diff --git a/llm_server/routes/stats.py b/llm_server/routes/stats.py index f77486b..572bb30 100644 --- a/llm_server/routes/stats.py +++ b/llm_server/routes/stats.py @@ -3,6 +3,7 @@ import time from datetime import datetime from threading import Semaphore, Thread +from llm_server import opts from llm_server.integer import ThreadSafeInteger from llm_server.opts import concurrent_generates @@ -22,7 +23,7 @@ class SemaphoreCheckerThread(Thread): def run(self): global proompters_1_min while True: - self.values.append(self.semaphore._value) + self.values.append(opts.concurrent_generates - concurrent_semaphore._value) proompters_1_min = sum(self.values) / len(self.values) time.sleep(1)