Hey, thanks for taking over the project and continuing development on it.

Can you please add File Search tool and provide some examples?

Thanks in advance!

Please add support for the new parameter parallel_tool_calls (see https://community.openai.com/t/new-api-feature-disable-parallel-function-calling-via-parallel-tool-calls-false/805405, https://platform.openai.com/docs/api-reference/runs/createThreadAndRun and https://platform.openai.com/docs/api-reference/chat/create).

Use Uploads to upload large files in multiple parts.
https://platform.openai.com/docs/api-reference/uploads

最近更新支持

Using the new FunctionDefinition returns this on azure's gpt-4-turbo

  "error": {
    "message": "1 validation error for Request\nbody -> tools\n  extra fields not permitted (type=value_error.extra)",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }

Looking at the converted JSON (I used ObjectMapper().writeValueAsString so I am not sure if it's accurate)

{
    "type": "function",
    "function": {
      "name": "wolframAlpha",
      "description": "Computes mathematical expressions using WolframAlpha",
      "parameters": {
        "$schema": "http://json-schema.org/draft-04/schema#",
        "title": "Wolfram Alpha Function",
        "type": "object",
        "additionalProperties": false,
        "properties": {
          "input": {
            "type": "string"
          }
        }
      }
    }
  }

seems like azure doesn't support the "$schema", "title", "aditionalProperties" fields as I can't find these in any documentation

another limitation with the new FunctionDefinition is that it doesn't allow parameterless functions, the parameters field is optional and can be omitted

Omitting parameters defines a function with an empty parameter list.

我在尝试将接收到的JSON请求体反序列化为ChatCompletionRequest对象时遇到了问题。尽管我发送的请求数据格式看起来是正确的，但是反序列化过程没有成功。

请求示例
{
"topP": 1,
"temperature": 1,
"maxTokens": 2500,
"messages": [
{
"role": "system",
"content": "你好"
},
{
"role": "user",
"content": [
{
"type": "text",
"text": "识别这张图片"
},
{
"type": "image_url",
"imageUrl": {
"url": "https://img.baigonglian.com/open_platform/generate/1ad71539c74715cfceb1c6b2634e9bf1.png"
}
}
]
}
],
"model": "gpt-4o",
"frequencyPenalty": 0.6299999952316284
}

Is it possible to implement GET in the threads API to find out all the threads you created.

It seems to be possible also if not officially documented using the following example:

def list_threads(api_key):
response = requests.get(
'https://api.openai.com/v1/threads?order=desc&limit=20',
headers={
'Content-Type': 'application/json',
'Authorization': f'Bearer {api_key}',
'OpenAI-Beta': 'assistants=v1'
}
)
if response.status_code == 200:
return response.json()['data']
return None

Please consider this new feature in the next release. Thanks

com.fasterxml.jackson.databind.exc.MismatchedInputException: Cannot deserialize value of type `java.lang.String` from Object value (token `JsonToken.START_OBJECT`)
 at [Source: (String)"{"id":"step_UZoC1MNqKPZ3VcuG2r2k","object":"thread.run.step.delta","delta":{"step_details":{"type":"tool_calls","tool_calls":[{"index":0,"id":"call_mYpvxh6V0XygaLpz1FX","type":"file_search","file_search":{"ranking_options":{"ranker":"default_2024_08_21","score_threshold":0.0},"results":[]}}]}}}"; line: 1, column: 233] (through reference chain: com.theokanning.openai.assistants.run_step.RunStepDelta["delta"]->com.theokanning.openai.assistants.run_step.Delta["step_details"]->com.theokanning.openai.assistants.run_step.StepDetails["tool_calls"]->java.util.ArrayList[0]->com.theokanning.openai.assistants.run.ToolCall["file_search"]->java.util.LinkedHashMap["ranking_options"])

"type":"file_search","file_search":{"ranking_options":{"ranker":"default_2024_08_21","score_threshold":0.0}

Class FileSearchTool needs to add the ranking_options field.

https://platform.openai.com/docs/assistants/tools/file-search/improve-file-search-result-relevance-with-chunk-ranking

        OpenAiService service = getOpenAiService();
        ThreadRequest threadRequest = ThreadRequest.builder().build();
        Thread thread = service.createThread(threadRequest);
        return thread.getId();

springboot 3.1.0,jdk18

developers can now supply a JSON Schema via json_schema, a new option for the response_format parameter.

https://openai.com/index/introducing-structured-outputs-in-the-api

We have been using TheoKanning/openai-java in our service. Since that is not being maintained anymore, we thought of switching over to this project.

