROCmGfxArch -and $torchFloorMap.ContainsKey($ROCmGfxArch)) { $torchFloorMap[$ROCmGfxArch] } else { "torch" }
$ROCmVisionSpec = if ($ROCmGfxArch -and $torchvisionFloorMap.ContainsKey($ROCmGfxArch)) { $torchvisionFloorMap[$ROCmGfxArch] } else { "torchvision" }
$ROCmAudioSpec = if ($ROCmGfxArch -and $torchaudioFloorMap.ContainsKey($ROCmGfxArch)) { $torchaudioFloorMap[$ROCmGfxArch] } else { "torchaudio" }
if ($archFamily) {
$ROCmIndexUrl = "$amdIndexBase/$archFamily/"
} elseif ($ROCmGfxArch) {
# GPU arch detected but not in the supported wheel map — warn explicitly
# so the user knows why they are getting CPU PyTorch instead of ROCm.
substep "[WARN] AMD GPU ($ROCmGfxArch) not in supported arch list -- falling back to CPU-only PyTorch" "Yellow"
substep " Supported: gfx1200/1201 (RDNA 4), gfx1150/1151 (RDNA 3.5), gfx1100-1103 (RDNA 3), gfx90a, gfx908" "Yellow"
} else {
# HIP SDK present ($HasROCm=true via amd-smi) but gcnArchName was not
# readable — warn rather than silently falling back to CPU PyTorch.
substep "[WARN] AMD GPU detected (HIP SDK present) but GPU arch could not be read -- falling back to CPU-only PyTorch" "Yellow"
substep " Arch detection requires hipinfo to report gcnArchName. Re-install the HIP SDK if this is unexpected." "Yellow"
}
}
$PyTorchWhlBase = if ($env:UNSLOTH_PYTORCH_MIRROR) { $env:UNSLOTH_PYTORCH_MIRROR.TrimEnd('/') } else { "https://download.pytorch.org/whl" }
if ($ROCmIndexUrl) {
substep "installing PyTorch (AMD ROCm, $ROCmGfxArch)..."
if ($ROCmTorchSpec -ne "torch") {
substep " enforcing $ROCmTorchSpec $ROCmVisionSpec $ROCmAudioSpec (known _grouped_mm bug in older wheels)" "Cyan"
}
if ($script:UnslothVerbose) {
Fast-Install $ROCmTorchSpec $ROCmVisionSpec $ROCmAudioSpec --force-reinstall --index-url $ROCmIndexUrl
$torchInstallExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install $ROCmTorchSpec $ROCmVisionSpec $ROCmAudioSpec --force-reinstall --index-url $ROCmIndexUrl | Out-String
$torchInstallExit = $LASTEXITCODE
}
if ($torchInstallExit -ne 0) {
Write-Host "[WARN] AMD ROCm PyTorch install failed -- falling back to CPU" -ForegroundColor Yellow
Write-Host $output -ForegroundColor Yellow
$ROCmIndexUrl = $null
} else {
# Tell install_python_stack.py to skip probe + suppress manual-install warning.
$env:UNSLOTH_ROCM_TORCH_INSTALLED = "1"
}
}
if (-not $ROCmIndexUrl -and $CuTag -eq "cpu") {
substep "installing PyTorch (CPU-only)..."
if ($script:UnslothVerbose) {
Fast-Install torch torchvision torchaudio --index-url "$PyTorchWhlBase/cpu"
$torchInstallExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install torch torchvision torchaudio --index-url "$PyTorchWhlBase/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
}
} elseif (-not $ROCmIndexUrl) {
substep "installing PyTorch with CUDA support ($CuTag)..."
substep "(This download is ~2.8 GB -- may take a few minutes)"
if ($script:UnslothVerbose) {
Fast-Install torch torchvision torchaudio --index-url "$PyTorchWhlBase/$CuTag"
$torchInstallExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install torch torchvision torchaudio --index-url "$PyTorchWhlBase/$CuTag" | Out-String
$torchInstallExit = $LASTEXITCODE
}
if ($torchInstallExit -ne 0) {
Write-Host "[FAILED] PyTorch CUDA install failed (exit code $torchInstallExit)" -ForegroundColor Red
Write-Host $output -ForegroundColor Red
exit 1
}
# Install Triton for Windows (enables torch.compile -- without it training can hang)
substep "installing Triton for Windows..."
if ($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 {
substep "Triton for Windows installed (enables torch.compile)"
}
}
# Rename running unsloth.exe so pip can replace it (Windows refuses to delete a mapped .exe).
$VenvScriptsDir = Join-Path $VenvDir "Scripts"
$RunningUnslothExe = Join-Path $VenvScriptsDir "unsloth.exe"
if (Test-Path -LiteralPath $RunningUnslothExe -PathType Leaf) {
$StaleUnslothExe = "$RunningUnslothExe.deleteme"
if (Test-Path -LiteralPath $StaleUnslothExe) {
Remove-Item -LiteralPath $StaleUnslothExe -Force -ErrorAction SilentlyContinue
}
try {
Rename-Item -LiteralPath $RunningUnslothExe -NewName "unsloth.exe.deleteme" -Force -ErrorAction Stop
} catch {
substep "could not rename unsloth.exe ($($_.Exception.Message)); pip may fail with WinError 32" "Yellow"
}
}
# Ordered heavy dependency installation -- shared cross-platform script
substep "running ordered dependency installation..."
