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
"""A requested MiniMax-H3 step is one denoiser evaluation on the real diffusers scheduler."""
from __future__ import annotations
import pytest
from core.inference import video_minimax_h3 as h3
def _scheduler():
pytest.importorskip("torch")
diffusers = pytest.importorskip("diffusers")
cls = getattr(diffusers, "MiniMaxH3Scheduler", None)
if cls is None:
pytest.skip("this diffusers has no MiniMaxH3Scheduler")
return cls(shift = 12.0)
@pytest.mark.parametrize("steps", [1, 2, 8, 20, 30, 50])
def test_requested_steps_are_model_evaluations(steps):
sched = _scheduler()
sched.set_timesteps(h3.h3_num_inference_steps(steps))
assert len(sched.timesteps) == steps
assert float(sched.sigmas[-1]) == 0.0 and float(sched.sigmas[0]) == 1.0
def test_kill_switch_keeps_the_grid_point_count(monkeypatch):
monkeypatch.setenv(h3.H3_STEP_EVALUATIONS_ENV, "0")
sched = _scheduler()
sched.set_timesteps(h3.h3_num_inference_steps(20))
assert len(sched.timesteps) == 19