OpenAI #Chat: 画像付きプロンプトに回答

// Script Example of Business Process Automation
// for 'engine type: 3' ("GraalJS standard mode")
// cf. 'engine type: 2' ("GraalJS Nashorn compatible mode") (renamed from "GraalJS" at 20230526)

//////// START "main()" /////////////////////////////////////////////////////////////////

main();
function main(){ 

////// == Config Retrieving / 工程コンフィグの参照 ==
const strAuthzSetting  = configs.get      ( "AuthzConfU1" );                       /// REQUIRED
  engine.log( " AutomatedTask Config: Authz Setting: " + strAuthzSetting );
const strOrgId         = configs.get      ( "StrConfU2" );                         // NotRequired
  engine.log( " AutomatedTask Config: OpenAI-Organization: " + strOrgId );
const strModel         = configs.get      ( "StrConfM" ) !== "" ?                  // NotRequired
                         configs.get      ( "StrConfM" ) : "gpt-4-vision-preview"; // (default)
  engine.log( " AutomatedTask Config: OpenAI Model: " + strModel );

const strTextPrompt    = configs.get      ( "StrConfA1" );                         /// REQUIRED
  if( strTextPrompt  === "" ){
    throw new Error( "\n AutomatedTask ConfigError:" +
                     " Config {A1: Prompt} must be non-empty \n" );
  }
const strImageUrls     = configs.get      ( "StrConfA2" );                         /// REQUIRED
  if( strImageUrls   === "" ){
    throw new Error( "\n AutomatedTask ConfigError:" +
                     " Config {A2: ImageUrls} must be non-empty \n" );
  }
const arrImageUrls     = strImageUrls.split("\n");

const strDetail        = configs.get      ( "StrConfB1" ) !== "" ?                 // NotRequired
                         configs.get      ( "StrConfB1" ) : "low";                 // (default)
const strMaxTokens     = configs.get      ( "StrConfB2" );                         // NotRequired
const numMaxTokens     = parseInt ( strMaxTokens, 10 );
  engine.log( " AutomatedTask Config: Max Tokens: " + numMaxTokens );

const strPocketGenerated = configs.getObject ( "SelectConfC1" );                   /// REQUIRED NotRequired
const numPocketPrompt    = configs.getObject ( "SelectConfC2" );                   // NotRequired
const numPocketTotal     = configs.getObject ( "SelectConfC3" );                   // NotRequired



////// == Data Retrieving / ワークフローデータの参照 ==
// (Nothing. Retrieved via Expression Language in Config Retrieving)


////// == Calculating / 演算 ==

//// OpenAI API > Documentation > API REFERENCE > CHAT
//// https://platform.openai.com/docs/api-reference/chat/create (not updated)
//// https://platform.openai.com/docs/guides/vision

/// prepare json
let strJson = {};
    strJson.model = strModel;
    if ( ! isNaN(numMaxTokens) ) {
      strJson.max_tokens       = numMaxTokens;
    }
//    strJson.response_format = {};
//    strJson.response_format.type = "json_object";

    strJson.messages = [];
    strJson.messages[0] = {};
    strJson.messages[0].role = "user";
    strJson.messages[0].content = [];
    strJson.messages[0].content[0] = {};
    strJson.messages[0].content[0].type = "text";
    strJson.messages[0].content[0].text = strTextPrompt;

    for ( let i = 0; i < arrImageUrls.length; i++ ) {
      const objTmp = {};
      objTmp.type = "image_url";
      objTmp.image_url = arrImageUrls[i];
      strJson.messages[0].content.push ( objTmp );
    }

/// prepare request1
let request1Uri = "https://api.openai.com/v1/chat/completions";
let request1 = httpClient.begin(); // HttpRequestWrapper
    request1 = request1.queryParam( "detail", strDetail );
    request1 = request1.authSetting( strAuthzSetting ); // with "Authorization: Bearer XX"
    request1 = request1.body( JSON.stringify( strJson ), "application/json" );
    if ( strOrgId !== "" ){
      request1 = request1.header( "OpenAI-Organization", strOrgId );
    }

/// try request1
const response1     = request1.post( request1Uri ); // HttpResponseWrapper
engine.log( " AutomatedTask ApiRequest1 Start: " + request1Uri );
const response1Code = response1.getStatusCode() + ""; // JavaNum to string
const response1Body = response1.getResponseAsString();
engine.log( " AutomatedTask ApiResponse1 Status: " + response1Code );
if( response1Code !== "200"){
  throw new Error( "\n AutomatedTask UnexpectedResponseError: " +
                    response1Code + "\n" + response1Body + "\n" );
}


