From 1cedb28acf318fcf84337aa905f2fe7da78e9685 Mon Sep 17 00:00:00 2001
From: Lucas Armand <lucas@vast.ai>
Date: Wed, 8 Oct 2025 16:54:18 -0700
Subject: [PATCH 1/2] Removed division by elapsed time, since autoscaler
 cur_load in units of workload

---
 lib/metrics.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/lib/metrics.py b/lib/metrics.py
index 166706b..5152d73 100644
--- a/lib/metrics.py
+++ b/lib/metrics.py
@@ -78,10 +78,10 @@ class Metrics:
             elapsed = time.time() - self.last_metric_update
             if self.system_metrics.model_is_loaded is False and elapsed >= 10:
                 log.debug(f"sending loading model metrics after {int(elapsed)}s wait")
-                self.__send_metrics_and_reset(elapsed)
+                self.__send_metrics_and_reset()
             elif self.update_pending or elapsed > 10:
                 log.debug(f"sending loaded model metrics after {int(elapsed)}s wait")
-                self.__send_metrics_and_reset(elapsed)
+                self.__send_metrics_and_reset()
 
     def _model_loaded(self, max_throughput: float) -> None:
         self.system_metrics.model_loading_time = (
@@ -96,13 +96,13 @@ class Metrics:
 
     #######################################Private#######################################
 
-    def __send_metrics_and_reset(self, elapsed):
+    def __send_metrics_and_reset(self):
 
         def compute_autoscaler_data() -> AutoScalaerData:
             return AutoScalaerData(
                 id=self.id,
                 loadtime=(self.system_metrics.model_loading_time or 0.0),
-                cur_load=(self.model_metrics.workload_processing / elapsed),
+                cur_load=(self.model_metrics.workload_processing),
                 max_perf=self.model_metrics.max_throughput,
                 cur_perf=self.model_metrics.cur_perf,
                 error_msg=self.model_metrics.error_msg or "",

From 055e346c8c63513c5016403e9d25255754b02fdc Mon Sep 17 00:00:00 2001
From: Lucas Armand <lucas@vast.ai>
Date: Thu, 9 Oct 2025 10:13:50 -0700
Subject: [PATCH 2/2] Send metrics on request start

---
 lib/metrics.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/lib/metrics.py b/lib/metrics.py
index 166706b..8bce1ff 100644
--- a/lib/metrics.py
+++ b/lib/metrics.py
@@ -45,6 +45,7 @@ class Metrics:
         self.model_metrics.workload_received += workload
         self.model_metrics.requests_recieved.add(reqnum)
         self.model_metrics.requests_working.add(reqnum)
+        self.update_pending = True
 
     def _request_end(self, workload: float, reqnum: int) -> None:
         """