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
import importlib
import sys
import types
from unittest.mock import MagicMock
import pytest
_STUBBED: list[str] = []
def _stub_if_missing(name, attrs):
if name in sys.modules:
return
try:
importlib.import_module(name)
return
except Exception:
pass
_STUBBED.append(name)
mod = types.ModuleType(name)
mod.__spec__ = None
for attr in attrs:
setattr(mod, attr, MagicMock())
sys.modules[name] = mod
parent, _, child = name.rpartition(".")
if parent and parent in sys.modules:
setattr(sys.modules[parent], child, mod)
_stub_if_missing("unsloth", ("FastLanguageModel", "FastVisionModel", "is_bfloat16_supported"))
_stub_if_missing("unsloth.chat_templates", ("get_chat_template",))
import core.inference.inference as inf # noqa: E402
# Leaving these installed changes _shared_policy in later tests.
for _name in reversed(_STUBBED):
sys.modules.pop(_name, None)
class _Lfm3AudioModel:
device = "cpu"
class _Lfm3AudioForConditionalGeneration(_Lfm3AudioModel):
def generate(self, **kwargs):
return None
class _Processor:
feature_extractor = object()
tokenizer = object()
def _config(**overrides):
fields = dict(
identifier = "LiquidAI/LFM3-Audio-2B",
path = "LiquidAI/LFM3-Audio-2B",
is_vision = False,
is_lora = False,
is_audio = False,
audio_type = "audio_vlm",
has_audio_input = True,
is_local = False,
base_model = None,
gguf_variant = None,
)
fields.update(overrides)
return types.SimpleNamespace(**fields)
@pytest.fixture
def multimodal_auto_class(monkeypatch):
import transformers
auto_class = type("AutoModelForMultimodalLM", (), {})
monkeypatch.setattr(transformers, "AutoModelForMultimodalLM", auto_class, raising = False)
return auto_class
@pytest.fixture
def backend(monkeypatch, multimodal_auto_class):
import utils.models.diffusion_lm as diffusion_lm
unsloth = types.ModuleType("unsloth")
monkeypatch.setitem(sys.modules, "unsloth", unsloth)
calls = {}
def fast_model(**kwargs):
calls["fast_model"] = kwargs
return _Lfm3AudioForConditionalGeneration(), _Processor()
def fast_language_model(**kwargs):
calls["fast_language_model"] = kwargs
return _Lfm3AudioModel(), object()
noop = lambda *a, **k: None
unsloth.FastModel = types.SimpleNamespace(from_pretrained = fast_model, for_inference = noop)
monkeypatch.setattr(
inf,
"FastLanguageModel",
types.SimpleNamespace(from_pretrained = fast_language_model, for_inference = noop),
)
monkeypatch.setattr(diffusion_lm, "diffusion_lm_profile_for_model", lambda *a, **k: None)
monkeypatch.setattr(diffusion_lm, "loaded_diffusion_profile", lambda *a, **k: None)
monkeypatch.setattr(inf, "get_device_map", lambda *a, **k: "sequential")
monkeypatch.setattr(inf, "get_visible_gpu_count", lambda: 1)
monkeypatch.setattr(inf, "log_gpu_memory", noop)
monkeypatch.setattr(inf, "_exact_model_name_for_load", lambda *a, **k: None)
monkeypatch.setattr(inf, "_offload_kwargs_for_fit", lambda *a, **k: {})
monkeypatch.setattr(inf, "raise_if_offloaded", noop)
monkeypatch.setattr(inf, "runtime_context_length", lambda *a, **k: 4096)
b = inf.InferenceBackend.__new__(inf.InferenceBackend)
b.models, b.loading_models, b.active_model_name = {}, set(), None
b._resolve_chat_eos = noop
b._load_chat_template_info = noop
return b, calls
def test_audio_input_chat_loads_through_the_multimodal_class(backend, multimodal_auto_class):
b, calls = backend
assert inf.InferenceBackend.load_model(b, _config(), trust_remote_code = True) is True
assert calls["fast_model"]["auto_model"] is multimodal_auto_class
assert "fast_language_model" not in calls
loaded = b.models["LiquidAI/LFM3-Audio-2B"]
assert isinstance(loaded["processor"], _Processor)
assert loaded["has_audio_input"] is True
def test_a_plain_text_model_still_takes_the_text_load(backend):
b, calls = backend
config = _config(audio_type = None, has_audio_input = False)
with pytest.raises(Exception, match = "cannot generate text"):
inf.InferenceBackend.load_model(b, config)
assert "fast_model" not in calls and "fast_language_model" in calls
def test_audio_chat_auto_model_needs_audio_vlm_without_vision(multimodal_auto_class):
assert inf._audio_chat_auto_model(_config()) is multimodal_auto_class
assert inf._audio_chat_auto_model(_config(is_vision = True)) is None
assert inf._audio_chat_auto_model(_config(audio_type = "whisper")) is None
assert inf._audio_chat_auto_model(_config(has_audio_input = False, audio_type = None)) is None
def test_an_unavailable_multimodal_api_keeps_the_fallback(monkeypatch):
import transformers
monkeypatch.setattr(transformers, "AutoModelForMultimodalLM", None, raising = False)
assert inf._audio_chat_auto_model(_config()) is None