Added response_format capabilities to chat completion request

api/src/main/java/com/theokanning/openai/completion/chat/ChatCompletionRequest.java

@@ -15,11 +15,6 @@
 @NoArgsConstructor
 public class ChatCompletionRequest {
 
-    /**
-     * ID of the model to use.
-     */
-    String model;
-
     /**
      * The messages to generate chat completions for, in the <a
      * href="https://platform.openai.com/docs/guides/chat/introduction">chat format</a>.<br>
@@ -28,36 +23,38 @@ public class ChatCompletionRequest {
     List<ChatMessage> messages;
 
     /**
-     * What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower
-     * values like 0.2 will make it more focused and deterministic.<br>
-     * We generally recommend altering this or top_p but not both.
-     */
-    Double temperature;
-
-    /**
-     * An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens
-     * with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.<br>
-     * We generally recommend altering this or temperature but not both.
+     * ID of the model to use.
      */
-    @JsonProperty("top_p")
-    Double topP;
+    String model;
 
     /**
-     * How many chat completion chatCompletionChoices to generate for each input message.
+     * Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far,
+     * decreasing the model's likelihood to repeat the same line verbatim.
      */
-    Integer n;
+    @JsonProperty("frequency_penalty")
+    Double frequencyPenalty;
 
     /**
-     * If set, partial message deltas will be sent, like in ChatGPT. Tokens will be sent as data-only <a
-     * href="https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format">server-sent
-     * events</a> as they become available, with the stream terminated by a data: [DONE] message.
+     * <p>An object specifying the format that the model must output.</p>
+     *
+     * <p>Setting to { "type": "json_object" } enables JSON mode, which guarantees the message the model generates is valid JSON.</p>
+     *
+     * <p><b>Important:</b> when using JSON mode, you must also instruct the model to produce JSON yourself via a system or user message.
+     * Without this, the model may generate an unending stream of whitespace until the generation reaches the token limit, resulting
+     * in a long-running and seemingly "stuck" request. Also note that the message content may be partially cut off if
+     * finish_reason="length", which indicates the generation exceeded max_tokens or the conversation exceeded the max context length.</p>
      */
-    Boolean stream;
+    @JsonProperty("response_format")
+    ResponseFormat responseFormat;
 
     /**
-     * Up to 4 sequences where the API will stop generating further tokens.
+     * Accepts a json object that maps tokens (specified by their token ID in the tokenizer) to an associated bias value from -100
+     * to 100. Mathematically, the bias is added to the logits generated by the model prior to sampling. The exact effect will
+     * vary per model, but values between -1 and 1 should decrease or increase likelihood of selection; values like -100 or 100
+     * should result in a ban or exclusive selection of the relevant token.
      */
-    List<String> stop;
+    @JsonProperty("logit_bias")
+    Map<String, Integer> logitBias;
 
     /**
      * The maximum number of tokens allowed for the generated answer. By default, the number of tokens the model can return will
@@ -66,6 +63,11 @@ public class ChatCompletionRequest {
     @JsonProperty("max_tokens")
     Integer maxTokens;
 
+    /**
+     * How many chat completion chatCompletionChoices to generate for each input message.
+     */
+    Integer n;
+
     /**
      * Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far,
      * increasing the model's likelihood to talk about new topics.
@@ -74,38 +76,48 @@ public class ChatCompletionRequest {
     Double presencePenalty;
 
     /**
-     * Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far,
-     * decreasing the model's likelihood to repeat the same line verbatim.
+     * Up to 4 sequences where the API will stop generating further tokens.
      */
-    @JsonProperty("frequency_penalty")
-    Double frequencyPenalty;
+    List<String> stop;
 
     /**
-     * Accepts a json object that maps tokens (specified by their token ID in the tokenizer) to an associated bias value from -100
-     * to 100. Mathematically, the bias is added to the logits generated by the model prior to sampling. The exact effect will
-     * vary per model, but values between -1 and 1 should decrease or increase likelihood of selection; values like -100 or 100
-     * should result in a ban or exclusive selection of the relevant token.
+     * If set, partial message deltas will be sent, like in ChatGPT. Tokens will be sent as data-only <a
+     * href="https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format">server-sent
+     * events</a> as they become available, with the stream terminated by a data: [DONE] message.
      */
-    @JsonProperty("logit_bias")
-    Map<String, Integer> logitBias;
+    Boolean stream;
 
