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 pytest
from datasets import Dataset
from PIL import Image
from utils.datasets import format_and_template_dataset, llm_assist
HELPER_SENTENCE = "Solve the math problem shown in the image."
GENERIC_SENTENCE = "Describe this image in detail."
@pytest.fixture(autouse = True)
def _helper_writes_one_sentence(monkeypatch):
monkeypatch.setattr(
llm_assist,
"llm_generate_vlm_instruction",
lambda **kwargs: {"instruction": HELPER_SENTENCE, "confidence": 0.85},
)
def _format(rows, mapping = None):
images = [Image.new("RGB", (8, 8), (i * 40, 0, 0)) for i in range(2)]
ds = Dataset.from_dict({"image": images, **rows})
info = format_and_template_dataset(
ds,
model_name = "unsloth/Qwen2.5-VL-3B-Instruct",
tokenizer = None,
is_vlm = True,
dataset_name = "org/ds",
custom_format_mapping = mapping,
)
assert info["success"], info["errors"]
return info["dataset"]
def _turns(sample):
user, assistant = sample["messages"]
user_text = next(part["text"] for part in user["content"] if part["type"] == "text")
return user_text, assistant["content"][0]["text"]
@pytest.mark.parametrize(
"question_col, answer_col, mapping",
[
("problem", "solution", True),
("problem", "answer", False),
("Question", "Answer", False),
("inputs", "outputs", True),
("input", "output", False),
],
)
def test_each_row_keeps_its_own_question(question_col, answer_col, mapping):
questions = ["Find x in the triangle.", "How many red cubes are left?"]
answers = ["x = 42", "3"]
out = _format(
{question_col: questions, answer_col: answers},
{"image": "image", answer_col: "text"} if mapping else None,
)
assert [_turns(sample) for sample in out] == list(zip(questions, answers))
def test_answer_column_is_not_reused_as_the_question():
prompts = ["a red square", "a dark red square"]
out = _format({"prompt": prompts}, {"image": "image", "prompt": "text"})
assert [_turns(sample) for sample in out] == [(HELPER_SENTENCE, p) for p in prompts]
def test_exact_question_column_wins_over_a_case_variant():
questions = ["Find x in the triangle.", "How many red cubes are left?"]
answers = ["x = 42", "3"]
out = _format(
{"Question": ["upper one", "upper two"], "question": questions, "answer": answers}
)
assert [_turns(sample) for sample in out] == list(zip(questions, answers))
def test_populated_case_variant_wins_over_an_empty_exact_column():
questions = ["Find x in the triangle.", "How many red cubes are left?"]
answers = ["x = 42", "3"]
out = _format({"Question": questions, "question": [None, None], "answer": answers})
assert [_turns(sample) for sample in out] == list(zip(questions, answers))
def test_structured_input_column_is_not_used_as_the_question():
answers = ["a red square", "a dark red square"]
out = _format({"inputs": [[101, 102], [103]], "caption": answers})
assert [_turns(sample) for sample in out] == [(HELPER_SENTENCE, a) for a in answers]
def test_missing_question_value_uses_the_generic_instruction():
questions = ["Find x in the triangle.", None]
answers = ["x = 42", "3"]
out = _format({"problem": questions, "answer": answers})
assert [_turns(sample) for sample in out] == [
(questions[0], answers[0]),
(GENERIC_SENTENCE, answers[1]),
]
def test_later_question_value_is_kept_when_the_first_is_missing():
questions = [None, "How many red cubes are left?"]
answers = ["x = 42", "3"]
out = _format({"problem": questions, "answer": answers})
assert [_turns(sample) for sample in out] == [
(GENERIC_SENTENCE, answers[0]),
(questions[1], answers[1]),
]