fix stream metrics timing

llmproxymetrics.go

@@ -40,6 +40,7 @@ func createProxy(target *url.URL) func(http.ResponseWriter, *http.Request) {
 			req.URL.Scheme = target.Scheme
 			req.URL.Host = target.Host
 		}
+		wasstreamed := -1
 		data, err := io.ReadAll(r.Body)
 		if err == nil {
 			r.Body = io.NopCloser(bytes.NewReader(data))
@@ -48,9 +49,9 @@ func createProxy(target *url.URL) func(http.ResponseWriter, *http.Request) {
 
 			if jsonData["stream"] != nil {
 				if !jsonData["stream"].(bool) {
-					notStreamed.Inc()
+					wasstreamed = 0
 				} else {
-					streamed.Inc()
+					wasstreamed = 1
 				}
 			}
 		} else {
@@ -71,13 +72,13 @@ func createProxy(target *url.URL) func(http.ResponseWriter, *http.Request) {
 					line, err := reader.ReadBytes('\n')
 					if err != nil {
 						if err == io.EOF {
-							handleJsonLine([]byte(string(line)))
+							handleJsonLine([]byte(string(line)), wasstreamed)
 							pw.Write(line)
 							break
 						}
 						return
 					}
-					handleJsonLine(line)
+					handleJsonLine(line, wasstreamed)
 					pw.Write(line)
 				}
 
@@ -96,7 +97,7 @@ func createProxy(target *url.URL) func(http.ResponseWriter, *http.Request) {
 	}
 }
 
-func handleJsonLine(line []byte) {
+func handleJsonLine(line []byte, wasstreamed int) {
 	if len(line) == 0 {
 		return
 	}
@@ -115,6 +116,11 @@ func handleJsonLine(line []byte) {
 		tokens_in.Add(jsonData["prompt_eval_count"].(float64))
 		eval_time.Observe(duration)
 	}
+	if wasstreamed == 1 {
+		streamed.Inc()
+	} else if wasstreamed == 0 {
+		notStreamed.Inc()
+	}
 }
 
 var opsProcessed = promauto.NewCounter(prometheus.CounterOpts{