Update config.json

config.json

@@ -23,5 +23,16 @@
   "torch_dtype": "bfloat16",
   "transformers_version": "4.46.3",
   "use_cache": true,
-  "vocab_size": 32768
+  "vocab_size": 32768,
+  "quantization_config": {
+        "quant_method": "exl2",
+        "version": "0.2.4",
+        "bits": 3.0,
+        "head_bits": 6,
+        "calibration": {
+            "rows": 100,
+            "length": 2048,
+            "dataset": "(default)"
+        }
+    }
 }