Studio: fix training output dir escaping outputs root for models on another drive (#6293)

* Studio: derive training output dir from model basename for local-drive models

A LoRA/QLoRA run started from a model loaded by absolute path (common when
models live on a non-system drive, e.g. G:\modelsAI\...\gemma-4-12B-it) seeded
the default output dir with that full path. resolve_output_dir then raised
"path escapes root ... is not under the studio outputs folder", so training
could not start from a model stored off the system drive.

Add default_run_dir_name(): Hugging Face repo ids keep their namespace
(org/model becomes org_model), while local paths collapse to their final
component so an absolute source path can no longer leak into the output dir.
Use it at the three worker derivation sites and add a regression test.

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

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

* Studio: cap run dir name length and drop redundant resolve

Apply PR review feedback: length-cap the auto-generated output dir component so an unusually long model name stays under the filesystem name limit, and drop the redundant double resolve_output_dir at the embedding site so all three derivation sites match.

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

studio/backend/core/training/worker.py

@@ -1687,12 +1687,12 @@ def _run_mlx_training(event_queue, stop_queue, config):
         warmup_steps = 5
 
     # ── 5. Build output dir ──
+    # Resolve to ~/.unsloth/studio/outputs/ so the export page finds it
+    from utils.paths import resolve_output_dir, ensure_dir, default_run_dir_name
+
     output_dir = config.get("output_dir", "")
     if not output_dir:
-        output_dir = f"{model_name.replace('/', '_')}_{int(time.time())}"
-    # Resolve to ~/.unsloth/studio/outputs/ so the export page finds it
-    from utils.paths import resolve_output_dir, ensure_dir
-
+        output_dir = f"{default_run_dir_name(model_name)}_{int(time.time())}"
     output_dir = str(resolve_output_dir(output_dir))
     ensure_dir(Path(output_dir))
 
@@ -2450,6 +2450,7 @@ def run_training_process(*, event_queue: Any, stop_queue: Any, config: dict) ->
             resolve_output_dir,
             resolve_tensorboard_dir,
             datasets_root,
+            default_run_dir_name,
         )
 
         import transformers
@@ -2773,7 +2774,7 @@ def run_training_process(*, event_queue: Any, stop_queue: Any, config: dict) ->
             resume_from_checkpoint
         )
         if not output_dir:
-            output_dir = f"{model_name.replace('/', '_')}_{int(time.time())}"
+            output_dir = f"{default_run_dir_name(model_name)}_{int(time.time())}"
         output_dir = str(resolve_output_dir(output_dir))
         ensure_dir(Path(output_dir))
 
@@ -2924,7 +2925,7 @@ def _run_embedding_training(event_queue: Any, stop_queue: Any, config: dict) ->
         from datasets import Dataset
         from utils.datasets.cache_safe import load_dataset_cache_safe as load_dataset
         from transformers import TrainerCallback
-        from utils.paths import datasets_root, resolve_output_dir
+        from utils.paths import datasets_root, resolve_output_dir, default_run_dir_name
     except ImportError as e:
         event_queue.put(
             {
@@ -3182,7 +3183,7 @@ def _run_embedding_training(event_queue: Any, stop_queue: Any, config: dict) ->
         resume_from_checkpoint
     )
     if not output_dir:
-        output_dir = str(resolve_output_dir(f"{model_name.replace('/', '_')}_{int(time.time())}"))
+        output_dir = f"{default_run_dir_name(model_name)}_{int(time.time())}"
     output_dir = str(resolve_output_dir(output_dir))
 
     num_epochs = config.get("num_epochs", 2)

