Adding service_tier for request and response ChatCompletions (Azure#44719)

* Adding service_tier for request and response ChatCompletions

* Alternative spelling

* Removed generated annotation

Author: jpalvarezl

sdk/openai/azure-ai-openai/CHANGELOG.md

@@ -4,6 +4,8 @@
 
 ### Features Added
 
+- Added `ServiceTier` and `ServiceTierOptions` for `ChatCompletions` and `ChatCompletionsOptions` respectively. This is exclusively a non-Azure OpenAI issue and was requested in [this issue](https://github.com/Azure/azure-sdk-for-java/issues/41695#issuecomment-2736037879).
+
 ### Breaking Changes
 
 ### Bugs Fixed

sdk/openai/azure-ai-openai/src/main/java/com/azure/ai/openai/models/ChatCompletions.java

@@ -43,6 +43,8 @@ public final class ChatCompletions implements JsonSerializable<ChatCompletions>
     @Generated
     private final CompletionsUsage usage;
 
+    private ServiceTier serviceTier;
+
     /**
      * Get the id property: A unique identifier associated with this chat completions response.
      *
@@ -95,6 +97,15 @@ public OffsetDateTime getCreatedAt() {
         return OffsetDateTime.ofInstant(Instant.ofEpochSecond(this.createdAt), ZoneOffset.UTC);
     }
 
+    /**
+     * This field is not available in Azure OpenAI. The service tier used for processing the request.
+     *
+     * @return the serviceTier value.
+     */
+    public ServiceTier getServiceTier() {
+        return serviceTier;
+    }
+
     /*
      * Content filtering results for zero or more prompts in the request. In a streaming request,
      * results for different prompts may arrive at different times or in different orders.
@@ -173,7 +184,6 @@ public String getModel() {
     /**
      * {@inheritDoc}
      */
-    @Generated
     @Override
     public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
         jsonWriter.writeStartObject();
@@ -185,6 +195,9 @@ public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
         jsonWriter.writeArrayField("prompt_filter_results", this.promptFilterResults,
             (writer, element) -> writer.writeJson(element));
         jsonWriter.writeStringField("system_fingerprint", this.systemFingerprint);
+        if (this.serviceTier != null) {
+            jsonWriter.writeStringField("service_tier", this.serviceTier.toString());
+        }
         return jsonWriter.writeEndObject();
     }
 
@@ -197,7 +210,6 @@ public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
      * @throws IllegalStateException If the deserialized JSON object was missing any required properties.
      * @throws IOException If an error occurs while reading the ChatCompletions.
      */
-    @Generated
     public static ChatCompletions fromJson(JsonReader jsonReader) throws IOException {
         return jsonReader.readObject(reader -> {
             String id = null;
@@ -206,6 +218,7 @@ public static ChatCompletions fromJson(JsonReader jsonReader) throws IOException
             CompletionsUsage usage = null;
             String model = null;
             List<ContentFilterResultsForPrompt> promptFilterResults = null;
+            ServiceTier serviceTier = null;
             String systemFingerprint = null;
             while (reader.nextToken() != JsonToken.END_OBJECT) {
                 String fieldName = reader.getFieldName();
@@ -224,6 +237,8 @@ public static ChatCompletions fromJson(JsonReader jsonReader) throws IOException
                     promptFilterResults = reader.readArray(reader1 -> ContentFilterResultsForPrompt.fromJson(reader1));
                 } else if ("system_fingerprint".equals(fieldName)) {
                     systemFingerprint = reader.getString();
+                } else if ("service_tier".equals(fieldName)) {
+                    serviceTier = ServiceTier.fromString(reader.getString());
                 } else {
                     reader.skipChildren();
                 }
@@ -232,6 +247,7 @@ public static ChatCompletions fromJson(JsonReader jsonReader) throws IOException
             deserializedChatCompletions.model = model;
             deserializedChatCompletions.promptFilterResults = promptFilterResults;
             deserializedChatCompletions.systemFingerprint = systemFingerprint;
+            deserializedChatCompletions.serviceTier = serviceTier;
             return deserializedChatCompletions;
         });
     }

sdk/openai/azure-ai-openai/src/main/java/com/azure/ai/openai/models/ChatCompletionsOptions.java

@@ -221,6 +221,11 @@ public final class ChatCompletionsOptions implements JsonSerializable<ChatComple
     @Generated
     private Boolean parallelToolCalls;
 
+    /*
+     * This field is not available in Azure OpenAI. Specifies the latency tier to use for processing the request.
+     */
+    private ServiceTierOptions serviceTierOptions;
+
     /**
      * Creates an instance of ChatCompletionsOptions class.
      *
@@ -903,10 +908,29 @@ public ChatCompletionsOptions setParallelToolCalls(Boolean parallelToolCalls) {
         return this;
     }
 
+    /**
+     * Get the {@link ServiceTierOptions} for this request.
+     *
+     * @return the {@link ServiceTierOptions} value.
+     */
+    public ServiceTierOptions getServiceTierOptions() {
+        return this.serviceTierOptions;
+    }
+
+    /**
+     * Set the {@link ServiceTierOptions} for this request.
+     *
+     * @param serviceTierOptions the {@link ServiceTierOptions} value to set.
+     * @return the ChatCompletionsOptions object itself.
+     */
+    public ChatCompletionsOptions setServiceTierOptions(ServiceTierOptions serviceTierOptions) {
+        this.serviceTierOptions = serviceTierOptions;
+        return this;
+    }
+
     /**
      * {@inheritDoc}
      */
-    @Generated
     @Override
     public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
         jsonWriter.writeStartObject();
@@ -950,6 +974,9 @@ public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
             (writer, element) -> writer.writeString(element == null ? null : element.toString()));
         jsonWriter.writeJsonField("prediction", this.prediction);
         jsonWriter.writeJsonField("audio", this.audio);
+        if (this.serviceTierOptions != null) {
+            jsonWriter.writeStringField("service_tier", this.serviceTierOptions.toString());
+        }
         return jsonWriter.writeEndObject();
     }
 
