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 datetime
import json
import pytest
from datasets import Dataset
from utils.datasets import format_dataset
_REPLY = {
"response": "Lower the batch size; l'OOM vient de là.",
"followups": [],
"cites": None,
"resolved": True,
}
_MAPPINGS = pytest.mark.parametrize(
"mapping",
[
None,
{"User": "user", "Assistant": "assistant"},
{"User": "user", "Assistant": "assistant", "__system_prompt": "Be brief."},
],
ids = ["auto_mapping", "user_mapping", "template_mapping"],
)
def _reply(cell, mapping):
row = Dataset.from_dict({"User": ["Training runs out of memory."], "Assistant": [cell]})
reply = format_dataset(row, custom_format_mapping = mapping)["dataset"][0]["conversations"][-1]
assert reply["role"] == "assistant"
return reply["content"]
@_MAPPINGS
def test_structured_cell_trains_as_json(mapping):
content = _reply(_REPLY, mapping)
assert json.loads(content) == _REPLY
assert "l'OOM vient de là" in content
@_MAPPINGS
def test_non_finite_numbers_in_a_structured_cell_train_as_null(mapping):
def reject(constant):
raise ValueError(constant)
content = _reply({"score": float("nan"), "top": [float("inf"), 1.5]}, mapping)
assert json.loads(content, parse_constant = reject) == {"score": None, "top": [None, 1.5]}
@_MAPPINGS
def test_list_cell_with_a_date_trains_as_json(mapping):
cell = [None, {"at": datetime.date(2024, 1, 2)}]
assert _reply(cell, mapping) == '[null, {"at": "2024-01-02"}]'