Function _run_predictions

eval/run_compare.py:527–576

(
    *,
    spec: AgentSpec,
    swebench_repo: Path,
    swebench_python: str,
    dataset_local: Path,
    split: str,
    prompt_field: str,
    instance_ids: list[str] | None,
    output_path: Path,
    request_timeout: int,
    max_patch_retries: int,
    patch_retry_backoff: float,
    trace_dir: Path | None = None,
    workers: int = 1,
)

def _run_predictions(
    *,
    spec: AgentSpec,
    swebench_repo: Path,
    swebench_python: str,
    dataset_local: Path,
    split: str,
    prompt_field: str,
    instance_ids: list[str] | None,
    output_path: Path,
    request_timeout: int,
    max_patch_retries: int,
    patch_retry_backoff: float,
    trace_dir: Path | None = None,
    workers: int = 1,
) -> None:
    cmd = [
        swebench_python,
        "scripts/run_custom_api.py",
        "--api_url",
        f"http://127.0.0.1:{spec.port}/generate",
        "--dataset_name_or_path",
        str(dataset_local),
        "--split",
        split,
        "--prompt_field",
        prompt_field,
        "--model_name_or_path",
        f"{spec.name}-local",
        "--output_file",
        str(output_path),
        "--timeout",
        str(request_timeout),
        "--append",
        "--max_patch_retries",
        str(max_patch_retries),
        "--patch_retry_backoff_seconds",
        str(patch_retry_backoff),
        "--extra_payload",
        json.dumps(spec.extra_payload),
        "--workers",
        str(workers),
    ]
    if instance_ids:
        cmd.extend(["--instance_ids", ",".join(instance_ids)])
    if trace_dir is not None:
        cmd.extend(["--trace_dir", str(trace_dir)])

    _info(f"  predictions: {spec.name} → {output_path}")
    subprocess.run(cmd, cwd=str(swebench_repo), check=True)  # noqa: S603


def _run_harness(