+    /**
+     * What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower
+     * values like 0.2 will make it more focused and deterministic.<br>
+     * We generally recommend altering this or top_p but not both.
+     */
+    Double temperature;
 
     /**
-     * A unique identifier representing your end-user, which will help OpenAI to monitor and detect abuse.
+     * An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens
+     * with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.<br>
+     * We generally recommend altering this or temperature but not both.
      */
-    String user;
+    @JsonProperty("top_p")
+    Double topP;
 
     /**
-     * A list of the available functions.
+     * A unique identifier representing your end-user, which will help OpenAI to monitor and detect abuse.
      */
-    List<?> functions;
+    String user;
 
     /**
      * Controls how the model responds to function calls, as specified in the <a href="https://platform.openai.com/docs/api-reference/chat/create#chat/create-function_call">OpenAI documentation</a>.
      */
     @JsonProperty("function_call")
     ChatCompletionRequestFunctionCall functionCall;
 
+    /**
+     * A list of the available functions.
+     */
+    List<?> functions;
+
     @Data
     @Builder
     @AllArgsConstructor
@@ -118,4 +130,18 @@ public static ChatCompletionRequestFunctionCall of(String name) {
         }
 
     }
+
+    @Data
+    @Builder
+    @AllArgsConstructor
+    @NoArgsConstructor
+    public static class ResponseFormat {
+        String type;
+
+        public static ResponseFormat of(String type) {
+            return new ResponseFormat(type);
+        }
+
+    }
+
 }

service/src/test/java/com/theokanning/openai/service/ChatCompletionTest.java

@@ -2,11 +2,14 @@
 
 import com.fasterxml.jackson.annotation.JsonProperty;
 import com.fasterxml.jackson.annotation.JsonPropertyDescription;
+import com.fasterxml.jackson.core.JsonParser;
 import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
 import com.fasterxml.jackson.databind.node.ObjectNode;
 import com.theokanning.openai.completion.chat.*;
 import org.junit.jupiter.api.Test;
 
+import java.io.IOException;
 import java.util.*;
 
 import static org.junit.jupiter.api.Assertions.*;
@@ -23,7 +26,7 @@ static class Weather {
     }
 
     enum WeatherUnit {
-        CELSIUS, FAHRENHEIT;
+        CELSIUS, FAHRENHEIT
     }
 
     static class WeatherResponse {
@@ -300,4 +303,46 @@ void streamChatCompletionWithDynamicFunctions() {
         assertNotNull(accumulatedMessage.getFunctionCall().getArguments().get("unit"));
     }
 
+    @Test
+    void streamChatCompletionWithJsonResponseFormat() {
+        final List<ChatMessage> messages = new ArrayList<>();
+
+        // The system message is deliberately vague in order to not give too much of a direction of how response should look like.
+        // The main gist there is that chat competition should always contain JSON content.
+        final ChatMessage systemMessage = new ChatMessage(
+                ChatMessageRole.SYSTEM.value(),
+                "You are a dog and will speak as such - but please do it in JSON."
+        );
+
+        messages.add(systemMessage);
+
+        ChatCompletionRequest chatCompletionRequest = ChatCompletionRequest
+                .builder()
+                .model("gpt-4-1106-preview")
+                .messages(messages)
+                .n(1)
+                .maxTokens(256)
+                .responseFormat(ChatCompletionRequest.ResponseFormat.of("json_object"))
+                .build();
+
+        ChatCompletionResult chatCompletion = service.createChatCompletion(chatCompletionRequest);
+
+        ChatCompletionChoice chatCompletionChoice = chatCompletion.getChoices().get(0);
+        String expectedJsonContent = chatCompletionChoice.getMessage().getContent();
+
+        assertTrue(isValidJSON(expectedJsonContent), "Invalid JSON response:\n\n" + expectedJsonContent);
+    }
+
+    private boolean isValidJSON(String json) {
+        try (final JsonParser parser = new ObjectMapper().createParser(json)) {
+            while (parser.nextToken() != null) {
+                // Just try to read all tokens in order to verify whether this is valid json.
+            }
+            return true;
+        } catch (IOException ioe) {
+            ioe.printStackTrace();
+            return false;
+        }
+    }
+
 }