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
from __future__ import annotations
import sys
import pytest
from core.inference.diffusion import DiffusionBackend
from .test_diffusion_backend import (
_FakePipeline,
_FakeTransformer,
_force_cuda_target,
fake_runtime as fake_runtime,
)
@pytest.mark.parametrize(
"token,hub_token,expected",
[
(False, None, False),
(None, None, "omitted"),
("test-token", None, "test-token"),
(None, False, False),
],
ids = ["anonymous-direct", "ambient", "explicit", "anonymous-worker"],
)
def test_hosted_vae_staged_fallback_keeps_pipeline_token(
fake_runtime, monkeypatch, token, hub_token, expected
):
import huggingface_hub
from core.inference import diffusion as dmod
from core.inference import diffusion_te_prequant
class ReachedPipeline(BaseException):
pass
backend = DiffusionBackend()
_force_cuda_target(backend, monkeypatch)
diffusers = sys.modules["diffusers"]
monkeypatch.setattr(diffusers, "QwenImage21Pipeline", _FakePipeline, raising = False)
monkeypatch.setattr(diffusers, "QwenImage21Transformer2DModel", _FakeTransformer, raising = False)
monkeypatch.setattr(dmod, "prefer_ungated_mirror", lambda base, *_a: base)
monkeypatch.setattr(diffusion_te_prequant, "te_prequant_mirror_path", lambda *_a, **_k: None)
seen = []
def download(repo, filename, **kwargs):
seen.append((filename, kwargs.get("token", "omitted")))
raise FileNotFoundError(filename)
def construct(base, **kwargs):
seen.append(("pipeline", kwargs.get("token", "omitted")))
raise ReachedPipeline
monkeypatch.setattr(huggingface_hub, "hf_hub_download", download)
monkeypatch.setattr(_FakePipeline, "from_pretrained", construct)
with pytest.raises(ReachedPipeline):
backend.load_pipeline(
"Qwen/Qwen-Image-2.1",
model_kind = "pipeline",
hf_token = token,
_hub_token = hub_token,
speed_mode = "off",
text_encoder_quant = "off",
transformer_quant = "off",
)
assert (
"vae/qwen_image_2.1_vae_bf16.safetensors",
None if expected == "omitted" else expected,
) in seen
assert seen[-1] == ("pipeline", expected)