python "$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 $stackExit)" -ForegroundColor Red
Write-Host " Re-run the installer or check the error above for details." -ForegroundColor Red
# Restore the pre-rename unsloth.exe so the user keeps a working CLI.
# Treat a zero-byte exe as "pip half-wrote a broken binary" -- prefer the
# stale-but-working copy in .deleteme.
if (Test-Path -LiteralPath "$RunningUnslothExe.deleteme") {
$needRestore = -not (Test-Path -LiteralPath $RunningUnslothExe)
if (-not $needRestore) {
try {
$needRestore = (Get-Item -LiteralPath $RunningUnslothExe -ErrorAction Stop).Length -eq 0
} catch { $needRestore = $true }
}
if ($needRestore) {
try {
if (Test-Path -LiteralPath $RunningUnslothExe) {
Remove-Item -LiteralPath $RunningUnslothExe -Force -ErrorAction SilentlyContinue
}
Rename-Item -LiteralPath "$RunningUnslothExe.deleteme" -NewName "unsloth.exe" -Force -ErrorAction Stop
substep "restored unsloth.exe after failed install"
} catch {
substep "could not restore unsloth.exe ($($_.Exception.Message))" "Yellow"
}
}
}
exit 1
}
} else {
step "python" "dependencies up to date"
# Restore ErrorActionPreference (was lowered for pip/python section)
$ErrorActionPreference = $prevEAP
}
# ── Pre-install transformers 5.x into .venv_t5_530/ and .venv_t5_550/ ──
# Runs outside the deps fast-path gate so that upgrades from the legacy
# single .venv_t5 are always migrated to the tiered layout.
# T5 sidecar venvs live under the resolved $StudioHome so custom installs are self-contained.
$VenvT5_530Dir = Join-Path $StudioHome ".venv_t5_530"
$VenvT5_550Dir = Join-Path $StudioHome ".venv_t5_550"
$VenvT5Legacy = Join-Path $StudioHome ".venv_t5"
$_NeedT5Install = $false
if (Test-Path -LiteralPath $VenvT5Legacy) {
Assert-StudioOwnedOrAbsent -Path $VenvT5Legacy -Label "legacy transformers sidecar venv"
Remove-Item -LiteralPath $VenvT5Legacy -Recurse -Force
$_NeedT5Install = $true
}
if (-not (Test-Path -LiteralPath $VenvT5_530Dir)) { $_NeedT5Install = $true }
if (-not (Test-Path -LiteralPath $VenvT5_550Dir)) { $_NeedT5Install = $true }
# Also reinstall when python deps were updated
if (-not $SkipPythonDeps) { $_NeedT5Install = $true }
if ($_NeedT5Install) {
Write-Host ""
$prevEAP_t5 = $ErrorActionPreference
$ErrorActionPreference = "Continue"
# --- .venv_t5_530 (transformers 5.3.0) ---
substep "pre-installing transformers 5.3.0 for newer model support..."
Assert-StudioOwnedOrAbsent -Path $VenvT5_530Dir -Label "transformers 5.3 sidecar venv"
if (Test-Path -LiteralPath $VenvT5_530Dir) { Remove-Item -LiteralPath $VenvT5_530Dir -Recurse -Force }
[System.IO.Directory]::CreateDirectory($VenvT5_530Dir) | Out-Null
Mark-StudioOwned -Path $VenvT5_530Dir
foreach ($pkg in @("transformers==5.3.0", "huggingface_hub==1.8.0", "hf_xet==1.4.2")) {
if ($script:UnslothVerbose) {
Fast-Install --target $VenvT5_530Dir --no-deps $pkg
$t5PkgExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install --target $VenvT5_530Dir --no-deps $pkg | Out-String
$t5PkgExit = $LASTEXITCODE
}
if ($t5PkgExit -ne 0) {
Write-Host "[FAIL] Could not install $pkg into .venv_t5_530/" -ForegroundColor Red
Write-Host $output -ForegroundColor Red
$ErrorActionPreference = $prevEAP_t5
exit 1
}
}
if ($script:UnslothVerbose) {
Fast-Install --target $VenvT5_530Dir tiktoken
$tiktokenInstallExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install --target $VenvT5_530Dir tiktoken | Out-String
$tiktokenInstallExit = $LASTEXITCODE
}
if ($tiktokenInstallExit -ne 0) {
substep "Could not install tiktoken into .venv_t5_530/ -- Qwen tokenizers may fail" "Yellow"
}
step "transformers" "5.3.0 pre-installed"
# --- .venv_t5_550 (transformers 5.5.0) ---
substep "pre-installing transformers 5.5.0 for Gemma 4 support..."
