* Studio: defer llama.cpp update probes and self-heal MLX on macOS Two macOS startup problems shared one root area in the FastAPI lifespan: - The llama.cpp capability + freshness probes ran inline before the server yielded, so a cold/slow/flaky network on the GitHub freshness check blocked 'Application startup complete' (~34s on CI, longer in the field). Move both probes to a daemon thread; app.state stays None until ready (status routes already re-probe at request time). Opt out with UNSLOTH_DISABLE_UPDATE_CHECK=1. - Train and Export were greyed out because mlx/mlx-lm/mlx-vlm arrive only transitively and a resolver backtrack silently drops them, so CHAT_ONLY stayed true. Add utils/mlx_repair.py: when Apple Silicon is detected without MLX, reinstall mlx/mlx-lm/mlx-vlm by name on a daemon thread and re-run hardware detection (opt out UNSLOTH_DISABLE_MLX_AUTOREPAIR=1). Surface a chat_only_reason in /api/health plus a sidebar tooltip so a greyed Train/Export explains itself instead of failing silently. * Studio: guard model defaults against a None model name load_model_defaults(None) called model_name.lower() with no guard, raising 'Error loading model defaults for None' before any model is selected. Return an empty dict for a falsy/non-str name. * Studio: drop obsolete upstream macOS + Windows Blackwell prebuilt pins Both pins worked around gaps in ggml-org upstream prebuilts, but Studio now routes every GPU host and all of macOS to the unslothai/llama.cpp fork (published_repo_for_host), which ships the needed bundles, so both pins are dead code on the default install path: - macOS b9415: macOS always routes to the fork (its own macOS bundles), and host_supports_macos_minos() is the backstop. The pin only fired under an explicit --published-repo ggml-org override. - Windows Blackwell b9360: Windows-NVIDIA routes to the fork, whose windows-x64-cuda13 bundle covers Blackwell (manifest max_sm 120, toolkit 13.3), so the pin's self-disable check makes it dormant on every default install; it could only activate under the same upstream override on a 13.0-13.2 driver. Remove the pin constants, functions, and call sites. Keep the Blackwell capability detection (_drop_blackwell_incapable_windows_cuda, _host_is_blackwell, _windows_cuda_attempt_covers_blackwell) that still drops a non-sm_120 cuda-12.4 build on a Blackwell host. After this, an explicit --published-repo ggml-org override on a Blackwell 13.0-13.2 host loses its GPU fallback and lands on CPU; the default fork path is unaffected. Update the install selection-logic and macOS-compat unit tests for the new no-pin behavior. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Studio: walk back deeper on the macOS upstream prebuilt path After removing the b9415 macOS pin, the explicit --published-repo ggml-org upstream path still used the default 2-release fallback, so a pre-macOS-26 host behind a run of macOS-26-only builds would exhaust two too-new plans (minos is only checked post-download) and drop to a source build before reaching a loadable older release. Walk back as deep as the fork macOS path (DEFAULT_MAX_MACOS_RELEASE_FALLBACKS), turning the removed static pin into dynamic discovery. Addresses review feedback on the macOS upstream fallback. * Studio: pin transformers during MLX self-heal so it cannot break Studio mlx-lm/mlx-vlm declare transformers>=5, but the single-env install pins transformers==4.57.6. The self-heal used --upgrade with no constraint, so it could upgrade transformers in the live venv and break the rest of Studio just to make import mlx.core pass. Pin transformers to the installed version via a constraint file: the resolver either finds an mlx build compatible with it or fails (we stay chat-only), never upgrading transformers underneath Studio. Addresses review feedback on the MLX repair install. * Studio: harden MLX self-heal against an unsupported mlx-vlm Pinning transformers alone made uv backtrack mlx-vlm to 0.3.9 (below unsloth-zoo's mlx-vlm>=0.4.4), which imports but breaks VLM Train/Export -- so the self-heal could clear chat-only onto a broken stack. Mirror the main installer: set UV_OVERRIDE=overrides-darwin-arm64.txt so a current mlx-vlm coexists with the transformers pin, require the same minimum versions unsloth-zoo declares, and gate/validate on a full mlx_stack_available() check (not a bare import) so an old or partial stack stays chat-only. Addresses PR review. * Studio: filter Blackwell-incapable CUDA in resolve_upstream_asset_choice