/// parse response1
/* engine.log( response1Body ); // debug
{
  "id": "chatcmpl-8I9gORGEvsLaVE10Dir0pOMZI5I37",
  "object": "chat.completion",
  "created": 1699337816,
  "model": "gpt-4-1106-vision-preview",
  "usage": {
    "prompt_tokens": 1887, 
    "completion_tokens": 16,
    "total_tokens": 1903
  },
  "choices": [{
    "message": {
      "role": "assistant", 
      "content": "\u6700\u521d\u306e\u753b\u50cf\u306b\u306f\u3001\"Fine tuning workflow\" \u3068\u3044\u3046"
    },
    "finish_details": {
      "type": "max_tokens"
    },
    "index": 0
    }]
}
*/
const response1Obj = JSON.parse( response1Body );
engine.log( " AutomatedTask ApiResponse1 finish_details: " + response1Obj.choices[0].finish_details.type );


////// == Data Updating / ワークフローデータへの代入 ==

if( strPocketGenerated !== null ){
  engine.setData( strPocketGenerated, response1Obj.choices[0].message.content );
}
if( numPocketPrompt !== null ){
  engine.setData( numPocketPrompt, new java.math.BigDecimal( response1Obj.usage.prompt_tokens ) );
}
if( numPocketTotal !== null ){
  engine.setData( numPocketTotal, new java.math.BigDecimal( response1Obj.usage.total_tokens ) );
}

} //////// END "main()" /////////////////////////////////////////////////////////////////


/*
Notes:
- This [Automated Step] obtains the answer text via OpenAI API (Chat endpoint).
    - Specify the instruction (prompt) using Text and Image-Url.
    - Also possible to specify multiple images (Image-Urls).
    - Compatible with GPT-4V preview version (gpt-4-vision-preview).
        - Specifications are subject to change.
        - https://platform.openai.com/docs/guides/vision
- If place this [Automated Atep] in the workflow diagram, communication will occur every time a process arrives.
    - Request from the Questetra BPM Suite server to the OpenAI server.
    - Analyzes the response from the OpenAI server and stores the necessary information.
- [HTTP Authz Settings] is required for workflow apps that include this [Automated Step].
    - An API key is required to use OpenAI API. Please obtain an API key in advance.
        - https://platform.openai.com/api-keys
    - Set 'Secret API Key' as communication token. [HTTP Authz Settings] > [Token Fixed Value]

APPENDIX
- For low res mode, a 512px x 512px image is expected.
- For high res mode,
    - the short side of the image should be less than 768px and
    - the long side of the image should be less than 2,000px.
- Supported type of files
    - PNG (.png), JPEG (.jpeg .jpg), WEBP (.webp), and non-animated GIF (.gif)
- An error may occur depending on the timing.
    - 429 error ('Too Many Requests')
    - Timeout script error `java.util.concurrent.TimeoutException`


Notes-ja:
- この［自動工程］は、OpenAI API （Chat エンドポイント）を通じて、回答文を取得します。
    - 指示文（プロンプト）は Text および Image-Url にて指定します。
    - 複数画像（Image-Urls）の指定も可能です。
    - GPT-4V プレビュー版（gpt-4-vision-preview）に対応します。
        - 仕様が変更される可能性があります。
        - https://platform.openai.com/docs/guides/vision
- この［自動工程］をワークフロー図に配置すれば、案件到達の度に通信が発生します。
    - Questetra BPM Suite サーバから OpenAI サーバに対してリクエストします。
    - OpenAI サーバからのレスポンスを解析し、必要情報を格納します。
- この［自動工程］を含むワークフローアプリには、［HTTP 認証設定］が必要です。
    - OpenAI API の利用には API key が必要です。あらかじめ API Key を取得しておいてください。
        - https://platform.openai.com/api-keys
    - 'Secret API Key' を通信トークンとしてセットします。［HTTP 認証設定］＞［トークン直接指定］

APPENDIX-ja
- 低解像度モード（low）の場合、512px x 512px 画像が推奨です。
- 高解像度モード（high）の場合、
    - 画像の短辺は 768px 未満、
    - 画像の長辺は 2,000px 未満である必要があります。
- サポートされる画像フォーマット
    - PNG (.png), JPEG (.jpeg .jpg), WEBP (.webp), and non-animated GIF (.gif)
- タイミングによって、エラーになる場合があります。
    - 429エラー（'Too Many Requests'）
    - Timeout スクリプトエラー `java.util.concurrent.TimeoutException`
*/