bucket for tokens out

llmproxymetrics.go

@@ -90,14 +90,21 @@ func handleJsonLine(line []byte) {
 		duration := jsonData["eval_duration"].(float64)
 		fmt.Printf("Duration: %.2f seconds\n", duration/1000000000.0)
 		opsProcessed.Inc()
+		tokens_out.Observe(jsonData["prompt_eval_count"].(float64))
 	}
 }
 
 var opsProcessed = promauto.NewCounter(prometheus.CounterOpts{
-	Name: "total_requests",
+	Name: "llmproxymetrics_total_requests",
 	Help: "The total number of processed events",
 })
 
+var tokens_out = prometheus.NewHistogram(prometheus.HistogramOpts{
+	Name:    "llmproxymetrics_tokens_out",
+	Help:    "Tokens generated.",
+	Buckets: prometheus.LinearBuckets(0, 100, 25),
+})
+
 func main() {
 	err := env.Parse(&cfg)
 	if err != nil {