Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

13.3 KB
Raw Blame
"""
Tests for benchmarks/runners.py

Tests cover:
- format_query function
- run_benchmark function
- process_example function
- Result saving and loading
"""

from unittest.mock import Mock, patch
from pathlib import Path
import tempfile

import pytest


@pytest.fixture(autouse=True)
def _pin_data_directory(tmp_path, monkeypatch):
    """Keep benchmark output out of the real user data directory.

    run_benchmark resolves its output directory inside LDR_DATA_DIR and replaces
    any other location with a default directory there. Pinning the data
    directory to this test's tmp_path keeps that default per test.
    """
    monkeypatch.setenv("LDR_DATA_DIR", str(tmp_path))


class TestFormatQuery:
    """Tests for the format_query function."""

    def test_format_query_simpleqa(self):
        """SimpleQA returns question unchanged."""
        from local_deep_research.benchmarks.runners import format_query

        question = "What is the capital of France?"
        result = format_query(question, "simpleqa")

        assert result == question

    def test_format_query_browsecomp(self):
        """BrowseComp formats with template."""
        from local_deep_research.benchmarks.runners import format_query

        question = "What is the capital of France?"
        result = format_query(question, "browsecomp")

        # Should contain the question
        assert question in result
        # Should be longer than the question (template added)
        assert len(result) > len(question)

    def test_format_query_default(self):
        """Default format returns question unchanged."""
        from local_deep_research.benchmarks.runners import format_query

        question = "What is the capital of France?"
        result = format_query(question, "unknown_type")

        assert result == question

    def test_format_query_case_insensitive(self):
        """Dataset type is case insensitive."""
        from local_deep_research.benchmarks.runners import format_query

        question = "What is the capital of France?"
        result1 = format_query(question, "BROWSECOMP")
        result2 = format_query(question, "BrowseComp")
        result3 = format_query(question, "browsecomp")

        assert result1 == result2 == result3


class TestRunBenchmark:
    """Tests for the run_benchmark function."""

    def test_run_benchmark_creates_output_dir(self, monkeypatch):
        """run_benchmark creates output directory."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            # run_benchmark output_dir is now confined to the LDR data directory,
            # so point the data dir at tmpdir to keep this out-of-tree output
            # location honoured rather than clamped to the default.
            monkeypatch.setenv("LDR_DATA_DIR", tmpdir)
            output_dir = Path(tmpdir) / "new_dir"

            # Mock the dataset loading and search
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                try:
                    run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=str(output_dir),
                        run_evaluation=False,
                    )
                except Exception:
                    pass  # May fail for other reasons, but dir should be created

                # Directory should be created
                assert output_dir.exists()

    def test_run_benchmark_default_search_config(self):
        """run_benchmark uses default search config when not provided."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                with patch(
                    "local_deep_research.benchmarks.runners.generate_report"
                ) as mock_report:
                    mock_report.return_value = "Test report"

                    result = run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=tmpdir,
                        run_evaluation=False,
                    )

                    # Should return a result dict
                    assert isinstance(result, dict)

    def test_run_benchmark_custom_search_config(self):
        """run_benchmark uses custom search config when provided."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                with patch(
                    "local_deep_research.benchmarks.runners.generate_report"
                ) as mock_report:
                    mock_report.return_value = "Test report"

                    custom_config = {
                        "iterations": 5,
                        "questions_per_iteration": 2,
                        "search_tool": "wikipedia",
                    }

                    result = run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=tmpdir,
                        search_config=custom_config,
                        run_evaluation=False,
                    )

                    assert isinstance(result, dict)

    def test_run_benchmark_with_progress_callback(self):
        """run_benchmark calls progress callback."""
        from local_deep_research.benchmarks.runners import run_benchmark

        callback = Mock()

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = [
                    {"problem": "Q1", "answer": "A1"}
                ]
                mock_registry.create_dataset.return_value = mock_dataset
                # Return a class that will fail isinstance check gracefully
                mock_registry.get_dataset_class.return_value = type(
                    "FakeDataset", (), {}
                )

                with patch(
                    "local_deep_research.benchmarks.runners.quick_summary"
                ) as mock_summary:
                    mock_summary.return_value = {"content": "Answer"}

                    with patch(
                        "local_deep_research.benchmarks.runners.grade_results"
                    ) as mock_grade:
                        mock_grade.return_value = []

