@jambonz/llm 0.5.0 → 0.5.2

package/dist/test-kit/index.cjs

@@ -117,6 +117,9 @@ async function* streamFromOpenAI(client, req, options) {
     const param = options.tokensParam ?? (needsMaxCompletionTokens(req.model) ? "max_completion_tokens" : "max_tokens");
     body[param] = effectiveMaxTokens;
   }
+  const t0 = process.hrtime.bigint();
+  let t1;
+  let t2;
   let stream;
   let responseHeaders;
   try {
@@ -129,6 +132,7 @@ async function* streamFromOpenAI(client, req, options) {
     } else {
       stream = await apiPromise;
     }
+    t1 = process.hrtime.bigint();
   } catch (err) {
     if (isAbortError(err)) {
       yield { type: "end", finishReason: "aborted" };
@@ -160,6 +164,7 @@ async function* streamFromOpenAI(client, req, options) {
       const delta = choice.delta;
       const finishReason = choice.finish_reason;
       if (delta?.content) {
+        if (t2 === void 0) t2 = process.hrtime.bigint();
         yield { type: "token", text: delta.content };
       }
       if (delta?.tool_calls) {
@@ -229,6 +234,14 @@ async function* streamFromOpenAI(client, req, options) {
       endEvent.vendorMetadata = metadata;
     }
   }
+  if (t1 !== void 0) {
+    const t3 = process.hrtime.bigint();
+    endEvent.clientTiming = {
+      requestToHeadersMs: Number(t1 - t0) / 1e6,
+      ...t2 !== void 0 ? { headersToFirstTokenMs: Number(t2 - t1) / 1e6 } : {},
+      totalMs: Number(t3 - t0) / 1e6
+    };
+  }
   yield endEvent;
 }
 function appendOpenAIAssistantToolCall(history, toolCalls) {
@@ -2660,6 +2673,7 @@ var groqManifest = {
 var GROQ_METADATA_EXTRACTOR = makeMetadataExtractor([
   { header: "x-request-id", key: "request_id" },
   { header: "x-groq-region", key: "region" },
+  { header: "openai-processing-ms", key: "processing_ms", numeric: true },
   { header: "x-ratelimit-remaining-requests", key: "requests_remaining", numeric: true },
   { header: "x-ratelimit-remaining-tokens", key: "tokens_remaining", numeric: true }
 ]);