@ui-tars/sdk
Version:
A powerful cross-platform(ANY device/platform) toolkit for building GUI automation agents for UI-TARS
107 lines (106 loc) • 4.49 kB
JavaScript
/**
* Copyright (c) 2025 Bytedance, Inc. and its affiliates.
* SPDX-License-Identifier: Apache-2.0
*/
import * as __WEBPACK_EXTERNAL_MODULE_openai__ from "openai";
import * as __WEBPACK_EXTERNAL_MODULE__ui_tars_action_parser_e6c10e92__ from "@ui-tars/action-parser";
import * as __WEBPACK_EXTERNAL_MODULE__context_useContext_mjs_c75ccb70__ from "./context/useContext.mjs";
import * as __WEBPACK_EXTERNAL_MODULE__types_mjs_4ad50757__ from "./types.mjs";
import * as __WEBPACK_EXTERNAL_MODULE__utils_mjs_25ece7d1__ from "./utils.mjs";
import * as __WEBPACK_EXTERNAL_MODULE__constants_mjs_225410ff__ from "./constants.mjs";
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 __WEBPACK_EXTERNAL_MODULE__types_mjs_4ad50757__.Model {
get factors() {
return __WEBPACK_EXTERNAL_MODULE__constants_mjs_225410ff__.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 __WEBPACK_EXTERNAL_MODULE_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, __WEBPACK_EXTERNAL_MODULE__context_useContext_mjs_c75ccb70__.useContext)();
const compressedImages = await Promise.all(images.map((image)=>(0, __WEBPACK_EXTERNAL_MODULE__utils_mjs_25ece7d1__.preprocessResizeImage)(image, __WEBPACK_EXTERNAL_MODULE__constants_mjs_225410ff__.MAX_PIXELS)));
const messages = (0, __WEBPACK_EXTERNAL_MODULE__utils_mjs_25ece7d1__.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, __WEBPACK_EXTERNAL_MODULE__ui_tars_action_parser_e6c10e92__.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;
}
}
export { UITarsModel };
//# sourceMappingURL=Model.mjs.map