Remove sampling and set temperature to zero.

generation_config.json

@@ -1,12 +1,12 @@
 {
   "bos_token_id": 128000,
-  "do_sample": true,
+  "do_sample": false,
   "eos_token_id": [
     128001,
     128009
   ],
   "max_length": 4096,
-  "temperature": 0.6,
-  "top_p": 0.9,
+  "temperature": 0,
+  "top_p": 1,
   "transformers_version": "4.40.0"
 }