Assert-StudioOwnedOrAbsent -Path $VenvT5_550Dir -Label "transformers 5.5 sidecar venv"
if (Test-Path -LiteralPath $VenvT5_550Dir) { Remove-Item -LiteralPath $VenvT5_550Dir -Recurse -Force }
[System.IO.Directory]::CreateDirectory($VenvT5_550Dir) | Out-Null
Mark-StudioOwned -Path $VenvT5_550Dir
foreach ($pkg in @("transformers==5.5.0", "huggingface_hub==1.8.0", "hf_xet==1.4.2")) {
if ($script:UnslothVerbose) {
Fast-Install --target $VenvT5_550Dir --no-deps $pkg
$t5PkgExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install --target $VenvT5_550Dir --no-deps $pkg | Out-String
$t5PkgExit = $LASTEXITCODE
}
if ($t5PkgExit -ne 0) {
Write-Host "[FAIL] Could not install $pkg into .venv_t5_550/" -ForegroundColor Red
Write-Host $output -ForegroundColor Red
$ErrorActionPreference = $prevEAP_t5
exit 1
}
}
if ($script:UnslothVerbose) {
Fast-Install --target $VenvT5_550Dir tiktoken
$tiktokenInstallExit = $LASTEXITCODE
$output = ""
} else {
$output = Fast-Install --target $VenvT5_550Dir tiktoken | Out-String
$tiktokenInstallExit = $LASTEXITCODE
}
if ($tiktokenInstallExit -ne 0) {
substep "Could not install tiktoken into .venv_t5_550/ -- Qwen tokenizers may fail" "Yellow"
}
$ErrorActionPreference = $prevEAP_t5
step "transformers" "5.5.0 pre-installed"
} # end $_NeedT5Install
# ==========================================================================
# PHASE 3.4: Prefer prebuilt llama.cpp bundles before source build
# ==========================================================================
# Nest llama.cpp under $StudioHome only for real env-overrides, never the
# legacy default. Reuses $StudioHomeIsCustom from the canonical comparison
# computed above so the llama.cpp nest matches ownership-guard semantics.
if ($StudioHomeIsCustom) {
$UnslothHome = $StudioHome
} else {
$UnslothHome = Join-Path $env:USERPROFILE ".unsloth"
}
if (-not (Test-Path -LiteralPath $UnslothHome)) { [System.IO.Directory]::CreateDirectory($UnslothHome) | Out-Null }
$LlamaCppDir = Join-Path $UnslothHome "llama.cpp"
$NeedLlamaSourceBuild = $false
$SkipPrebuiltInstall = $false
$RequestedLlamaTag = if ($env:UNSLOTH_LLAMA_TAG) { $env:UNSLOTH_LLAMA_TAG } else { $DefaultLlamaTag }
$HelperReleaseRepo = "ggml-org/llama.cpp"
$LlamaPr = if ($env:UNSLOTH_LLAMA_PR) { $env:UNSLOTH_LLAMA_PR.Trim() } else { "" }
$LlamaPrForce = if ($env:UNSLOTH_LLAMA_PR_FORCE) { $env:UNSLOTH_LLAMA_PR_FORCE.Trim() } else { $DefaultLlamaPrForce }
$LlamaSource = $DefaultLlamaSource
if ($LlamaSource.EndsWith('.git')) { $LlamaSource = $LlamaSource.Substring(0, $LlamaSource.Length - 4) }
$ResolvedSourceUrl = $LlamaSource
$ResolvedSourceRef = $RequestedLlamaTag
$ResolvedSourceRefKind = "tag"
$ResolvedLlamaTag = $RequestedLlamaTag
if ($env:UNSLOTH_LLAMA_FORCE_COMPILE -eq "1") {
$NeedLlamaSourceBuild = $true
$SkipPrebuiltInstall = $true
}
function Invoke-LlamaHelper {
param(
[string[]]$Arguments,
[string]$StderrPath = $null
)
$previousErrorActionPreference = $ErrorActionPreference
$previousNativeErrorPreference = $null
$restoreNativeErrorPreference = $false
$ErrorActionPreference = "Continue"
if ($PSVersionTable.PSVersion.Major -ge 7) {
$previousNativeErrorPreference = $PSNativeCommandUseErrorActionPreference
$PSNativeCommandUseErrorActionPreference = $false
$restoreNativeErrorPreference = $true
}
try {
# Capture all output (stdout + stderr) so that PowerShell does not
# convert stderr lines into visible ErrorRecord objects. Separate
# stdout from stderr afterwards.
$allOutput = & python "$PSScriptRoot\install_llama_prebuilt.py" @Arguments 2>&1
$exitCode = $LASTEXITCODE
$stdoutLines = @()
$stderrLines = @()
foreach ($line in $allOutput) {
if ($line -is [System.Management.Automation.ErrorRecord]) {
$stderrLines += $line.ToString()
} else {
$stdoutLines += $line
}
}
if ($StderrPath -and $stderrLines.Count -gt 0) {
$stderrLines | Out-File -FilePath $StderrPath -Encoding utf8
}
return [pscustomobject]@{
Output = $stdoutLines
ExitCode = $exitCode
}
} finally {
$ErrorActionPreference = $previousErrorActionPreference
if ($restoreNativeErrorPreference) {
$PSNativeCommandUseErrorActionPreference = $previousNativeErrorPreference
}
}
}
if ($LlamaSource -ne "https://github.com/ggml-org/llama.cpp") {
step "llama.cpp" "custom source: $LlamaSource -- forcing source build" "Yellow"
$NeedLlamaSourceBuild = $true
$SkipPrebuiltInstall = $true
}
if (-not $LlamaPr -and $LlamaPrForce -and $LlamaPrForce -match '^\d+$' -and [int]$LlamaPrForce -gt 0) {
$LlamaPr = $LlamaPrForce
step "llama.cpp" "baked-in PR_FORCE=$LlamaPrForce" "Yellow"
}
if ($LlamaPr) {
if ($LlamaPr -notmatch '^\d+$' -or [int]$LlamaPr -le 0) {
Write-Host "[ERROR] UNSLOTH_LLAMA_PR=$LlamaPr is not a valid PR number" -ForegroundColor Red
exit 1
}
step "llama.cpp" "UNSLOTH_LLAMA_PR=$LlamaPr -- will build from PR head" "Yellow"
$ResolvedLlamaTag = "pr-$LlamaPr"
$ResolvedSourceUrl = $LlamaSource
$ResolvedSourceRef = "pr-$LlamaPr"
$ResolvedSourceRefKind = "pull"
$NeedLlamaSourceBuild = $true
$SkipPrebuiltInstall = $true
}
if ($env:UNSLOTH_LLAMA_FORCE_COMPILE -eq "1") {
Write-Host ""
substep "UNSLOTH_LLAMA_FORCE_COMPILE=1 -- skipping prebuilt llama.cpp install" "Yellow"
$NeedLlamaSourceBuild = $true
} elseif ($SkipPrebuiltInstall) {
Write-Host ""
substep "Skipping prebuilt install -- falling back to source build" "Yellow"
} else {
Write-Host ""
substep "installing prebuilt llama.cpp bundle (preferred path)..."
