Add stream_options param; add usage in ChatCompletionChunk

openapi.yaml

@@ -7212,6 +7212,8 @@ components:
           type: boolean
           nullable: true
           default: false
+        stream_options:
+          $ref: "#/components/schemas/ChatCompletionStreamOptions"
         suffix:
           description: |
             The suffix that comes after a completion of inserted text.
@@ -7697,6 +7699,18 @@ components:
         - tool
         - function
 
+    ChatCompletionStreamOptions:
+      description: Options for stream response
+      type: object
+      nullable: true
+      default: null
+      properties:
+        include_usage:
+          type: boolean
+          description: |
+            If set, an extra chunk will be returned before the `data: [Done]` message with a `usage` field that shows usage data for all chunks in the streamed request.
+            All other chunks will also include a `usage` field but with a null value. In the case of an error, the extra (usage) chunk might be missing.
+
     ChatCompletionResponseMessage:
       type: object
       description: A chat completion message generated by the model.
@@ -7891,6 +7905,8 @@ components:
           type: boolean
           nullable: true
           default: false
+        stream_options:
+          $ref: "#/components/schemas/ChatCompletionStreamOptions"
         temperature:
           type: number
           minimum: 0
@@ -8191,6 +8207,8 @@ components:
           type: string
           description: The object type, which is always `chat.completion.chunk`.
           enum: [chat.completion.chunk]
+        usage:
+          $ref: "#/components/schemas/CompletionUsage"
       required:
         - choices
         - created