resolve_upstream_asset_choice returned the first windows-cuda choice unfiltered, so a Blackwell host could be handed an sm_120-incapable cuda-12.4 build while the sibling planners drop it. Apply _drop_blackwell_incapable_windows_cuda here too and fall through to the CPU bundle on a Blackwell host with no capable GPU asset. Addresses PR review. * Studio: re-poll health so MLX self-heal reaches an open UI The sidebar cached the initial /api/health, so a successful background MLX self-heal (chat_only flips false) did not re-enable Train/Export until a manual reload. While chat-only for the recoverable mlx_unavailable reason, re-poll /api/health and stop once Train/Export become available. Addresses PR review. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Studio: make the disabled Train/Export tooltip reachable The greyed Train/Export items pass a tooltip explaining why (e.g. MLX missing), but a disabled <button> fires no pointer events and SidebarMenuButton only showed tooltips while collapsed, so the explanation never appeared. Wrap a disabled button in a focusable span and show its tooltip while expanded too; enabled items keep the collapsed-only behavior. Addresses PR review. * Studio: gate Train/Export on the full MLX stack, not bare mlx.core detect_hardware enabled MLX training whenever `import mlx.core` worked, but the MLX self-heal (utils/mlx_repair) treats a stack without mlx-lm/mlx-vlm at the versions unsloth-zoo requires as inadequate. That asymmetry let the UI enable Train/Export on exactly the partial/backtracked stack the self-heal is trying to repair (greyed-in-but-broken VLM export). Gate on the same mlx_stack_available() criterion so a partial stack stays chat-only (reason mlx_unavailable) and the background repair restores it. Addresses PR review. * Fix MLX repair and health auth for PR #6494 * Fix macOS upstream prebuilt fallback for PR #6494 * Fix MLX stack validation for PR #6494 * Fix MLX self-heal validation for PR #6494 * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Review fixes: isolate hardware-state test, robust transformers pin - test_chat_only_reason.py: detect_hardware() assigns module globals directly, which monkeypatch does not revert; the autouse fixture now saves and restores DEVICE/CHAT_ONLY/CHAT_ONLY_REASON/IS_ROCM so a chat-only verdict here cannot leak into other backend tests (e.g. test_utils.py) on a GPU host. - mlx_repair.py: read the transformers version from importlib.metadata instead of importing transformers, so the install pin is not silently dropped when transformers has valid metadata but fails to import. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix CI: model full MLX stack in dispatch tests, keep selection test offline dispatch (macOS) job: - detect_hardware now gates MLX on the full stack (mlx_stack_available imports mlx_lm/mlx_vlm and checks dist versions), so faking only mlx.core makes the apple_silicon_mlx profile resolve to CPU. The dispatch tests assert the routing decision when the stack IS usable, so model a complete stack: test_hardware_dispatch_matrix patches utils.mlx_repair.mlx_stack_available and test_is_mlx_dispatch_gate patches hardware._has_usable_mlx_stack. The stack predicate's own internals stay covered by test_mlx_repair.py. Repo tests (CPU) job: - test_no_cuda_attempt_on_published_path_for_13_1 fell through to a live github_release_assets() upstream fetch after the Blackwell filter dropped every published attempt, which the offline security scanner blocks. Stub that fetch so the walk-back deterministically finds no usable CUDA build and raises PrebuiltFallback without network. * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Harden MLX self-heal: prepare transformers constraint inside the try attempt_mlx_repair runs on a daemon thread, but _transformers_constraint_args was called before the try. A failure there (e.g. tempfile.mkstemp on a full disk or a bad TMPDIR) would propagate unhandled and silently kill the self-heal thread. Move the call inside the try and initialize constraint_path so any such failure is caught and leaves Studio chat-only instead of crashing the thread. --------- Co-authored-by: Daniel Han <michaelhan2050@gmail.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: wasimysaid <wasimysdev@gmail.com>
279 lines
10 KiB
Python
279 lines
10 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-only
|
|
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
|
|
|
|
"""Best-effort MLX self-heal for Apple Silicon.
