Collect garbage after quantization to reduce peak RAM

app/ml/model_loader.py

@@ -1,3 +1,5 @@
+import gc
+
 import torch
 from detoxify import Detoxify
 from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
@@ -33,6 +35,7 @@ def load_system():
                 dtype=torch.qint8,
             )
             model_o.eval()
+            gc.collect()
         except Exception:
             pass
 
@@ -55,6 +58,7 @@ def load_system():
                 {torch.nn.Linear},
                 dtype=torch.qint8,
             )
+            gc.collect()
         except Exception:
             pass
         try:
@@ -63,6 +67,7 @@ def load_system():
                 {torch.nn.Linear},
                 dtype=torch.qint8,
             )
+            gc.collect()
         except Exception:
             pass