One immediate issue we encountered was deprecation warning here

What is the recommended alternative to TikTokensUtil.java ?
We want to count tokens before hand to make a decision whether to even make a request to open ai or not.

我现在会把之前的历史都发给gpt，但是回复的好像有问题
第一次问的是，使用Java读取excel,
第二次问的你好，然后也会把第一次记录发送给gpt，返回的是这种

Returns an error when using the o1-mini model, no such error with 4o-mini

https://pastebin.com/VgZyjjF6

文档关联到assistant，启用file_search和stream

[LogHandler] SSE: {"id":"msg_xxx","object":"thread.message.delta","delta":{"content":[{"index":0,"type":"text","text":{"value":"【272:4†source】","annotations":[{"index":0,"type":"file_citation","text":"【272:4†source】","start_index":357,"end_index":371,"file_citation":{"file_id":"file-xxx","quote":""}}]}}]}} event: THREAD_MESSAGE_DELTA
[LogHandler] SSE: {"id":"msg_xxx","object":"thread.message.delta","delta":{"content":[{"index":0,"type":"text","text":{"value":"."}}]}} event: THREAD_MESSAGE_DELTA
[LogHandler] SSE: {"id":"msg_xxx","object":"thread.message","created_at":1723696331,"assistant_id":"asst_xxx","thread_id":"thread_xxx","run_id":"run_xxx","status":"completed","incomplete_details":null,"incomplete_at":null,"completed_at":1723696333,"role":"assistant","content":[{"type":"text","text":{"value":"To ... on【272:4†source】.","annotations":[{"type":"file_citation","text":"【272:4†source】","start_index":357,"end_index":371,"file_citation":{"file_id":"file-xxx"}}]}}],"attachments":[],"metadata":{}} event: THREAD_MESSAGE_COMPLETED
[LogHandler] message completed
[LogHandler] SSE: {"id":"step_xxx","object":"thread.run.step","created_at":1723696331,"run_id":"run_xxx","assistant_id":"asst_xxx","thread_id":"thread_xxx","type":"message_creation","status":"completed","cancelled_at":null,"completed_at":1723696333,"expires_at":1723696927,"failed_at":null,"last_error":null,"step_details":{"type":"message_creation","message_creation":{"message_id":"msg_xxx"}},"usage":{"prompt_tokens":6064,"completion_tokens":92,"total_tokens":6156}} event: THREAD_RUN_STEP_COMPLETED
[LogHandler] error: Unknown event name: thread.run.incomplete

如果没有【272:4†source】等类似标注的回复，就不会发生这个错误

应该共享一个配置好的ObjectMapper实例。

Customizing File Search settings
The newes change support customize how the file_search tool chunks your data and how many chunks it returns to the model context.

Creating an user image message from a URL is straight forward using UserMessage.buildImageMessage.

Creating an user image message from a local image file is much more tedious, here's some sample code:

public void postDiagramImage(String userMessageText, Path diagramImagePath) {
  List<ChatMessage> messages = ...
  UserMessage imageMessage = createImageMessage(userMessageText, diagramImagePath);
  messages.add(imageMessage);
  ...
}

private UserMessage createImageMessage(String userMessageText, Path imagePath) {
  ImageContent userMessageContent = new ImageContent(userMessageText);
  String imagePathString = imagePath.toAbsolutePath().toString();
  String extension = imagePathString.substring(imagePathString.lastIndexOf('.') + 1);
  String imageUrl = "data:image/" + extension + ";base64," + encodeImage(imagePath);
  ImageContent imageContent = new ImageContent(new ImageUrl(imageUrl));
  UserMessage imageMessage = new UserMessage(List.of(userMessageContent, imageContent));
  return imageMessage;
}
	
private String encodeImage(Path imagePath) {
  byte[] fileContent;
      try {
            fileContent = Files.readAllBytes(imagePath);
            return Base64.getEncoder().encodeToString(fileContent);
      } catch (IOException e){
        throw new RuntimeException(e);
      }
}

It would be great if code like this could be encapsulated, and provided to the user in a friendly manner.

As of 0.21.0, it looks like the parameter max_num_results is not available in FileSearchTool. Is this on the roadmap?

