FoundationAgents/MetaGPT · error · FileNotFoundError

Workflow file not found: {graph_path}

Error message

Workflow file not found: {graph_path}

What it means

AFlow's interface builds the workflow path as <optimized_path>/<dataset>/workflows/round_<round>/graph.py and raises FileNotFoundError when it does not exist. The round is either the one you passed or load_best_round()'s pick, so the error means that round's optimized workflow artifact was never produced (or lives under a different dataset/root path).

Source

Thrown at metagpt/ext/aflow/scripts/interface.py:68

        dataset: 数据集名称
        question: 输入问题
        round: 指定使用的轮次，如果为None则使用最佳轮次
        llm_name: 使用的LLM模型名称
        optimized_path: 优化结果保存路径

    Returns:
        (答案, 成本)的元组
    """
    # 如果没有指定轮次，使用最佳轮次
    if round is None:
        round = load_best_round(dataset, optimized_path)

    logger.info(f"Using round {round} for inference")

    # 构建工作流路径并加载
    graph_path = Path(optimized_path) / dataset / "workflows" / f"round_{round}" / "graph.py"
    if not graph_path.exists():
        raise FileNotFoundError(f"Workflow file not found: {graph_path}")

    # 动态加载工作流类
    WorkflowClass = load_workflow_class(str(graph_path))

    # 创建工作流实例
    llm_config = ModelsConfig.default().get(llm_name)
    workflow = WorkflowClass(
        name=f"{dataset}_workflow",
        llm_config=llm_config,
        dataset=dataset,
    )

    # 执行推理
    if dataset in ["MBPP", "HumanEval"]:
        # 代码类任务需要额外的entry_point参数
        answer, cost = await workflow(question, entry_point=entry_point)
    else:
        answer, cost = await workflow(question)

View on GitHub (pinned to 11cdf466d0)

Solutions

1. List the actual rounds: ls <optimized_path>/<dataset>/workflows/ and pass an existing round number explicitly.
2. Verify optimized_path and dataset strings match the real folder names exactly (path is case-sensitive on Linux).
3. If optimization was interrupted before writing graph.py for the best round, rerun optimization or point at the last complete round.
4. Check load_best_round's source (best.json/experience file) and correct it if it references a round that was never written.

Example fix

# before
result, cost = await run_workflow(dataset="HotpotQA", question=q, optimized_path="../optimized", round=5)
# FileNotFoundError: Workflow file not found: ../optimized/HotpotQA/workflows/round_5/graph.py

# after
from pathlib import Path
rounds = sorted(int(p.name.split("_")[1]) for p in Path("../optimized/HotpotQA/workflows").glob("round_*") if p.joinpath("graph.py").exists())
assert rounds, "no completed workflow rounds; run optimization first"
result, cost = await run_workflow(dataset="HotpotQA", question=q, optimized_path="../optimized", round=rounds[-1])

from pathlib import Path

def latest_complete_round(optimized_path: str, dataset: str) -> int | None:
    wf_dir = Path(optimized_path) / dataset / "workflows"
    rounds = sorted(int(p.name.split("_")[1]) for p in wf_dir.glob("round_*") if (p / "graph.py").is_file())
    return rounds[-1] if rounds else None

rnd = round if round is not None else latest_complete_round(optimized_path, dataset)
if rnd is None:
    raise FileNotFoundError(f"no completed rounds under {optimized_path}/{dataset}; run optimization first")

try:
    result, cost = await run_workflow(dataset=dataset, question=q, optimized_path=optimized_path, round=rnd)
except FileNotFoundError as e:
    raise FileNotFoundError(f"round {rnd} has no graph.py; list rounds via ls {optimized_path}/{dataset}/workflows") from e

When it happens

Trigger: Calling run_aflow or the interface's inference entry with optimized_path/dataset/round combinations where no round_<N>/graph.py exists: round=5 when optimization only completed 3 rounds; wrong optimized_path root; dataset name not matching the folder created during optimization.

Common situations: Inference run before/with an interrupted optimization loop; round=None and the metadata best-round file points at a missing round (crashed optimization); directory renamed or dataset string casing differs from the folder name; copying optimized dirs between machines incompletely.

Related errors

• No requirement document found.
• Unsupported dataset: {dataset}
• Missing fields: {missing_fields}
• use `review` after `fill`
• File {path} not found.

AI-assisted analysis of FoundationAgents/MetaGPT@11cdf466d0 (2026-08-14). Data as JSON: /api/errors/df8e8ee5773a72c2. Report an issue: GitHub.