if (Test-Path -LiteralPath $LlamaCppDir) {
substep "Existing llama.cpp install detected -- validating staged prebuilt update before replacement"
# If the existing install is the wrong kind (e.g. windows-cpu on a ROCm
# machine that should have windows-hip), remove it so the installer is
# forced to download the correct variant rather than skipping on tag match.
$existingMetaPath = Join-Path $LlamaCppDir "UNSLOTH_PREBUILT_INFO.json"
if (Test-Path $existingMetaPath) {
try {
$existingMeta = Get-Content $existingMetaPath -Raw | ConvertFrom-Json
$existingKind = $existingMeta.install_kind
$expectedKind = if ($HasROCm) { "windows-hip" } elseif ($HasNvidiaSmi) { "windows-cuda" } else { "windows-cpu" }
if ($existingKind -and $existingKind -ne $expectedKind) {
substep "Removing mismatched llama.cpp install (found '$existingKind', need '$expectedKind')..."
Remove-Item -Recurse -Force -LiteralPath $LlamaCppDir -ErrorAction SilentlyContinue
}
} catch {
# unreadable metadata -- let the installer handle it
}
}
}
# why: install_llama_prebuilt.py uses os.replace(), which would displace
# an unrelated $env:UNSLOTH_STUDIO_HOME\llama.cpp before the source-build
# ownership check below ever runs.
if ($StudioHomeIsCustom) {
Assert-StudioOwnedOrAbsent -Path $LlamaCppDir -Label "llama.cpp install"
}
$prebuiltArgs = @(
"$PSScriptRoot\install_llama_prebuilt.py",
"--install-dir", $LlamaCppDir,
"--llama-tag", $RequestedLlamaTag,
"--published-repo", $HelperReleaseRepo,
"--simple-policy"
)
if ($HasROCm) {
$prebuiltArgs += "--has-rocm"
}
if ($env:UNSLOTH_LLAMA_RELEASE_TAG) {
$prebuiltArgs += @("--published-release-tag", $env:UNSLOTH_LLAMA_RELEASE_TAG)
}
$prevEAPPrebuilt = $ErrorActionPreference
$ErrorActionPreference = "Continue"
$previousNativeErrorPreference = $null
$restoreNativeErrorPreference = $false
if ($PSVersionTable.PSVersion.Major -ge 7) {
$previousNativeErrorPreference = $PSNativeCommandUseErrorActionPreference
$PSNativeCommandUseErrorActionPreference = $false
$restoreNativeErrorPreference = $true
}
try {
if ($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 = $LASTEXITCODE
}
} finally {
if ($restoreNativeErrorPreference) {
$PSNativeCommandUseErrorActionPreference = $previousNativeErrorPreference
}
}
$ErrorActionPreference = $prevEAPPrebuilt
if ($prebuiltExit -eq 0) {
if ($prebuiltOutput -match "already matches") {
step "llama.cpp" "prebuilt up to date and validated"
} else {
step "llama.cpp" "prebuilt installed and validated"
}
if ($StudioHomeIsCustom -and (Test-Path -LiteralPath $LlamaCppDir -PathType Container)) {
Mark-StudioOwned -Path $LlamaCppDir
}
$installedRelease = Get-InstalledLlamaPrebuiltRelease -InstallDir $LlamaCppDir
if ($installedRelease) {
substep $installedRelease
}
} elseif ($prebuiltExit -eq 3) {
step "llama.cpp" "install blocked by active llama.cpp process" "Yellow"
Write-LlamaFailureLog -Output $prebuiltOutput
if (Test-Path -LiteralPath $LlamaCppDir) {
substep "Existing install was restored" "Yellow"
}
substep "Close Studio or other llama.cpp users and retry" "Yellow"
exit 3
} else {
step "llama.cpp" "prebuilt install failed (continuing)" "Yellow"
Write-LlamaFailureLog -Output $prebuiltOutput
if (Test-Path -LiteralPath $LlamaCppDir) {
substep "Prebuilt update failed; existing install was restored or cleaned before source build fallback" "Yellow"
}
substep "Prebuilt llama.cpp path unavailable or failed validation -- falling back to source build" "Yellow"
