FoundationAgents/MetaGPT · error · ValueError

Rollouts must be greater than 2 if there is no tree to load

Error message

Rollouts must be greater than 2 if there is no tree to load

What it means

Raised by MCTS tree search when no tree could be loaded from disk and the requested rollout count is too small: rollouts is decremented by 2 because two rollouts are consumed to build the initial tree (simulate root + expand_and_simulate), and a negative remainder is rejected. Effectively rollouts must be >= 2 (message says greater than 2; rollouts < 2 after subtracting triggers it).

Source

Thrown at metagpt/ext/sela/search/tree_search.py:434

        rollouts = args.rollouts
        from_scratch = args.from_scratch
        role, root = initialize_di_root_node(state, reflection=reflection)
        self.root_node = root
        self.instruction_generator = InstructionGenerator(
            state=state, use_fixed_insights=self.use_fixed_insights, from_scratch=from_scratch
        )
        await self.instruction_generator.initialize()

        tree_loaded = False
        if load_tree:
            tree_loaded = self.load_tree()
            mcts_logger.log("MCTS", f"Number of simulations: {self.get_num_simulations()}")
            mcts_logger.log("MCTS", f"Tree loaded: {tree_loaded}")

        if not tree_loaded:
            rollouts -= 2  # 2 rollouts for the initial tree
            if rollouts < 0:
                raise ValueError("Rollouts must be greater than 2 if there is no tree to load")
            self.children[root] = []
            reward = await self.simulate(root, role)
            self.backpropagate(root, reward)
            node, reward = await self.expand_and_simulate(root)
            # self.backpropagate(node, reward)
            self.save_node_order(root.id)
            self.save_node_order(node.id)
        else:
            root = self.root_node
            self.load_node_order()

        for _ in range(rollouts):  # number of rollouts
            mcts_logger.log("MCTS", f"Start the next rollout {_+1}")
            node = self.select(root)
            if node.is_terminal():
                if node.raw_value == 0:
                    reward = await self.simulate(node)
                else:

View on GitHub (pinned to 11cdf466d0)

Solutions

1. Increase --rollouts to at least 3 (2 are consumed for initial tree construction)
2. Verify the tree file path in args so a previously saved tree is actually loaded
3. For quick tests, keep rollouts small but >= 3 on a fresh tree

Example fix

# before
--rollouts 1

# after
--rollouts 10

MIN_ROLLOUTS = 3
if not tree_exists(load_tree_path) and rollouts < MIN_ROLLOUTS:
    raise SystemExit("increase --rollouts to >= 3 for a fresh tree")

When it happens

Trigger: Starting a fresh search (no saved tree JSON in the tree save path) with --rollouts 0 or 1; or load_tree failing silently (missing/corrupt tree file) while a small rollout count is used.

Common situations: Smoke-testing with a tiny rollout count; wrong tree save/load directory so an existing tree is not found; deleted tree files.

Related errors

• please check MilvusConnection, uri must be set.
• Dataset {dataset_name} not found in config file. Available d
• Dataset {task_name} not found in config file. Available data
• Number of classes {num_classes} not supported
• Target column not provided

AI-assisted analysis of FoundationAgents/MetaGPT@11cdf466d0 (2026-08-14). Data as JSON: /api/errors/e1a92eb0d7194865. Report an issue: GitHub.