@arizeai/openinference-instrumentation-openai 4.2.6 → 4.3.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arizeai/openinference-instrumentation-openai",
-  "version": "4.2.6",
+  "version": "4.3.0",
   "private": false,
   "description": "OpenInference instrumentation for OpenAI",
   "keywords": [],
@@ -28,8 +28,8 @@
     "@opentelemetry/api": "^1.9.0",
     "@opentelemetry/core": "^2.8.0",
     "@opentelemetry/instrumentation": "^0.46.0",
-    "@arizeai/openinference-core": "2.7.0",
-    "@arizeai/openinference-semantic-conventions": "2.11.0"
+    "@arizeai/openinference-core": "2.7.2",
+    "@arizeai/openinference-semantic-conventions": "2.13.0"
   },
   "devDependencies": {
     "@opentelemetry/exporter-trace-otlp-proto": "^0.50.0",

package/src/instrumentation.ts

@@ -44,7 +44,7 @@ import {
   getResponsesOutputMessagesAttributes,
   getResponsesUsageAttributes,
 } from "./responsesAttributes";
-import { assertUnreachable, isString } from "./typeUtils";
+import { assertUnreachable, getNumberProperty, isString } from "./typeUtils";
 import { VERSION } from "./version";
 
 const MODULE_NAME = "openai";
@@ -81,6 +81,7 @@ export const HOST_SUFFIX_TO_PROVIDER: Record<string, LLMProvider> = {
   "api.minimax.io": LLMProvider.MINIMAX,
   "api.minimaxi.com": LLMProvider.MINIMAX,
   "api.minimax.chat": LLMProvider.MINIMAX,
+  "oci.oraclecloud.com": LLMProvider.ORACLE,
 };
 
 /**
@@ -788,7 +789,9 @@ function getCompletionInputValueAndMimeType(body: CompletionCreateParamsBase): A
 /**
  * Get usage attributes
  */
-function getUsageAttributes(completion: ChatCompletion | Completion): Attributes {
+function getUsageAttributes(
+  completion: ChatCompletion | Completion | ChatCompletionChunk,
+): Attributes {
   if (completion.usage) {
     const usageAttributes: Attributes = {
       [SemanticConventions.LLM_TOKEN_COUNT_COMPLETION]: completion.usage.completion_tokens,
@@ -796,6 +799,10 @@ function getUsageAttributes(completion: ChatCompletion | Completion): Attributes
       [SemanticConventions.LLM_TOKEN_COUNT_TOTAL]: completion.usage.total_tokens,
       [SemanticConventions.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_READ]:
         completion.usage.prompt_tokens_details?.cached_tokens,
+      [SemanticConventions.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_WRITE]: getNumberProperty(
+        completion.usage.prompt_tokens_details,
+        "cache_write_tokens",
+      ),
       [SemanticConventions.LLM_TOKEN_COUNT_PROMPT_DETAILS_AUDIO]:
         completion.usage.prompt_tokens_details?.audio_tokens,
       [SemanticConventions.LLM_TOKEN_COUNT_COMPLETION_DETAILS_AUDIO]:
@@ -938,6 +945,11 @@ async function consumeChatCompletionStreamChunks(stream: Stream<ChatCompletionCh
   let finishReason: string | undefined;
   // The first message is for the assistant response so we start at 1
   for await (const chunk of stream) {
+    // With `stream_options.include_usage`, the final chunk carries the request's
+    // usage and an empty `choices` array, so read it before the guard below.
+    if (chunk.usage) {
+      span.setAttributes(getUsageAttributes(chunk));
+    }
     if (chunk.choices.length <= 0) {
       continue;
     }

package/src/responsesAttributes.ts

@@ -11,6 +11,8 @@ import type { Stream } from "openai/streaming";
 import { safelyJSONStringify } from "@arizeai/openinference-core";
 import { SemanticConventions } from "@arizeai/openinference-semantic-conventions";
 
+import { getNumberProperty } from "./typeUtils";
+
 /**
  * Get attributes for responses api Items that are not typical messages with role
  * @param item - The item to get attributes for
@@ -217,6 +219,10 @@ export function getResponsesUsageAttributes(response: ResponseType): Attributes
       [SemanticConventions.LLM_TOKEN_COUNT_TOTAL]: response.usage.total_tokens,
       [SemanticConventions.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_READ]:
         response.usage.input_tokens_details?.cached_tokens,
+      [SemanticConventions.LLM_TOKEN_COUNT_PROMPT_DETAILS_CACHE_WRITE]: getNumberProperty(
+        response.usage.input_tokens_details,
+        "cache_write_tokens",
+      ),
       [SemanticConventions.LLM_TOKEN_COUNT_COMPLETION_DETAILS_REASONING]:
         response.usage.output_tokens_details?.reasoning_tokens,
       // no audio tokens for response inputs or outputs

package/src/typeUtils.ts

@@ -11,3 +11,14 @@ export function assertUnreachable(_: never): never {
 export function isString(value: unknown): value is string {
   return typeof value === "string";
 }
+
+/**
+ * Read a numeric property that the installed OpenAI SDK types may not declare yet
+ * (for example `prompt_tokens_details.cache_write_tokens`). Returns undefined when
+ * the object is missing or the property is not a number.
+ */
+export function getNumberProperty(obj: object | null | undefined, key: string): number | undefined {
+  if (obj == null) return undefined;
+  const value: unknown = Reflect.get(obj, key);
+  return typeof value === "number" ? value : undefined;
+}

package/src/version.ts

@@ -1,2 +1,2 @@
 // this is an auto-generated file. see js/scripts/version-update.js
-export const VERSION = "4.2.6";
+export const VERSION = "4.3.0";