See https://platform.openai.com/docs/api-reference/assistants/createAssistant#assistants-createassistant-tools (request body, under tools -> FileSearch)

看异常应该是有些字段的序列化方式没设置对。

重现测试用例：

public class ChatCompletionExample {

    private static final String BASE_URL = "https://api.openai.com/v1/";
    private static final String API_KEY = "sk-1234567890";

    private static final ObjectMapper objectMapper = new ObjectMapper();

    public static void main(String[] args) throws JsonProcessingException {
        String json = "{\"model\":\"gpt-4o-mini\",\"max_tokens\":150,\"response_format\":{\"type\":\"json_object\"},\"messages\":[{\"role\":\"system\",\"content\":\"You are a helpful assistant and you should reply in JSON format.\"},{\"role\":\"user\",\"content\":\"What is your name?\"}]}";
        ChatCompletionRequest completionRequest = objectMapper.readValue(json, ChatCompletionRequest.class);
        OpenAiService openAiService = new OpenAiService(API_KEY, BASE_URL);
        ChatCompletionResult completionResult = openAiService.createChatCompletion(completionRequest);
        System.out.println(completionResult);
    }
}

报了如下异常：

Exception in thread "main" com.fasterxml.jackson.databind.exc.MismatchedInputException: Cannot construct instance of `com.theokanning.openai.completion.chat.ChatResponseFormat` (although at least one Creator exists): cannot deserialize from Object value (no delegate- or property-based Creator)
 at [Source: (String)"{"model":"gpt-4o-mini","max_tokens":150,"response_format":{"type":"json_object"},"messages":[{"role":"system","content":"You are a helpful assistant and you should reply in JSON format."},{"role":"user","content":"What is your name?"}]}"; line: 1, column: 60] (through reference chain: com.theokanning.openai.completion.chat.ChatCompletionRequest["response_format"])
	at com.fasterxml.jackson.databind.exc.MismatchedInputException.from(MismatchedInputException.java:63)
	at com.fasterxml.jackson.databind.DeserializationContext.reportInputMismatch(DeserializationContext.java:1733)
	at com.fasterxml.jackson.databind.DeserializationContext.handleMissingInstantiator(DeserializationContext.java:1358)
	at com.fasterxml.jackson.databind.deser.BeanDeserializerBase.deserializeFromObjectUsingNonDefault(BeanDeserializerBase.java:1420)
	at com.fasterxml.jackson.databind.deser.BeanDeserializer.deserializeFromObject(BeanDeserializer.java:352)
	at com.fasterxml.jackson.databind.deser.BeanDeserializer.deserialize(BeanDeserializer.java:185)
	at com.fasterxml.jackson.databind.deser.impl.MethodProperty.deserializeAndSet(MethodProperty.java:129)
	at com.fasterxml.jackson.databind.deser.BeanDeserializer.vanillaDeserialize(BeanDeserializer.java:314)
	at com.fasterxml.jackson.databind.deser.BeanDeserializer.deserialize(BeanDeserializer.java:177)
	at com.fasterxml.jackson.databind.deser.DefaultDeserializationContext.readRootValue(DefaultDeserializationContext.java:323)
	at com.fasterxml.jackson.databind.ObjectMapper._readMapAndClose(ObjectMapper.java:4730)
	at com.fasterxml.jackson.databind.ObjectMapper.readValue(ObjectMapper.java:3677)
	at com.fasterxml.jackson.databind.ObjectMapper.readValue(ObjectMapper.java:3645)
	at example.ChatCompletionExample.main(ChatCompletionExample.java:21)

微调的endpoint参数,和最新版openai有一些差异.不是最新的

For example, purpose

Hi, Is there any plan to package the latest version of maven dependencies to upload to remote maven central repository.
This will be handy for me to add dependency packages to the project, thanks again!

I use this API for a Discord bot with ChatGPT
After switching from the gpt-3.5-turbo model to the gpt-4o, sometimes an error started appearing during requests

Log:
https://pastes.dev/b0IFXKBXG1

执行测试用例时，对应messageChunk获取的arguments类型还是objectNode，导致解析为空

我需要在初始化时做什么操作吗，设定不同的序列化方式？

service version

0.20.1

<dependency>
  <groupId>io.github.lambdua</groupId>
  <artifactId>service</artifactId>
  <version>0.20.1</version>
</dependency>

异常信息：

java.lang.ClassNotFoundException: com.fasterxml.jackson.databind.PropertyNamingStrategies
	at java.net.URLClassLoader.findClass(URLClassLoader.java:382)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:357)
	at com.theokanning.openai.service.OpenAiService.defaultObjectMapper(OpenAiService.java:180)
	at com.theokanning.openai.service.OpenAiService.<clinit>(OpenAiService.java:82)
	

目前想要自定义注入拦截器比较复杂,应该在构造函数上添加更灵活的定制化需求

service应该默认支持重试机制

When StreamOption.includeUsage = true, the last usage chunk contains no choices, resulting in an IndexOutOfBoundsException.