|
|
|
|
On macOS, Studio enables Train/Export only when the MLX training/export stack is
|
|
usable (see utils.hardware.hardware.detect_hardware -> CHAT_ONLY). MLX is pulled
|
|
only transitively via unsloth-zoo, and a resolver backtrack (mlx-vlm ->
|
|
transformers>=5 vs the single-env transformers pin) can silently drop it, leaving
|
|
Train/Export greyed out after a reinstall/update. This reinstalls mlx by name on
|
|
a background thread, then re-detects so the gate re-opens without a manual
|
|
`unsloth studio update`.
|
|
|
|
The install mirrors the main Apple Silicon installer (install_python_stack.py):
|
|
it points UV_OVERRIDE at overrides-darwin-arm64.txt so the resolver keeps the
|
|
Studio transformers pin AND installs a current mlx-vlm, and it requires the same
|
|
minimum versions unsloth-zoo declares so a backtracked old mlx-vlm (which still
|
|
imports but breaks VLM Train/Export) is never accepted as healthy.
|
|
|
|
Mirrors the runtime backend self-heal already used for tilelang
|
|
(core.training.worker._ensure_tilelang_backend_unconditional): default-on,
|
|
best-effort, opt out with UNSLOTH_DISABLE_MLX_AUTOREPAIR=1.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import importlib
|
|
import os
|
|
import platform
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
import tempfile
|
|
import threading
|
|
from pathlib import Path
|
|
|
|
import structlog
|
|
|
|
logger = structlog.get_logger(__name__)
|
|
|
|
DISABLE_ENV_VAR = "UNSLOTH_DISABLE_MLX_AUTOREPAIR"
|
|
# Minimum versions unsloth-zoo requires on Apple Silicon (its pyproject darwin
|
|
# deps). mlx-vlm especially must be >=0.4.4: an older one still imports but
|
|
# breaks VLM Train/Export, so installing it would wrongly clear chat-only.
|
|
_MLX_MIN_VERSIONS = {"mlx": "0.22.0", "mlx-lm": "0.22.0", "mlx-vlm": "0.4.4"}
|
|
_MLX_PACKAGE_NAMES = tuple(_MLX_MIN_VERSIONS)
|
|
_MLX_RUNTIME_IMPORTS = ("mlx.core", "mlx_lm", "mlx_lm.sample_utils", "mlx_vlm")
|
|
MLX_PACKAGES = tuple(f"{name}>={version}" for name, version in _MLX_MIN_VERSIONS.items())
|
|
_MLX_REINSTALL_ARGS = tuple(
|
|
arg for name in _MLX_PACKAGE_NAMES for arg in ("--reinstall-package", name)
|
|
)
|
|
_REPAIR_TIMEOUT_S = 900
|
|
|
|
# Attempt at most once per process; success is sticky (mlx then imports and the
|
|
# guard short-circuits on the next boot).
|
|
_attempted = False
|
|
_attempted_lock = threading.Lock()
|
|
|
|
|
|
def is_apple_silicon() -> bool:
|
|
return platform.system() == "Darwin" and platform.machine() == "arm64"
|
|
|
|
|
|
def mlx_available() -> bool:
|
|
try:
|
|
import mlx.core # noqa: F401
|
|
return True
|
|
except Exception:
|
|
return False
|
|
|
|
|
|
def _mlx_runtime_imports_available() -> bool:
|
|
for module in _MLX_RUNTIME_IMPORTS:
|
|
try:
|
|
importlib.import_module(module)
|
|
except Exception:
|
|
return False
|
|
return True
|
|
|
|
|
|
def _mlx_versions_satisfy_minimums() -> bool:
|
|
try:
|
|
from importlib.metadata import PackageNotFoundError
|
|
from importlib.metadata import version as _dist_version
|
|
|
|
from packaging.version import Version
|
|
except Exception:
|
|
return False
|
|
for name, minimum in _MLX_MIN_VERSIONS.items():
|
|
try:
|
|
if Version(_dist_version(name)) < Version(minimum):
|
|
return False
|
|
except PackageNotFoundError:
|
|
return False
|
|
except Exception:
|
|
return False
|
|
return True
|
|
|
|
|
|
def mlx_stack_available() -> bool:
|
|
"""`import mlx.core` works AND mlx/mlx-lm/mlx-vlm meet unsloth-zoo's minimums.
