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.

6.2 KB
Raw Blame
#!/usr/bin/env python
"""
Example script for running benchmarks using the Local Deep Research benchmarking framework.

This script demonstrates how to run SimpleQA and BrowseComp benchmarks programmatically.
"""

import argparse
import os


from local_deep_research.api.benchmark_functions import (
    compare_configurations,
    evaluate_browsecomp,
    evaluate_simpleqa,
)


def main():
    """Run benchmark examples."""
    parser = argparse.ArgumentParser(description="LDR Benchmark Examples")
    parser.add_argument(
        "--benchmark",
        choices=["simpleqa", "browsecomp", "compare"],
        default="simpleqa",
        help="Benchmark to run",
    )
    parser.add_argument(
        "--examples", type=int, default=10, help="Number of examples to use"
    )
    parser.add_argument(
        "--iterations", type=int, default=3, help="Number of search iterations"
    )
    parser.add_argument(
        "--questions", type=int, default=3, help="Questions per iteration"
    )
    parser.add_argument(
        "--search-tool", default="searxng", help="Search tool to use"
    )
    parser.add_argument(
        "--human-eval", action="store_true", help="Use human evaluation"
    )
    parser.add_argument(
        "--output-dir",
        default="benchmark_results",
        help="Directory to save results",
    )

    args = parser.parse_args()

    # Create output directory if it doesn't exist
    os.makedirs(args.output_dir, exist_ok=True)

    print(f"Running {args.benchmark} benchmark with {args.examples} examples")

    # Run the specified benchmark
    if args.benchmark == "simpleqa":
        run_simpleqa_example(args)
    elif args.benchmark == "browsecomp":
        run_browsecomp_example(args)
    elif args.benchmark == "compare":
        run_comparison_example(args)
    else:
        print(f"Unknown benchmark: {args.benchmark}")


def run_simpleqa_example(args):
    """Run SimpleQA benchmark."""
    print("\n=== SimpleQA Benchmark ===")
    print(f"Running with {args.examples} examples")
    print(f"Search iterations: {args.iterations}")
    print(f"Questions per iteration: {args.questions}")
    print(f"Search tool: {args.search_tool}")
    print(f"Human evaluation: {args.human_eval}")
    print(f"Output directory: {args.output_dir}")
    print("=" * 30)

    # Run benchmark
    result = evaluate_simpleqa(
        num_examples=args.examples,
        search_iterations=args.iterations,
        questions_per_iteration=args.questions,
        search_tool=args.search_tool,
        human_evaluation=args.human_eval,
        output_dir=args.output_dir,
    )

    # Print results
    if "metrics" in result:
        print("\nResults:")
        print(f"  Accuracy: {result['metrics'].get('accuracy', 0):.3f}")
        print(f"  Total examples: {result['total_examples']}")
        print(f"  Correct answers: {result['metrics'].get('correct', 0)}")
        print(
            f"  Average time: {result['metrics'].get('average_processing_time', 0):.2f}s"
        )
        print(f"\nReport saved to: {result.get('report_path', 'N/A')}")
    else:
        print("\nBenchmark completed without evaluation")
        print(f"  Results saved to: {result.get('results_path', 'N/A')}")


def run_browsecomp_example(args):
    """Run BrowseComp benchmark."""
    print("\n=== BrowseComp Benchmark ===")
    print(f"Running with {args.examples} examples")
    print(f"Search iterations: {args.iterations}")
    print(f"Questions per iteration: {args.questions}")
    print(f"Search tool: {args.search_tool}")
    print(f"Human evaluation: {args.human_eval}")
    print(f"Output directory: {args.output_dir}")
    print("=" * 30)

    # Run benchmark
    result = evaluate_browsecomp(
        num_examples=args.examples,
        search_iterations=args.iterations,
        questions_per_iteration=args.questions,
        search_tool=args.search_tool,
        human_evaluation=args.human_eval,
        output_dir=args.output_dir,
    )

    # Print results
    if "metrics" in result:
        print("\nResults:")
        print(f"  Accuracy: {result['metrics'].get('accuracy', 0):.3f}")
        print(f"  Total examples: {result['total_examples']}")
        print(f"  Correct answers: {result['metrics'].get('correct', 0)}")
        print(
            f"  Average time: {result['metrics'].get('average_processing_time', 0):.2f}s"
        )
        print(f"\nReport saved to: {result.get('report_path', 'N/A')}")
    else:
        print("\nBenchmark completed without evaluation")
        print(f"  Results saved to: {result.get('results_path', 'N/A')}")


def run_comparison_example(args):
    """Run configuration comparison."""
    print("\n=== Configuration Comparison ===")
    print(f"Dataset: {args.benchmark}")
    print(f"Examples per configuration: {args.examples}")
    print(f"Output directory: {args.output_dir}")
    print("=" * 30)

    # Define configurations to compare
    configurations = [
        {
            "name": "Base Config",
            "search_tool": args.search_tool,
            "iterations": 1,
            "questions_per_iteration": 3,
        },
        {
            "name": "More Iterations",
            "search_tool": args.search_tool,
            "iterations": 3,
            "questions_per_iteration": 3,
        },
        {
            "name": "More Questions",
            "search_tool": args.search_tool,
            "iterations": 1,
            "questions_per_iteration": 5,
        },
    ]

    # Run comparison
    result = compare_configurations(
        dataset_type="simpleqa",  # Use SimpleQA for faster comparison
        num_examples=args.examples,
        configurations=configurations,
        output_dir=args.output_dir,
    )

    # Print results
    print("\nComparison Results:")
    print(f"  Configurations tested: {result['configurations_tested']}")
    print(f"  Report saved to: {result['report_path']}")

    # Print brief comparison table
    print("\nResults Summary:")
    print("Configuration   | Accuracy | Avg. Time")
    print("--------------- | -------- | ---------")
    for res in result["results"]:
        name = res["configuration_name"]
        acc = res.get("metrics", {}).get("accuracy", 0)
        time = res.get("metrics", {}).get("average_processing_time", 0)
        print(f"{name:15} | {acc:.3f}    | {time:.2f}s")


if __name__ == "__main__":
    main()