Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

5.9 KB
Raw Blame
# SPDX-License-Identifier: AGPL-3.0-only

"""Canaries that MLX support never hijacks CUDA, AMD or Intel users via _IS_MLX or detect_hardware."""

import ast
import importlib
import sys
import types
from pathlib import Path


REPO_ROOT = Path(__file__).resolve().parents[2]
UNSLOTH_INIT = REPO_ROOT / "unsloth" / "__init__.py"


def test_is_mlx_gate_uses_three_required_predicates():
    """_IS_MLX must AND Darwin+arm64+importable-mlx; dropping any breaks dispatch."""
    tree = ast.parse(UNSLOTH_INIT.read_text(encoding = "utf-8"))

    target = None
    for node in ast.walk(tree):
        if (
            isinstance(node, ast.Assign)
            and len(node.targets) == 1
            and isinstance(node.targets[0], ast.Name)
            and node.targets[0].id == "_IS_MLX"
        ):
            target = node.value
            break
    assert target is not None, "_IS_MLX assignment not found in unsloth/__init__.py"
    assert isinstance(target, ast.Call), "_IS_MLX must call the shared MLX helper"
    expr_src = ast.unparse(target)
    assert expr_src == "_is_mlx_available()", "_IS_MLX must delegate to the shared MLX runtime gate"

    helper = None
    for node in ast.walk(tree):
        if isinstance(node, ast.FunctionDef) and node.name == "_is_mlx_available":
            helper = node
            break
    assert helper is not None, "_is_mlx_available helper not found"

    helper_src = ast.unparse(helper)
    assert (
        "platform.system()" in helper_src
        and "'Darwin'" in helper_src
        and "platform.machine()" in helper_src
        and "'arm64'" in helper_src
        and "find_spec" in helper_src
        and "'mlx'" in helper_src
        and "from unsloth_zoo.mlx import is_mlx_available" in helper_src
    ), "_IS_MLX helper must precheck local MLX predicates before importing zoo"
    assert (
        "from unsloth_zoo.mlx import is_mlx_available" in helper_src
        and "return is_mlx_available()" in helper_src
    ), "_IS_MLX helper must delegate final detection to the shared zoo MLX runtime gate"
    assert helper_src.index("UNSLOTH_FORCE_GPU_PATH") < helper_src.index(
        "from unsloth_zoo.mlx import is_mlx_available"
    ), "_IS_MLX helper must run the local MLX precheck before importing zoo"


def _evaluate_is_mlx_precheck(platform_module, importlib_util, os_module):
    """Re-evaluate the local _is_mlx_available precheck with injected deps."""
    return (
        os_module.environ.get("UNSLOTH_FORCE_GPU_PATH", "0") != "1"
        and platform_module.system() == "Darwin"
        and platform_module.machine() == "arm64"
        and importlib_util.find_spec("mlx") is not None
    )


def test_is_mlx_gate_true_on_apple_silicon_with_mlx_present(monkeypatch):
    import platform
    import importlib.util

    fake_mlx = types.ModuleType("mlx")
    fake_mlx.__spec__ = importlib.machinery.ModuleSpec("mlx", loader = None)
    fake_mlx.__path__ = []
    monkeypatch.setitem(sys.modules, "mlx", fake_mlx)

    monkeypatch.setattr(platform, "system", lambda: "Darwin")
    monkeypatch.setattr(platform, "machine", lambda: "arm64")

    import os

    assert _evaluate_is_mlx_precheck(platform, importlib.util, os) is True


def test_is_mlx_gate_false_when_mlx_missing(monkeypatch):
    import platform
    import importlib.util

    monkeypatch.delitem(sys.modules, "mlx", raising = False)
    monkeypatch.setattr(platform, "system", lambda: "Darwin")
    monkeypatch.setattr(platform, "machine", lambda: "arm64")

    real_find_spec = importlib.util.find_spec

    def _no_mlx(name, *args, **kwargs):
        if name == "mlx":
            return None
        return real_find_spec(name, *args, **kwargs)

    monkeypatch.setattr(importlib.util, "find_spec", _no_mlx)

    import os

    assert _evaluate_is_mlx_precheck(platform, importlib.util, os) is False


def test_is_mlx_gate_false_on_non_apple_silicon():
    """On the real Linux+CUDA / AMD / Intel test host, the gate stays False."""
    import platform
    import importlib.util

    if platform.system() == "Darwin" and platform.machine() == "arm64":
        import pytest
        pytest.skip("Test host is Apple Silicon; CUDA-side canary doesn't apply.")

    import os

    assert _evaluate_is_mlx_precheck(platform, importlib.util, os) is False


def _import_studio_hardware():
    """Lazy import of the Unsloth hardware module (studio/backend on sys.path)."""
    studio_backend = REPO_ROOT / "studio" / "backend"
    if str(studio_backend) not in sys.path:
        sys.path.insert(0, str(studio_backend))
    from utils.hardware import hardware as hw  # type: ignore

    return hw


def test_detect_hardware_picks_mlx_when_only_apple_silicon_available(monkeypatch):
    hw = _import_studio_hardware()

    import torch

    monkeypatch.setattr(torch.cuda, "is_available", lambda: False)
    if hasattr(torch, "xpu"):
        monkeypatch.setattr(torch.xpu, "is_available", lambda: False)

    import platform

    monkeypatch.setattr(platform, "system", lambda: "Darwin")
    monkeypatch.setattr(platform, "machine", lambda: "arm64")

    fake_mlx = types.ModuleType("mlx")
    fake_mlx_core = types.ModuleType("mlx.core")
    fake_mlx.core = fake_mlx_core
    monkeypatch.setitem(sys.modules, "mlx", fake_mlx)
    monkeypatch.setitem(sys.modules, "mlx.core", fake_mlx_core)

    # detect_hardware gates on the full MLX stack, so faking mlx.core alone is not enough.
    monkeypatch.setattr(hw, "_has_usable_mlx_stack", lambda: True)

    detected = hw.detect_hardware()
    assert detected == hw.DeviceType.MLX, f"expected MLX, got {detected!r}"


def test_detect_hardware_picks_cuda_on_real_host():
    """Canary: a real CUDA host must dispatch to CUDA even if mlx is importable."""
    import torch

    if not torch.cuda.is_available():
        import pytest
        pytest.skip("No CUDA available on this host; canary not applicable.")

    hw = _import_studio_hardware()
    detected = hw.detect_hardware()
    assert detected == hw.DeviceType.CUDA, f"CUDA host must dispatch to CUDA, got {detected!r}"