back to 2

app.py

@@ -346,7 +346,7 @@ def load_image( infilename ) :
 def throttled_inference(text):
     global inferences_running
     current = inferences_running
-    if current >= 3:
+    if current >= 2:
         print(f"Rejected inference when we already had {current} running")
         return load_image("./gpu.jpg")