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.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""unsloth-cli.py reads --use_gradient_checkpointing as a string, but get_peft_model wants True, False or "unsloth"."""
import importlib.util
import sys
import types
from pathlib import Path
import pytest
_REPO_ROOT = Path(__file__).resolve().parents[2]
class _Stop(Exception):
pass
def _load_legacy_cli():
spec = importlib.util.spec_from_file_location(
"unsloth_legacy_cli_gradient_checkpointing", _REPO_ROOT / "unsloth-cli.py"
)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
def _module(name, **attrs):
module = types.ModuleType(name)
module.__dict__.update(attrs)
return module
def _setup(monkeypatch, spelling):
seen = {}
class FastLanguageModel:
@staticmethod
def from_pretrained(**kwargs):
seen["loaded"] = True
return object(), types.SimpleNamespace(eos_token = "</s>")
@staticmethod
def get_peft_model(model, **kwargs):
seen.update(kwargs)
raise _Stop
for name, module in {
"unsloth": _module(
"unsloth",
FastLanguageModel = FastLanguageModel,
is_bfloat16_supported = lambda: False,
RawTextDataLoader = object,
),
"unsloth.models": _module("unsloth.models"),
"unsloth.models.loader_utils": _module(
"unsloth.models.loader_utils", prepare_device_map = lambda: (None, False)
),
"datasets": _module("datasets", load_dataset = None),
"transformers": _module("transformers"),
"transformers.utils": _module("transformers.utils", strtobool = None),
"trl": _module("trl", SFTTrainer = None, SFTConfig = None),
}.items():
monkeypatch.setitem(sys.modules, name, module)
args = types.SimpleNamespace(
model_name = "unsloth/llama-3-8b",
max_seq_length = 2048,
dtype = None,
load_in_4bit = True,
use_dora = False,
r = 16,
lora_alpha = 16,
lora_dropout = 0,
bias = "none",
use_gradient_checkpointing = spelling,
random_state = 3407,
use_rslora = False,
loftq_config = None,
)
return _load_legacy_cli(), args, seen
@pytest.mark.parametrize(
"spelling, expected",
[
("unsloth", "unsloth"),
("Unsloth", "unsloth"),
("", "unsloth"),
("mlx", "unsloth"),
("true", True),
("True", True),
("1", True),
("yes", True),
("false", False),
("False", False),
("0", False),
("no", False),
("none", False),
("off", False),
],
)
def test_gradient_checkpointing_reaches_get_peft_model_as_a_supported_value(
monkeypatch, spelling, expected
):
legacy_cli, args, seen = _setup(monkeypatch, spelling)
with pytest.raises(_Stop):
legacy_cli.run(args)
got = seen["use_gradient_checkpointing"]
assert type(got) is type(expected) and got == expected
def test_unknown_gradient_checkpointing_fails_before_the_model_loads(monkeypatch):
legacy_cli, args, seen = _setup(monkeypatch, "flase")
with pytest.raises(ValueError, match = "unsloth, true or false"):
legacy_cli.run(args)
assert "loaded" not in seen