ChatCompletionInterface.md

Description of the ChatGPT Chat Completion Interface for code generation

Links

https://platform.openai.com/docs/api-reference/chat/create https://platform.openai.com/docs/guides/vision We currently ignore tools and logprobs.

Basic implementation decisions

We use GSON for JSON serialization and deserialization. We use the @SerializedName annotation to map the JSON attributes to the Java attributes if the attribute name has to be different due to Java naming conventions.

Package com.composum.ai.backend.base.service.chat.impl.chatmodel (folder backend/base/src/main/java/com/composum/ai/backend/base/service/chat/impl/chatmodel) contains the generated Java classes for the JSON objects. Use the JavaBean conventions. The class names should be prefixed with ChatCompletion to avoid name clashes. Enumerations should be used for the fixed values and these should be inner classes in the class where they are used.

Request

Attribute role has the fixed values user, assistant and system and should be an enum in the generated code. Attribute type can have the fixed values text and image_url and should be an enum in the generated code.

{
  "model": "gpt-3.5-turbo",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "assistant",
      "content": "Hello!"
    },
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "What’s in this image?"
        },
        {
          "type": "image_url",
          "image_url": {
            "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
          }
        }
      ]
    }
  ],
  "max_tokens": 300,
  "stream": true
}

Response

Attribute role has the fixed values user, assistant and system and should be an enum in the generated code. Attribute logprobs should be ignored in the generated code. Attribute finish_reason has the fixed values stop, length, content_filter and should be an enum in the generated code.

The normal response and the streaming response should be mapped to the same Java class.

{
  "id": "chatcmpl-123",
  "object": "chat.completion",
  "created": 1677652288,
  "model": "gpt-3.5-turbo-0613",
  "system_fingerprint": "fp_44709d6fcb",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "\n\nHello there, how may I assist you today?"
      },
      "logprobs": null,
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 9,
    "completion_tokens": 12,
    "total_tokens": 21
  }
}

Streaming Response

{
  "id": "chatcmpl-123",
  "object": "chat.completion.chunk",
  "created": 1694268190,
  "model": "gpt-3.5-turbo-0613",
  "system_fingerprint": "fp_44709d6fcb",
  "choices": [
    {
      "index": 0,
      "delta": {
        "role": "assistant",
        "content": ""
      },
      "logprobs": null,
      "finish_reason": null
    }
  ]
}

{
  "id": "chatcmpl-123",
  "object": "chat.completion.chunk",
  "created": 1694268190,
  "model": "gpt-3.5-turbo-0613",
  "system_fingerprint": "fp_44709d6fcb",
  "choices": [
    {
      "index": 0,
      "delta": {
        "content": "Hello"
      },
      "logprobs": null,
      "finish_reason": null
    }
  ]
}

....

{
  "id": "chatcmpl-123",
  "object": "chat.completion.chunk",
  "created": 1694268190,
  "model": "gpt-3.5-turbo-0613",
  "system_fingerprint": "fp_44709d6fcb",
  "choices": [
    {
      "index": 0,
      "delta": {
        "content": "?"
      },
      "logprobs": null,
      "finish_reason": null
    }
  ]
}

{
  "id": "chatcmpl-123",
  "object": "chat.completion.chunk",
  "created": 1694268190,
  "model": "gpt-3.5-turbo-0613",
  "system_fingerprint": "fp_44709d6fcb",
  "choices": [
    {
      "index": 0,
      "delta": {},
      "logprobs": null,
      "finish_reason": "stop"
    }
  ]
}