$NeedLlamaSourceBuild = $true
}
}
# ==========================================================================
# PHASE 3.5: Install OpenSSL dev (for HTTPS support in llama-server)
# ==========================================================================
# llama-server needs OpenSSL to download models from HuggingFace via -hf.
# ShiningLight.OpenSSL.Dev includes headers + libs that cmake can find.
$OpenSslAvailable = $false
if ($NeedLlamaSourceBuild) {
# Check if OpenSSL dev is already installed (look for include dir)
$OpenSslRoots = @(
'C:\Program Files\OpenSSL-Win64',
'C:\Program Files\OpenSSL',
'C:\OpenSSL-Win64'
)
$OpenSslRoot = $null
foreach ($root in $OpenSslRoots) {
if (Test-Path (Join-Path $root 'include\openssl\ssl.h')) {
$OpenSslRoot = $root
break
}
}
if ($OpenSslRoot) {
$OpenSslAvailable = $true
substep "OpenSSL dev found at $OpenSslRoot"
} else {
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
# Re-check after install
foreach ($root in $OpenSslRoots) {
if (Test-Path (Join-Path $root 'include\openssl\ssl.h')) {
$OpenSslRoot = $root
$OpenSslAvailable = $true
substep "OpenSSL dev installed at $OpenSslRoot"
break
}
}
}
if (-not $OpenSslAvailable) {
substep "OpenSSL dev not available -- llama-server will be built without HTTPS" "Yellow"
}
}
} else {
substep "OpenSSL dev install skipped -- prebuilt llama.cpp already validated" "Yellow"
}
# ==========================================================================
# PHASE 4: Build llama.cpp with CUDA for GGUF inference + export
# ==========================================================================
# Builds at ~/.unsloth/llama.cpp — a single shared location under the user's
# home directory. This is used by both the inference server and the GGUF
# export pipeline (unsloth-zoo).
# We build:
# - llama-server: for GGUF model inference (with HTTPS if OpenSSL available)
# - llama-quantize: for GGUF export quantization
# Prerequisites (git, cmake, VS Build Tools, CUDA Toolkit) already installed in Phase 1.
$OriginalLlamaCppDir = $LlamaCppDir
$BuildDir = Join-Path $LlamaCppDir "build"
$LlamaServerBin = Join-Path $BuildDir "bin\Release\llama-server.exe"
$HasCmakeForBuild = $null -ne (Get-Command cmake -ErrorAction SilentlyContinue)
# Check if existing llama-server matches current GPU mode. A CUDA-built binary
# on a now-CPU-only machine (or vice versa) needs to be rebuilt.
$NeedRebuild = $false
if (Test-Path -LiteralPath $LlamaServerBin) {
$CmakeCacheFile = Join-Path $BuildDir "CMakeCache.txt"
if (Test-Path -LiteralPath $CmakeCacheFile) {
$cachedCuda = Select-String -LiteralPath $CmakeCacheFile -Pattern 'GGML_CUDA:BOOL=ON' -Quiet
if ($HasNvidiaSmi -and -not $cachedCuda) {
Write-Host " Existing llama-server is CPU-only but GPU is available -- rebuilding" -ForegroundColor Yellow
$NeedRebuild = $true
} elseif (-not $HasNvidiaSmi -and $cachedCuda) {
Write-Host " Existing llama-server was built with CUDA but no GPU detected -- rebuilding" -ForegroundColor Yellow
$NeedRebuild = $true
}
}
}
if (-not $NeedLlamaSourceBuild) {
Write-Host ""
step "llama.cpp" "prebuilt (validated)"
} elseif ((Test-Path -LiteralPath $LlamaServerBin) -and -not $NeedRebuild -and $RequestedLlamaTag -ne "master") {
# Skip rebuild only for pinned tags (e.g. b8635). When the requested
# tag is "master" (a moving target), always rebuild so the binary picks
# up new model architecture support (e.g. Gemma 4).
Write-Host ""
step "llama.cpp" "already built"
} elseif (-not $HasCmakeForBuild) {
Write-Host ""
if (-not $HasNvidiaSmi) {
# CPU-only machines depend entirely on llama-server for GGUF chat -- cmake is required
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"
}
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) {
substep "building llama.cpp with CUDA support..."
} else {
substep "building llama.cpp (CPU-only, no NVIDIA GPU detected)..."
}
substep "This typically takes 5-10 minutes on first build."
Write-Host ""
# Start total build timer
$totalSw = [System.Diagnostics.Stopwatch]::StartNew()
# Native commands (git, cmake) write to stderr even on success.
# With $ErrorActionPreference = "Stop" (set at top of script), PS 5.1
# converts stderr lines into terminating ErrorRecords, breaking output.
# Lower to "Continue" for the build section.
$prevEAP = $ErrorActionPreference
$ErrorActionPreference = "Continue"
$BuildOk = $true
$FailedStep = ""
