Create quantize_config.json

This is to make it work with engines like vLLM

quantize_config.json

@@ -0,0 +1,6 @@
+{
+  "bits": 4,
+  "desc_act": false,
+  "group_size": 128,
+  "true_sequential": true
+}