@arizeai/openinference-instrumentation-openai 4.4.1 → 4.4.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arizeai/openinference-instrumentation-openai",
-  "version": "4.4.1",
+  "version": "4.4.3",
   "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": "2.8.0",
-    "@arizeai/openinference-semantic-conventions": "2.14.0"
+    "@arizeai/openinference-core": "2.8.1",
+    "@arizeai/openinference-semantic-conventions": "2.15.0"
   },
   "devDependencies": {
     "@opentelemetry/exporter-trace-otlp-proto": "^0.50.0",

package/src/instrumentation.ts

@@ -485,6 +485,7 @@ export class OpenAIInstrumentation extends InstrumentationBase<typeof openai> {
               span.setAttributes({
                 // Do not record the output data as it can be large
                 ...getEmbeddingEmbeddingsAttributes(result),
+                ...getEmbeddingUsageAttributes(result),
               });
             }
             span.setStatus({ code: SpanStatusCode.OK });
@@ -555,7 +556,11 @@ export class OpenAIInstrumentation extends InstrumentationBase<typeof openai> {
                   ...getResponsesOutputMessagesAttributes(result),
                   ...getResponsesUsageAttributes(result),
                 });
-                span.setStatus({ code: SpanStatusCode.OK });
+                span.setStatus(
+                  result.status === "failed"
+                    ? { code: SpanStatusCode.ERROR, message: result.error?.message }
+                    : { code: SpanStatusCode.OK },
+                );
                 span.end();
               };
               if (isResponseCreateResponse(result)) {
@@ -981,6 +986,19 @@ function getEmbeddingEmbeddingsAttributes(response: CreateEmbeddingResponse): At
   }, {});
 }
 
+/**
+ * Get the token usage of an embedding response
+ */
+function getEmbeddingUsageAttributes(response: CreateEmbeddingResponse): Attributes {
+  if (!response.usage) {
+    return {};
+  }
+  return {
+    [SemanticConventions.LLM_TOKEN_COUNT_PROMPT]: response.usage.prompt_tokens,
+    [SemanticConventions.LLM_TOKEN_COUNT_TOTAL]: response.usage.total_tokens,
+  };
+}
+
 /**
  * Consumes the stream chunks and adds them to the span
  */

package/src/responsesAttributes.ts

@@ -255,7 +255,11 @@ export async function consumeResponseStreamEvents(
 
   for await (const event of stream) {
     switch (event.type) {
-      case "response.completed": {
+      // A stream can end on any of these; the incomplete and failed ones still carry the
+      // partial output and the usage.
+      case "response.completed":
+      case "response.incomplete":
+      case "response.failed": {
         response = event.response;
         break;
       }

package/src/version.ts

@@ -1,2 +1,2 @@
 // this is an auto-generated file. see js/scripts/version-update.js
-export const VERSION = "4.4.1";
+export const VERSION = "4.4.3";