Example of the Last Usage Chunk:
{ "id": "chatcmpl-A4yfRYHaCuVyGtk5r0DnV9donhXja", "object": "chat.completion.chunk", "created": 1725749881, "model": "gpt-4o-2024-05-13", "choices": [], "usage": { "prompt_tokens": 611, "completion_tokens": 13, "total_tokens": 624 }, "system_fingerprint": "fp_992d1ea92d" }

The OpenAiService.mapStreamToAccumulator method attempts to access the first choice in the choices array with the following code:
ChatCompletionChoice firstChoice = chunk.getChoices().get(0);

However, this code assumes that there is always at least one choice available. When choices is empty, this leads to a java.lang.IndexOutOfBoundsException.

Although my understanding of the model is limited, it seems reasonable to either:

1. Ignore the usage chunk during mapStreamToAccumulator.
2. Add usage data to AssistantMessage model

Such as official Doc giving :

it is available in py sdk.

Hi @Lambdua, I'm using your library specially for the Assistants API. Since yesterday the added a parallel_tool_calls param, I am getting this error:

Caused by: com.fasterxml.jackson.databind.exc.UnrecognizedPropertyException: Unrecognized field "parallel_tool_calls" (class com.theokanning.openai.assistants.run.Run), not marked as ignorable...

What I did locally as a workaround was to make the JsonUtil internal Mapper a bit more robust, so that unknown properties would not make the parsing fail:

private static class ObjectMapperHolder {
		private static final ObjectMapper MAPPER = new ObjectMapper();
		static {
			MAPPER.configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES, false);
			MAPPER.setSerializationInclusion(JsonInclude.Include.NON_NULL);
			MAPPER.setPropertyNamingStrategy(PropertyNamingStrategies.LOWER_CAMEL_CASE);
		}
	}

What do you think of this change?

ChatFunction ,FunctionExecutor and ChatFunctionDynamic 这几个类以及AssistantFunction导致目前api注入function变得很混乱

There are multiple settings related to function/tool in the code: ChatFunction/AssistantFunction/ChatFunctionDynamic

When using this code:

The messages returned from listMessages don't have an ImageUrl object. They only have a MessageContent object with type image_url but no image_url. The Assistant Playground does show that this data is returned from the API when listing thread messages.

Is it possibly because the messages returned from the Threads API are being coerced to Assistant MessageContent which doesn't cater for image_url?

Structured Outputs via tools is available by setting strict: true within your function definition.

I think there is no documentation about how to send Image from ByteArray or Stream... but I think the ChatGPT allow this ? It is possible to send by url or "text"... what is a text image ??
Thank you for this great API !

现有的token工具好像只能统计实际文字所使用的token，但搜索文件使用的token没有被统计。有什么办法统计搜索文件使用的token数量吗？

The latest assistant supports image recognition

I am the maintainer of this open-source library, and I developed it based on the openai Java library. Currently, the JDK 8 version is no longer suitable for my own needs. I have the idea of upgrading to a higher version of the JDK to use more of its features
However, this idea depends on the thoughts of the people using this library. I hope those who see this issue can provide their own opinions

delete assistant message endpoint
现在api包里面没有呢

感谢作者开源openai4j.

我使用CloudFlare 的AI GateWay 代理的Azure OpenAI.
需要填写defalut_header 和 deafalut_query.

OpenAI的官方python库是支持这个添加的。

我想问下这个库是否支持？
如果不支持的话，能否指导一下如何修改代码？
非常感谢。

ChatCompleteRequest的属性:
@JsonSerialize(using = ChatCompletionRequestFunctionCall.Serializer.class)
@JsonDeserialize(using = ChatCompletionRequestFunctionCall.Deserializer.class)
Object functionCall;

使用时不够友好,应该转为显示类型,减少学习成本

我看现在文件只支持了路径上传，如果字节流的话，现在有方法可以支持吗？

a new refusal string value on API responses which allows developers to programmatically detect if the model has generated a refusal instead of output matching the schema. When the response does not include a refusal and the model’s response has not been prematurely interrupted (as indicated by finish_reason), then the model’s response will reliably produce valid JSON matching the supplied schema.

missing timestamp_granularities parameter