# Re-sanitize CUDA_PATH_V* vars — Refresh-Environment (called during
# Node/Python installs above) may have repopulated conflicting versioned
# vars from the Machine registry.
if ($HasNvidiaSmi -and $CudaToolkitRoot) {
$cudaPathVars2 = @([Environment]::GetEnvironmentVariables('Process').Keys | Where-Object { $_ -match '^CUDA_PATH_V' })
foreach ($v2 in $cudaPathVars2) {
[Environment]::SetEnvironmentVariable($v2, $null, 'Process')
}
$tkDirName2 = Split-Path $CudaToolkitRoot -Leaf
if ($tkDirName2 -match '^v(\d+)\.(\d+)') {
[Environment]::SetEnvironmentVariable("CUDA_PATH_V$($Matches[1])_$($Matches[2])", $CudaToolkitRoot, 'Process')
}
# Also re-assert CUDA_PATH and CudaToolkitDir in case they were overwritten
[Environment]::SetEnvironmentVariable('CUDA_PATH', $CudaToolkitRoot, 'Process')
[Environment]::SetEnvironmentVariable('CudaToolkitDir', "$CudaToolkitRoot\", 'Process')
}
if (-not $LlamaPr) {
$ResolvedSourceUrl = $LlamaSource
if ($env:UNSLOTH_LLAMA_FORCE_COMPILE -eq "1") {
if ($RequestedLlamaTag -eq "latest") {
$ResolvedSourceRef = if ($env:UNSLOTH_LLAMA_FORCE_COMPILE_REF) {
$env:UNSLOTH_LLAMA_FORCE_COMPILE_REF
} else {
$DefaultLlamaForceCompileRef
}
$ResolvedSourceRefKind = "branch"
} else {
$ResolvedSourceRef = $RequestedLlamaTag
$ResolvedSourceRefKind = "tag"
}
} elseif ($RequestedLlamaTag -eq "latest") {
$resolveTagArgs = @("--resolve-llama-tag", "latest", "--published-repo", "ggml-org/llama.cpp", "--output-format", "json")
$resolveTagResult = Invoke-LlamaHelper -Arguments $resolveTagArgs
$resolveTagOutput = $resolveTagResult.Output
$resolveTagExit = $resolveTagResult.ExitCode
if ($resolveTagExit -eq 0 -and $resolveTagOutput) {
try {
$ResolvedSourceRef = (($resolveTagOutput | Out-String) | ConvertFrom-Json).llama_tag
} catch {
$ResolvedSourceRef = ""
}
} else {
$ResolvedSourceRef = ""
}
if ([string]::IsNullOrWhiteSpace($ResolvedSourceRef)) {
$ResolvedSourceRef = "latest"
}
$ResolvedSourceRefKind = "tag"
} else {
$ResolvedSourceRef = $RequestedLlamaTag
$ResolvedSourceRefKind = "tag"
}
if ([string]::IsNullOrWhiteSpace($ResolvedSourceUrl)) { $ResolvedSourceUrl = $LlamaSource }
if ([string]::IsNullOrWhiteSpace($ResolvedSourceRef)) { $ResolvedSourceRef = $RequestedLlamaTag }
}
# -- Step A: Clone or pull llama.cpp --
$UseConcreteRef = ($ResolvedSourceRef -ne "latest" -and -not [string]::IsNullOrWhiteSpace($ResolvedSourceRef))
if (Test-Path -LiteralPath (Join-Path $LlamaCppDir ".git")) {
# why: in-place git mutation (remote set-url, checkout -B, clean -fdx)
# rewrites $LlamaCppDir; mirror the prebuilt and temp-dir-swap guards
# so an unrelated workspace .git tree is never silently overwritten.
if ($StudioHomeIsCustom) {
Assert-StudioOwnedOrAbsent -Path $LlamaCppDir -Label "llama.cpp install"
}
Write-Host " Syncing llama.cpp to $ResolvedSourceRef..." -ForegroundColor Gray
# Always sync the remote URL so switching between default/fork sources works
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir remote set-url origin "$ResolvedSourceUrl.git" } | Out-Null
if ($LlamaPr) {
$gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin "pull/$LlamaPr/head" }
if ($gitFetchExit -ne 0) {
$BuildOk = $false
$FailedStep = "git fetch PR #$LlamaPr"
} else {
$gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B "pr-$LlamaPr" FETCH_HEAD }
if ($gitCheckoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout PR #$LlamaPr"
} else {
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
}
}
} elseif ($ResolvedSourceRefKind -eq "pull") {
$gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin $ResolvedSourceRef }
if ($gitFetchExit -ne 0) {
substep "git fetch failed -- using existing source" "Yellow"
} else {
$gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD }
if ($gitCheckoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout"
} else {
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
}
}
} elseif ($ResolvedSourceRefKind -eq "commit") {
$gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin $ResolvedSourceRef }
if ($gitFetchExit -ne 0) {
substep "git fetch failed -- using existing source" "Yellow"
} else {
$gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD }
if ($gitCheckoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout"
} else {
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
}
}
} elseif ($UseConcreteRef) {
$gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin $ResolvedSourceRef }
if ($gitFetchExit -ne 0) {
substep "git fetch failed -- using existing source" "Yellow"
} else {
$gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD }
if ($gitCheckoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout"
} else {
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
}
}
} else {
$gitFetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir fetch --depth 1 origin }
if ($gitFetchExit -ne 0) {
substep "git fetch failed -- using existing source" "Yellow"
} else {
$gitCheckoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir checkout -B unsloth-llama-build FETCH_HEAD }
if ($gitCheckoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout"