|
|
|
|
Check distribution versions before imports so a too-old but importable MLX
|
|
module is not loaded into this process before repair can replace it."""
|
|
if not _mlx_versions_satisfy_minimums():
|
|
return False
|
|
return _mlx_runtime_imports_available()
|
|
|
|
|
|
def _uv_executable() -> str | None:
|
|
"""Find uv even when macOS GUI launchers start with a minimal PATH."""
|
|
found = shutil.which("uv")
|
|
if found:
|
|
return found
|
|
for candidate in (
|
|
Path.home() / ".local" / "bin" / "uv",
|
|
Path.home() / ".cargo" / "bin" / "uv",
|
|
Path("/opt/homebrew/bin/uv"),
|
|
Path("/usr/local/bin/uv"),
|
|
):
|
|
try:
|
|
if candidate.is_file() and os.access(candidate, os.X_OK):
|
|
return str(candidate)
|
|
except OSError:
|
|
continue
|
|
return None
|
|
|
|
|
|
def _uv_install_cmd(*args: str) -> list[str] | None:
|
|
uv = _uv_executable()
|
|
if not uv:
|
|
return None
|
|
return [uv, "pip", "install", "--python", sys.executable, *args]
|
|
|
|
|
|
def _mlx_install_env() -> dict[str, str]:
|
|
"""Environment for the mlx install. Mirror the main installer
|
|
(install_python_stack.py) by pointing UV_OVERRIDE at overrides-darwin-arm64.txt,
|
|
which relaxes mlx-vlm/mlx-lm's transformers>=5 requirement to >=4.57.6. Without
|
|
it, uv keeps the Studio transformers pin only by silently backtracking mlx-vlm
|
|
to an old, unsupported version (uv honours UV_OVERRIDE; plain pip ignores it,
|
|
so the transformers constraint below is the pip-path safety net)."""
|
|
env = dict(os.environ)
|
|
override = (
|
|
Path(__file__).resolve().parents[1]
|
|
/ "requirements"
|
|
/ "single-env"
|
|
/ "overrides-darwin-arm64.txt"
|
|
)
|
|
if override.is_file():
|
|
env.setdefault("UV_OVERRIDE", str(override))
|
|
return env
|
|
|
|
|
|
def _transformers_constraint_args() -> tuple[list[str], str | None]:
|
|
"""Pin transformers to the running version for the mlx install.
|
|
|
|
The install must never upgrade transformers underneath a running Studio
|
|
(the single-env install pins transformers==4.57.6). With UV_OVERRIDE set this
|
|
is belt-and-suspenders; on the plain-pip path (no UV_OVERRIDE support) it is
|
|
the actual guard -- the resolver either finds an mlx build compatible with the
|
|
pin or fails, leaving us chat-only rather than breaking Studio. Returns
|
|
(pip args, temp file path to clean up).
|
|
|
|
Read the version from installed metadata rather than `import transformers`:
|
|
transformers can have valid metadata yet fail to import (e.g. an incompatible
|
|
huggingface_hub), and in that case we still want to pin it so the mlx install
|
|
cannot quietly upgrade it out from under Studio."""
|
|
from importlib.metadata import PackageNotFoundError, version as _dist_version
|
|
|
|
try:
|
|
transformers_version = _dist_version("transformers")
|
|
except PackageNotFoundError:
|
|
return [], None
|
|
except Exception:
|
|
return [], None
|
|
fd, path = tempfile.mkstemp(prefix = "mlx_repair_", suffix = ".txt")
|
|
with os.fdopen(fd, "w") as fh:
|
|
fh.write(f"transformers=={transformers_version}\n")
|
|
return ["--constraint", path], path
|
|
|
|
|
|
def attempt_mlx_repair(*, timeout: int = _REPAIR_TIMEOUT_S) -> bool:
|
|
"""Install a usable mlx/mlx-lm/mlx-vlm stack by name into the running venv.
|
|
Best-effort; returns True iff the resulting stack meets unsloth-zoo's minimums
|
|
(so a backtracked old mlx-vlm is rejected, not accepted). transformers is held
|
|
at its pinned version so the install can never upgrade it underneath Studio."""