                        with patch(
                            "local_deep_research.benchmarks.runners.generate_report"
                        ) as mock_report:
                            mock_report.return_value = "Report"

                            run_benchmark(
                                dataset_type="simpleqa",
                                num_examples=1,
                                output_dir=tmpdir,
                                progress_callback=callback,
                                run_evaluation=False,
                            )

                            # Callback should be called
                            assert callback.call_count >= 1


class TestDatasetRegistry:
    """Tests for DatasetRegistry interaction."""

    def test_dataset_registry_get_available_datasets(self):
        """DatasetRegistry returns available datasets."""
        from local_deep_research.benchmarks.datasets.base import (
            DatasetRegistry,
        )

        # Check that registry has registered datasets
        registered = DatasetRegistry.get_available_datasets()
        assert isinstance(registered, list)

    def test_dataset_registry_create_dataset_method_exists(self):
        """DatasetRegistry has create_dataset method."""
        from local_deep_research.benchmarks.datasets.base import (
            DatasetRegistry,
        )

        assert hasattr(DatasetRegistry, "create_dataset")
        assert callable(DatasetRegistry.create_dataset)

    def test_dataset_registry_load_dataset_method_exists(self):
        """DatasetRegistry has load_dataset method."""
        from local_deep_research.benchmarks.datasets.base import (
            DatasetRegistry,
        )

        assert hasattr(DatasetRegistry, "load_dataset")
        assert callable(DatasetRegistry.load_dataset)


class TestResultsSaving:
    """Tests for results saving functionality."""

    def test_results_saved_as_json(self):
        """Results are saved as JSON files."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                with patch(
                    "local_deep_research.benchmarks.runners.generate_report"
                ) as mock_report:
                    mock_report.return_value = "Test report"

                    run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=tmpdir,
                        run_evaluation=False,
                    )

                    # Check for JSON files in output directory
                    files = list(Path(tmpdir).iterdir())
                    json_files = [f for f in files if f.suffix == ".json"]
                    # May or may not have files depending on results
                    assert isinstance(json_files, list)


class TestBrowseCompSpecificBehavior:
    """Tests for BrowseComp-specific benchmark behavior."""

    def test_browsecomp_uses_template(self):
        """BrowseComp benchmark uses the template."""
        from local_deep_research.benchmarks.runners import format_query
        from local_deep_research.benchmarks.templates import (
            BROWSECOMP_QUERY_TEMPLATE,
        )

        question = "Test question"
        result = format_query(question, "browsecomp")

        # Result should be the template with question substituted
        expected = BROWSECOMP_QUERY_TEMPLATE.format(question=question)
        assert result == expected


class TestEvaluationConfig:
    """Tests for evaluation configuration."""

    def test_run_benchmark_with_evaluation_config(self):
        """run_benchmark accepts evaluation config."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                with patch(
                    "local_deep_research.benchmarks.runners.generate_report"
                ) as mock_report:
                    mock_report.return_value = "Test report"

                    eval_config = {
                        "model_name": "gpt-4",
                        "temperature": 0,
                    }

                    result = run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=tmpdir,
                        run_evaluation=False,
                        evaluation_config=eval_config,
                    )

                    assert isinstance(result, dict)

    def test_run_benchmark_human_evaluation_flag(self):
        """run_benchmark accepts human_evaluation flag."""
        from local_deep_research.benchmarks.runners import run_benchmark

        with tempfile.TemporaryDirectory() as tmpdir:
            with patch(
                "local_deep_research.benchmarks.runners.DatasetRegistry"
            ) as mock_registry:
                mock_dataset = Mock()
                mock_dataset.load.return_value = []
                mock_registry.create_dataset.return_value = mock_dataset

                with patch(
                    "local_deep_research.benchmarks.runners.generate_report"
                ) as mock_report:
                    mock_report.return_value = "Test report"

                    result = run_benchmark(
                        dataset_type="simpleqa",
                        num_examples=0,
                        output_dir=tmpdir,
                        run_evaluation=False,
                        human_evaluation=True,
                    )

                    assert isinstance(result, dict)