} else {
Invoke-SetupCommand -AlwaysQuiet { git -C $LlamaCppDir clean -fdx } | Out-Null
}
}
}
# why: in-place git-sync (the temp-dir clone path calls Mark-StudioOwned
# at swap-time) must mark the existing tree so a subsequent prebuilt
# update path's Assert-StudioOwnedOrAbsent does not exit on the same root.
if ($BuildOk -and $StudioHomeIsCustom) {
Mark-StudioOwned -Path $LlamaCppDir
}
} else {
Write-Host " Cloning llama.cpp @ $ResolvedSourceRef..." -ForegroundColor Gray
$buildTmp = "$LlamaCppDir.build.$PID"
$null = [System.IO.Directory]::CreateDirectory((Split-Path -LiteralPath $LlamaCppDir))
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
if ($LlamaPr) {
$cloneExit = Invoke-SetupCommand -AlwaysQuiet { git clone --depth 1 "$LlamaSource.git" $buildTmp }
if ($cloneExit -ne 0) {
$BuildOk = $false
$FailedStep = "git clone"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
if ($BuildOk) {
$fetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp fetch --depth 1 origin "pull/$LlamaPr/head:pr-$LlamaPr" }
if ($fetchExit -ne 0) {
$BuildOk = $false
$FailedStep = "git fetch PR #$LlamaPr"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
if ($BuildOk) {
$checkoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp checkout "pr-$LlamaPr" }
if ($checkoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout PR #$LlamaPr"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
} elseif ($ResolvedSourceRefKind -eq "pull") {
$cloneExit = Invoke-SetupCommand -AlwaysQuiet { git clone --depth 1 "$ResolvedSourceUrl.git" $buildTmp }
if ($cloneExit -ne 0) {
$BuildOk = $false
$FailedStep = "git clone"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
if ($BuildOk) {
$fetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp fetch --depth 1 origin $ResolvedSourceRef }
if ($fetchExit -ne 0) {
$BuildOk = $false
$FailedStep = "git fetch source PR ref"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
if ($BuildOk) {
$checkoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp checkout -B unsloth-llama-build FETCH_HEAD }
if ($checkoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout source PR ref"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
} elseif ($ResolvedSourceRefKind -eq "commit") {
$cloneExit = Invoke-SetupCommand -AlwaysQuiet { git clone --depth 1 "$ResolvedSourceUrl.git" $buildTmp }
if ($cloneExit -ne 0) {
$BuildOk = $false
$FailedStep = "git clone"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
if ($BuildOk) {
$fetchExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp fetch --depth 1 origin $ResolvedSourceRef }
if ($fetchExit -ne 0) {
$BuildOk = $false
$FailedStep = "git fetch source commit"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
if ($BuildOk) {
$checkoutExit = Invoke-SetupCommand -AlwaysQuiet { git -C $buildTmp checkout -B unsloth-llama-build FETCH_HEAD }
if ($checkoutExit -ne 0) {
$BuildOk = $false
$FailedStep = "git checkout source commit"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
} else {
$cloneArgs = @("clone", "--depth", "1")
if ($UseConcreteRef) {
$cloneArgs += @("--branch", $ResolvedSourceRef)
}
$cloneArgs += @("$ResolvedSourceUrl.git", $buildTmp)
$cloneExit = Invoke-SetupCommand -AlwaysQuiet { git @cloneArgs }
if ($cloneExit -ne 0) {
$BuildOk = $false
$FailedStep = "git clone"
if (Test-Path -LiteralPath $buildTmp) { Remove-Item -LiteralPath $buildTmp -Recurse -Force }
}
}
# Use temp dir for build; swap into $LlamaCppDir only after build succeeds
if ($BuildOk) {
$LlamaCppDir = $buildTmp
$BuildDir = Join-Path $LlamaCppDir "build"
}
}
# -- Step B: cmake configure --
if ($BuildOk) {
Write-Host ""
Write-Host "--- cmake configure ---" -ForegroundColor Cyan
$CmakeArgs = @(
'-S', $LlamaCppDir,
'-B', $BuildDir,
'-G', $CmakeGenerator,
'-Wno-dev'
)
# Tell cmake exactly where VS is (bypasses registry lookup)
if ($VsInstallPath) {
$CmakeArgs += "-DCMAKE_GENERATOR_INSTANCE=$VsInstallPath"
}
# Common flags
$CmakeArgs += '-DBUILD_SHARED_LIBS=OFF'
$CmakeArgs += '-DLLAMA_BUILD_TESTS=OFF'
$CmakeArgs += '-DLLAMA_BUILD_EXAMPLES=OFF'
$CmakeArgs += '-DLLAMA_BUILD_SERVER=ON'
$CmakeArgs += '-DGGML_NATIVE=ON'
# HTTPS support via OpenSSL
if ($OpenSslAvailable -and $OpenSslRoot) {
$CmakeArgs += "-DOPENSSL_ROOT_DIR=$OpenSslRoot"
$CmakeArgs += '-DLLAMA_OPENSSL=ON'
} else {
$CmakeArgs += '-DLLAMA_CURL=OFF'
}
$CmakeArgs += '-DCMAKE_EXE_LINKER_FLAGS=/NODEFAULTLIB:LIBCMT'
# CUDA flags -- only if GPU available, otherwise explicitly disable
if ($HasNvidiaSmi -and $NvccPath) {
$CmakeArgs += '-DGGML_CUDA=ON'
# Accept a host MSVC newer than nvcc's whitelist; a fresh toolkit
# (e.g. CUDA 13.3) otherwise aborts with "#error -- unsupported
# Microsoft Visual Studio version!". Mirrors the Linux fix. Via env
# (covers the configure probe + build), after Refresh-Environment, idempotent.
$nvccAllowFlag = '-allow-unsupported-compiler'
if ([string]::IsNullOrEmpty($env:NVCC_PREPEND_FLAGS)) {
$env:NVCC_PREPEND_FLAGS = $nvccAllowFlag
} elseif ($env:NVCC_PREPEND_FLAGS -notlike "*$nvccAllowFlag*") {
$env:NVCC_PREPEND_FLAGS = "$($env:NVCC_PREPEND_FLAGS) $nvccAllowFlag"
}
substep "NVCC_PREPEND_FLAGS = $env:NVCC_PREPEND_FLAGS"
$CmakeArgs += "-DCUDAToolkit_ROOT=$CudaToolkitRoot"
$CmakeArgs += "-DCUDA_TOOLKIT_ROOT_DIR=$CudaToolkitRoot"
$CmakeArgs += "-DCMAKE_CUDA_COMPILER=$NvccPath"
if ($CudaArch) {
# Validate nvcc actually supports this architecture
if (Test-NvccArchSupport -NvccExe $NvccPath -Arch $CudaArch) {
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$CudaArch"