studio/backend/tests/test_default_output_dir_name.py

@@ -0,0 +1,65 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+"""Auto-generated training output dir names stay inside outputs_root.
+
+Regression for local-model training: a model loaded by absolute path (e.g.
+``G:\\modelsAI\\...\\gemma-4-12B-it`` on a non-system drive) used to seed the
+default run dir with that full path, so ``resolve_output_dir`` raised
+``path escapes root`` because the result was not under ``<studio>/outputs``.
+"""
+
+import importlib.util
+from pathlib import Path
+
+import pytest
+
+
+_BACKEND_DIR = Path(__file__).resolve().parent.parent
+
+
+def _load_storage_roots():
+    path = _BACKEND_DIR / "utils/paths/storage_roots.py"
+    spec = importlib.util.spec_from_file_location("storage_roots_under_test", path)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_repo_id_keeps_namespace():
+    sr = _load_storage_roots()
+    assert sr.default_run_dir_name("unsloth/gemma-3-4b") == "unsloth_gemma-3-4b"
+    assert sr.default_run_dir_name("gemma-3-4b") == "gemma-3-4b"
+
+
+def test_local_paths_collapse_to_basename():
+    sr = _load_storage_roots()
+    assert sr.default_run_dir_name(r"G:\modelsAI\gguf\test\gemma-4-12B-it") == "gemma-4-12B-it"
+    assert sr.default_run_dir_name("/data/models/gemma-3-4b") == "gemma-3-4b"
+    assert sr.default_run_dir_name("~/models/gemma-3-4b") == "gemma-3-4b"
+    assert sr.default_run_dir_name("C:/Users/me/models/gemma-3-4b") == "gemma-3-4b"
+
+
+def test_empty_falls_back_to_model():
+    sr = _load_storage_roots()
+    assert sr.default_run_dir_name("") == "model"
+    assert sr.default_run_dir_name("   ") == "model"
+
+
+def test_very_long_name_is_capped():
+    sr = _load_storage_roots()
+    name = sr.default_run_dir_name("a" * 500)
+    assert 0 < len(name) <= 200
+
+
+def test_derived_name_resolves_under_outputs_root(tmp_path, monkeypatch):
+    sr = _load_storage_roots()
+    outputs = tmp_path / "outputs"
+    outputs.mkdir()
+    monkeypatch.setattr(sr, "outputs_root", lambda: outputs)
+
+    name = sr.default_run_dir_name(r"G:\modelsAI\gguf\test\gemma-4-12B-it")
+    resolved = sr.resolve_output_dir(f"{name}_1781327234")
+    assert resolved == outputs / "gemma-4-12B-it_1781327234"
+    # No escape: the absolute G: source no longer leaks into the output path.
+    assert "modelsAI" not in str(resolved)

studio/backend/utils/paths/__init__.py

@@ -41,6 +41,7 @@ from .storage_roots import (
     ensure_dir,
     ensure_studio_directories,
     resolve_under_root,
+    default_run_dir_name,
     resolve_output_dir,
     resolve_export_dir,
     resolve_export_write_dir,
@@ -88,6 +89,7 @@ __all__ = [
     "ensure_dir",
     "ensure_studio_directories",
     "resolve_under_root",
+    "default_run_dir_name",
     "resolve_output_dir",
     "resolve_export_dir",
     "resolve_export_write_dir",

studio/backend/utils/paths/storage_roots.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 
 import json
 import os
+import re
 import sys
 from pathlib import Path, PurePosixPath, PureWindowsPath
 import tempfile
@@ -384,6 +385,23 @@ def resolve_under_root(
     return candidate
 
 
+def default_run_dir_name(model_name: str) -> str:
+    # Folder-safe run name for an auto-created output dir. Repo ids keep their
+    # namespace (org/model -> org_model); local paths (incl. G:\dir\model)
+    # collapse to their final component so an absolute source can't escape
+    # outputs_root. Length-capped to stay under the filesystem name limit.
+    raw = str(model_name or "").strip()
+    is_path = (
+        "\\" in raw
+        or raw.startswith(("/", "~", "."))
+        or os.path.isabs(raw)
+        or (len(raw) >= 2 and raw[1] == ":")
+    )
+    base = PureWindowsPath(raw).name if is_path else raw.replace("/", "_")
+    base = re.sub(r"[^A-Za-z0-9._-]+", "_", base)[:200].strip("._-")
+    return base or "model"
+
+
 def resolve_output_dir(path_value: str | None = None) -> Path:
     return resolve_under_root(
         path_value,