@ui-tars/sdk/dist/Model.js

A powerful cross-platform(ANY device/platform) toolkit for building GUI automation agents for UI-TARS

/**
* Copyright (c) 2025 Bytedance, Inc. and its affiliates.
* SPDX-License-Identifier: Apache-2.0
*/
"use strict";
var __webpack_require__ = {};
(()=>{
    __webpack_require__.n = (module)=>{
        var getter = module && module.__esModule ? ()=>module['default'] : ()=>module;
        __webpack_require__.d(getter, {
            a: getter
        });
        return getter;
    };
})();
(()=>{
    __webpack_require__.d = (exports1, definition)=>{
        for(var key in definition)if (__webpack_require__.o(definition, key) && !__webpack_require__.o(exports1, key)) Object.defineProperty(exports1, key, {
            enumerable: true,
            get: definition[key]
        });
    };
})();
(()=>{
    __webpack_require__.o = (obj, prop)=>Object.prototype.hasOwnProperty.call(obj, prop);
})();
(()=>{
    __webpack_require__.r = function(exports1) {
        if ('undefined' != typeof Symbol && Symbol.toStringTag) Object.defineProperty(exports1, Symbol.toStringTag, {
            value: 'Module'
        });
        Object.defineProperty(exports1, '__esModule', {
            value: true
        });
    };
})();
var __webpack_exports__ = {};
__webpack_require__.r(__webpack_exports__);
__webpack_require__.d(__webpack_exports__, {
    UITarsModel: ()=>UITarsModel
});
const external_openai_namespaceObject = require("openai");
var external_openai_default = /*#__PURE__*/ __webpack_require__.n(external_openai_namespaceObject);
const action_parser_namespaceObject = require("@ui-tars/action-parser");
const useContext_js_namespaceObject = require("./context/useContext.js");
const external_types_js_namespaceObject = require("./types.js");
const external_utils_js_namespaceObject = require("./utils.js");
const external_constants_js_namespaceObject = require("./constants.js");
function _define_property(obj, key, value) {
    if (key in obj) Object.defineProperty(obj, key, {
        value: value,
        enumerable: true,
        configurable: true,
        writable: true
    });
    else obj[key] = value;
    return obj;
}
class UITarsModel extends external_types_js_namespaceObject.Model {
    get factors() {
        return external_constants_js_namespaceObject.DEFAULT_FACTORS;
    }
    get modelName() {
        return this.modelConfig.model ?? 'unknown';
    }
    async invokeModelProvider(params, options) {
        var _result_choices__message, _result_choices_, _result_choices;
        const { messages } = params;
        const { baseURL, apiKey, model, max_tokens = 1000, temperature = 0, top_p = 0.7, ...restOptions } = this.modelConfig;
        const openai = new (external_openai_default())({
            ...restOptions,
            maxRetries: 0,
            baseURL,
            apiKey
        });
        const result = await openai.chat.completions.create({
            model,
            messages,
            stream: false,
            seed: null,
            stop: null,
            frequency_penalty: null,
            presence_penalty: null,
            max_tokens,
            temperature,
            top_p
        }, options);
        return {
            prediction: (null === (_result_choices = result.choices) || void 0 === _result_choices ? void 0 : null === (_result_choices_ = _result_choices[0]) || void 0 === _result_choices_ ? void 0 : null === (_result_choices__message = _result_choices_.message) || void 0 === _result_choices__message ? void 0 : _result_choices__message.content) ?? ''
        };
    }
    async invoke(params) {
        const { conversations, images, screenContext, scaleFactor } = params;
        const { logger, signal } = (0, useContext_js_namespaceObject.useContext)();
        const compressedImages = await Promise.all(images.map((image)=>(0, external_utils_js_namespaceObject.preprocessResizeImage)(image, external_constants_js_namespaceObject.MAX_PIXELS)));
        const messages = (0, external_utils_js_namespaceObject.convertToOpenAIMessages)({
            conversations,
            images: compressedImages
        });
        const startTime = Date.now();
        const result = await this.invokeModelProvider({
            messages
        }, {
            signal
        }).catch((e)=>{
            null == logger || logger.error('[UITarsModel] error', e);
            throw e;
        }).finally(()=>{
            null == logger || logger.info(`[UITarsModel cost]: ${Date.now() - startTime}ms`);
        });
        if (!result.prediction) {
            const err = new Error();
            err.name = 'vlm response error';
            err.stack = JSON.stringify(result) ?? 'no message';
            null == logger || logger.error(err);
            throw err;
        }
        const { prediction } = result;
        try {
            const { parsed: parsedPredictions } = await (0, action_parser_namespaceObject.actionParser)({
                prediction,
                factor: this.factors,
                screenContext,
                scaleFactor
            });
            return {
                prediction,
                parsedPredictions
            };
        } catch (error) {
            null == logger || logger.error('[UITarsModel] error', error);
            return {
                prediction,
                parsedPredictions: []
            };
        }
    }
    constructor(modelConfig){
        super(), _define_property(this, "modelConfig", void 0), this.modelConfig = modelConfig;
        this.modelConfig = modelConfig;
    }
}
var __webpack_export_target__ = exports;
for(var __webpack_i__ in __webpack_exports__)__webpack_export_target__[__webpack_i__] = __webpack_exports__[__webpack_i__];
if (__webpack_exports__.__esModule) Object.defineProperty(__webpack_export_target__, '__esModule', {
    value: true
});

//# sourceMappingURL=Model.js.map