} else {
# GPU arch too new for this toolkit -- fall back to highest supported.
# PTX forward-compatibility will JIT-compile for the actual GPU at runtime.
$maxArch = Get-NvccMaxArch -NvccExe $NvccPath
if ($maxArch) {
$CmakeArgs += "-DCMAKE_CUDA_ARCHITECTURES=$maxArch"
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
}
}
} else {
$CmakeArgs += '-DGGML_CUDA=OFF'
}
$cmakeOutput = cmake @CmakeArgs 2>&1 | Out-String
$cmakeConfigureExit = $LASTEXITCODE
if ($cmakeConfigureExit -ne 0) {
$BuildOk = $false
$FailedStep = "cmake configure"
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
Write-Host " Copy contents of:" -ForegroundColor Yellow
Write-Host " <CUDA_PATH>\extras\visual_studio_integration\MSBuildExtensions" -ForegroundColor Yellow
Write-Host " into:" -ForegroundColor Yellow
Write-Host " <VS_PATH>\MSBuild\Microsoft\VC\v170\BuildCustomizations" -ForegroundColor Yellow
}
}
}
# -- Step C: Build llama-server --
$NumCpu = [Environment]::ProcessorCount
if ($NumCpu -lt 1) { $NumCpu = 4 }
if ($BuildOk) {
Write-Host ""
Write-Host "--- cmake build (llama-server) ---" -ForegroundColor Cyan
Write-Host " Parallel jobs: $NumCpu" -ForegroundColor Gray
Write-Host ""
$output = cmake --build $BuildDir --config Release --target llama-server -j $NumCpu 2>&1 | Out-String
$cmakeBuildServerExit = $LASTEXITCODE
if ($cmakeBuildServerExit -ne 0) {
$BuildOk = $false
$FailedStep = "cmake build (llama-server)"
Write-LlamaFailureLog -Output $output
}
}
# -- Step D: Build llama-quantize (optional, best-effort) --
if ($BuildOk) {
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
$cmakeBuildQuantizeExit = $LASTEXITCODE
if ($cmakeBuildQuantizeExit -ne 0) {
substep "llama-quantize build failed (GGUF export may be unavailable)" "Yellow"
Write-LlamaFailureLog -Output $output
}
}
# Swap temp build dir into final location (only if we built in a temp dir)
if ($BuildOk -and $LlamaCppDir -ne $OriginalLlamaCppDir) {
Assert-StudioOwnedOrAbsent -Path $OriginalLlamaCppDir -Label "llama.cpp install"
if (Test-Path -LiteralPath $OriginalLlamaCppDir) { Remove-Item -LiteralPath $OriginalLlamaCppDir -Recurse -Force }
Move-Item -LiteralPath $LlamaCppDir -Destination $OriginalLlamaCppDir
$LlamaCppDir = $OriginalLlamaCppDir
$BuildDir = Join-Path $LlamaCppDir "build"
$LlamaServerBin = Join-Path $BuildDir "bin\Release\llama-server.exe"
Mark-StudioOwned -Path $LlamaCppDir
} elseif (-not $BuildOk -and $LlamaCppDir -ne $OriginalLlamaCppDir) {
# Build failed -- clean up temp dir, preserve existing install
if (Test-Path -LiteralPath $LlamaCppDir) { Remove-Item -LiteralPath $LlamaCppDir -Recurse -Force }
$LlamaCppDir = $OriginalLlamaCppDir
$BuildDir = Join-Path $LlamaCppDir "build"
$LlamaServerBin = Join-Path $BuildDir "bin\Release\llama-server.exe"
}
# Restore ErrorActionPreference
$ErrorActionPreference = $prevEAP
# Stop timer
$totalSw.Stop()
$totalMin = [math]::Floor($totalSw.Elapsed.TotalMinutes)
$totalSec = [math]::Round($totalSw.Elapsed.TotalSeconds % 60, 1)
# -- Summary --
if ($BuildOk -and (Test-Path -LiteralPath $LlamaServerBin)) {
step "llama.cpp" "built"
$QuantizeBin = Join-Path $BuildDir "bin\Release\llama-quantize.exe"
if (Test-Path -LiteralPath $QuantizeBin) {
step "llama-quantize" "built"
}
step "build time" "${totalMin}m ${totalSec}s" "DarkGray"
} else {
$altBin = Join-Path $BuildDir "bin\llama-server.exe"
if ($BuildOk -and (Test-Path -LiteralPath $altBin)) {
step "llama.cpp" "built"
step "build time" "${totalMin}m ${totalSec}s" "DarkGray"
} else {
step "llama.cpp" "build failed at: $FailedStep (${totalMin}m ${totalSec}s); continuing" "Yellow"
substep "To retry: delete $LlamaCppDir and re-run setup." "Yellow"
$script:LlamaCppDegraded = $true
}
}
}
# ─────────────────────────────────────────────
# 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), $Rule, (Get-StudioAnsi Reset))
if ($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 " $Rule" -ForegroundColor DarkGray
if ($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
}