@ui-tars/action-parser/dist/actionParser.mjs

Action parser SDK for UI-TARS

function actionParser(params) {
  const { prediction, factor } = params;
  const parsed = parseActionVlm(prediction, factor);
  return {
    parsed
  };
}
function parseActionVlm(text, factor = 1e3, mode = "bc") {
  let reflection = null;
  let thought = null;
  let actionStr = "";
  text = text.trim();
  if (mode === "bc") {
    if (text.startsWith("Thought:")) {
      const thoughtMatch = text.match(/Thought: ([\s\S]+?)(?=\s*Action:|$)/);
      if (thoughtMatch) {
        thought = thoughtMatch[1].trim();
      }
    } else if (text.startsWith("Reflection:")) {
      const reflectionMatch = text.match(
        /Reflection: ([\s\S]+?)Action_Summary: ([\s\S]+?)(?=\s*Action:|$)/
      );
      if (reflectionMatch) {
        thought = reflectionMatch[2].trim();
        reflection = reflectionMatch[1].trim();
      }
    } else if (text.startsWith("Action_Summary:")) {
      const summaryMatch = text.match(/Action_Summary: (.+?)(?=\s*Action:|$)/);
      if (summaryMatch) {
        thought = summaryMatch[1].trim();
      }
    }
    if (!text.includes("Action:")) {
      actionStr = text;
    } else {
      const actionParts = text.split("Action:");
      actionStr = actionParts[actionParts.length - 1];
    }
  } else if (mode === "o1") {
    const thoughtMatch = text.match(/<Thought>\s*(.*?)\s*<\/Thought>/);
    const actionSummaryMatch = text.match(
      /\nAction_Summary:\s*(.*?)\s*Action:/
    );
    const actionMatch = text.match(/\nAction:\s*(.*?)\s*<\/Output>/);
    const thoughtContent = thoughtMatch ? thoughtMatch[1] : null;
    const actionSummaryContent = actionSummaryMatch ? actionSummaryMatch[1] : null;
    const actionContent = actionMatch ? actionMatch[1] : null;
    thought = `${thoughtContent}
<Action_Summary>
${actionSummaryContent}`;
    actionStr = actionContent || "";
  }
  const allActions = actionStr.split("\n\n");
  const actions = [];
  for (const rawStr of allActions) {
    const actionInstance = parseAction(rawStr.replace(/\n/g, String.raw`\n`).trimStart());
    let actionType = "";
    let actionInputs = {};
    if (actionInstance) {
      actionType = actionInstance.function;
      const params = actionInstance.args;
      actionInputs = {};
      for (const [paramName, param] of Object.entries(params)) {
        if (!param) continue;
        const trimmedParam = param.trim();
        actionInputs[paramName.trim()] = trimmedParam;
        if (paramName.includes("start_box") || paramName.includes("end_box")) {
          const oriBox = trimmedParam;
          const numbers = oriBox.replace(/[()[\]]/g, "").split(",");
          const floatNumbers = numbers.map(
            (num) => Number.parseFloat(num) / factor
          );
          if (floatNumbers.length === 2) {
            floatNumbers.push(floatNumbers[0], floatNumbers[1]);
          }
          actionInputs[paramName.trim()] = JSON.stringify(floatNumbers);
        }
      }
    }
    actions.push({
      reflection,
      thought: thought || "",
      action_type: actionType,
      action_inputs: actionInputs
    });
  }
  return actions;
}
function parseAction(actionStr) {
  try {
    const functionPattern = /^(\w+)\((.*)\)$/;
    const match = actionStr.trim().match(functionPattern);
    if (!match) {
      throw new Error("Not a function call");
    }
    const [_, functionName, argsStr] = match;
    const kwargs = {};
    if (argsStr.trim()) {
      const argPairs = argsStr.match(/([^,']|'[^']*')+/g) || [];
      for (const pair of argPairs) {
        const [key, ...valueParts] = pair.split("=");
        if (!key) continue;
        const value = valueParts.join("=").trim().replace(/^['"]|['"]$/g, "");
        kwargs[key.trim()] = value;
      }
    }
    return {
      function: functionName,
      args: kwargs
    };
  } catch (e) {
    console.error(`Failed to parse action '${actionStr}': ${e}`);
    return null;
  }
}
export {
  actionParser,
  parseActionVlm
};