|
|
# Prepare the constraint inside the try: this runs on a daemon thread, so an
|
|
# exception here (e.g. tempfile.mkstemp failing on a full disk or bad TMPDIR)
|
|
# must leave Studio chat-only, not crash the background self-heal thread.
|
|
constraint_path = None
|
|
try:
|
|
constraint_args, constraint_path = _transformers_constraint_args()
|
|
cmd = _uv_install_cmd("--upgrade", *_MLX_REINSTALL_ARGS, *constraint_args, *MLX_PACKAGES)
|
|
if cmd is None:
|
|
logger.warning(
|
|
"MLX self-heal requires uv so Studio can apply dependency overrides; "
|
|
"staying chat-only. Run `unsloth studio update` to restore uv."
|
|
)
|
|
return False
|
|
logger.info("MLX self-heal: installing %s", ", ".join(MLX_PACKAGES))
|
|
result = subprocess.run(
|
|
cmd,
|
|
env = _mlx_install_env(),
|
|
stdout = subprocess.PIPE,
|
|
stderr = subprocess.STDOUT,
|
|
text = True,
|
|
timeout = timeout,
|
|
)
|
|
except subprocess.TimeoutExpired:
|
|
logger.warning("MLX self-heal timed out after %ss; staying chat-only", timeout)
|
|
return False
|
|
except Exception as exc: # pragma: no cover - environment dependent
|
|
logger.warning("MLX self-heal could not start: %s", exc)
|
|
return False
|
|
finally:
|
|
if constraint_path and os.path.exists(constraint_path):
|
|
try:
|
|
os.remove(constraint_path)
|
|
except OSError:
|
|
pass
|
|
if result.returncode != 0:
|
|
tail = (result.stdout or "")[-2000:]
|
|
logger.warning("MLX self-heal failed (staying chat-only):\n%s", tail)
|
|
return False
|
|
importlib.invalidate_caches()
|
|
if not mlx_stack_available():
|
|
logger.warning(
|
|
"MLX self-heal produced an incomplete or too-old MLX stack "
|
|
"(need %s); staying chat-only.",
|
|
", ".join(f"{name}>={ver}" for name, ver in _MLX_MIN_VERSIONS.items()),
|
|
)
|
|
return False
|
|
return True
|
|
|
|
|
|
def _run_repair_and_redetect() -> None:
|
|
if not attempt_mlx_repair():
|
|
return
|
|
try:
|
|
from utils.hardware import hardware as hw
|
|
hw.detect_hardware() # flips CHAT_ONLY / DEVICE now that mlx imports
|
|
logger.info(
|
|
"MLX self-heal succeeded; Train/Export enabled (reload the page). chat_only=%s",
|
|
hw.CHAT_ONLY,
|
|
)
|
|
except Exception as exc: # pragma: no cover - defensive
|
|
logger.warning("MLX installed but hardware re-detection failed: %s", exc)
|
|
|
|
|
|
def start_mlx_autorepair_if_needed() -> bool:
|
|
"""If this is an Apple Silicon host whose MLX stack is missing or too old,
|
|
reinstall it on a daemon thread (off the startup critical path) and re-detect
|
|
on success. Returns True iff a repair thread was started. No-op (returns False)
|
|
off Apple Silicon, when the stack is already adequate, when already attempted
|
|
this process, or when disabled via UNSLOTH_DISABLE_MLX_AUTOREPAIR=1."""
|
|
global _attempted
|
|
if os.environ.get(DISABLE_ENV_VAR) == "1":
|
|
return False
|
|
if not is_apple_silicon():
|
|
return False
|
|
if mlx_stack_available():
|
|
return False
|
|
with _attempted_lock:
|
|
if _attempted:
|
|
return False
|
|
_attempted = True
|
|
logger.warning(
|
|
"Apple Silicon without a usable MLX stack; attempting a one-time background "
|
|
"reinstall of mlx/mlx-lm/mlx-vlm to re-enable Train/Export. "
|
|
"Set %s=1 to disable.",
|
|
DISABLE_ENV_VAR,
|
|
)
|
|
threading.Thread(
|
|
target = _run_repair_and_redetect,
|
|
daemon = True,
|
|
name = "mlx-autorepair",
|
|
).start()
|
|
return True
|