@@ -962,7 +989,6 @@ public JsonWriter toJson(JsonWriter jsonWriter) throws IOException {
      * @throws IllegalStateException If the deserialized JSON object was missing any required properties.
      * @throws IOException If an error occurs while reading the ChatCompletionsOptions.
      */
-    @Generated
     public static ChatCompletionsOptions fromJson(JsonReader jsonReader) throws IOException {
         return jsonReader.readObject(reader -> {
             List<ChatRequestMessage> messages = null;
@@ -997,6 +1023,7 @@ public static ChatCompletionsOptions fromJson(JsonReader jsonReader) throws IOEx
             List<ChatCompletionModality> modalities = null;
             PredictionContent prediction = null;
             AudioOutputParameters audio = null;
+            ServiceTierOptions serviceTierOptions = null;
             while (reader.nextToken() != JsonToken.END_OBJECT) {
                 String fieldName = reader.getFieldName();
                 reader.nextToken();
@@ -1066,6 +1093,8 @@ public static ChatCompletionsOptions fromJson(JsonReader jsonReader) throws IOEx
                     prediction = PredictionContent.fromJson(reader);
                 } else if ("audio".equals(fieldName)) {
                     audio = AudioOutputParameters.fromJson(reader);
+                } else if ("service_tier".equals(fieldName)) {
+                    serviceTierOptions = ServiceTierOptions.fromString(reader.getString());
                 } else {
                     reader.skipChildren();
                 }
@@ -1102,6 +1131,7 @@ public static ChatCompletionsOptions fromJson(JsonReader jsonReader) throws IOEx
             deserializedChatCompletionsOptions.modalities = modalities;
             deserializedChatCompletionsOptions.prediction = prediction;
             deserializedChatCompletionsOptions.audio = audio;
+            deserializedChatCompletionsOptions.serviceTierOptions = serviceTierOptions;
             return deserializedChatCompletionsOptions;
         });
     }

sdk/openai/azure-ai-openai/src/main/java/com/azure/ai/openai/models/ServiceTier.java

@@ -0,0 +1,51 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+package com.azure.ai.openai.models;
+
+import com.azure.core.util.ExpandableStringEnum;
+
+import java.util.Collection;
+
+/**
+ * The service tier used for processing the request.
+ */
+public final class ServiceTier extends ExpandableStringEnum<ServiceTier> {
+
+    /**
+     * Service tier option for scale.
+     */
+    public static final ServiceTier SCALE = fromString("scale");
+
+    /**
+     * Service tier option for default.
+     */
+    public static final ServiceTier DEFAULT = fromString("default");
+
+    /**
+     * Creates a new instance of ServiceTier value.
+     *
+     * @deprecated Use the {@link #fromString(String)} factory method.
+     */
+    @Deprecated
+    public ServiceTier() {
+    }
+
+    /**
+     * Creates or finds a ServiceTier from its string representation.
+     *
+     * @param name a name to look for.
+     * @return the corresponding ServiceTier.
+     */
+    public static ServiceTier fromString(String name) {
+        return fromString(name, ServiceTier.class);
+    }
+
+    /**
+     * Get known ServiceTier values.
+     *
+     * @return a collection of known ServiceTier values.
+     */
+    public static Collection<ServiceTier> values() {
+        return values(ServiceTier.class);
+    }
+}

sdk/openai/azure-ai-openai/src/main/java/com/azure/ai/openai/models/ServiceTierOptions.java

@@ -0,0 +1,57 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+package com.azure.ai.openai.models;
+
+import com.azure.core.util.ExpandableStringEnum;
+
+import java.util.Collection;
+
+/**
+ * Specifies the latency tier to use for processing the request.
+ * This parameter is relevant for customers subscribed to the scale tier service:
+ *  - If set to 'auto', and the Project is Scale tier enabled, the system will utilize scale tier credits until they are exhausted.
+ *  - If set to 'auto', and the Project is not Scale tier enabled, the request will be processed using the default service tier with a lower uptime SLA and no latency guaranty.
+ *  - If set to 'default', the request will be processed using the default service tier with a lower uptime SLA and no latency guaranty.
+ *  - When not set, the default behavior is 'auto'.
+ * When this parameter is set, the response body will include the `service_tier` utilized.
+ */
+public final class ServiceTierOptions extends ExpandableStringEnum<ServiceTierOptions> {
+
+    /**
+     * Service tier option for auto.
+     */
+    public static final ServiceTierOptions AUTO = fromString("auto");
+
+    /**
+     * Service tier option for default.
+     */
+    public static final ServiceTierOptions DEFAULT = fromString("default");
+
+    /**
+     * Creates a new instance of ServiceTierOptions value.
+     *
+     * @deprecated Use the {@link #fromString(String)} factory method.
+     */
+    @Deprecated
+    public ServiceTierOptions() {
+    }
+
+    /**
+     * Creates or finds a ServiceTierOptions from its string representation.
+     *
+     * @param name a name to look for.
+     * @return the corresponding ServiceTierOptions.
+     */
+    public static ServiceTierOptions fromString(String name) {
+        return fromString(name, ServiceTierOptions.class);
+    }
+
+    /**
+     * Get known ServiceTierOptions values.
+     *
+     * @return a collection of known ServiceTierOptions values.
+     */
+    public static Collection<ServiceTierOptions> values() {
+        return values(ServiceTierOptions.class);
+    }
+}