Create utils/quantize.py

utils/quantize.py

@@ -0,0 +1,8 @@
+import torch.quantization as tq
+
+def apply_int8(model):
+    model.qconfig = tq.get_default_qconfig("fbgemm")
+    tq.prepare(model, inplace=True)
+    # calibration pass must already be done
+    tq.convert(model, inplace=True)
+    return model