Update 'Recommended Inference Parameters' section

README.md

@@ -67,10 +67,10 @@ widget:
       <|im_start|>assistant
 inference:
   parameters:
-    add_special_tokens: true
     max_new_tokens: 250
     penalty_alpha: 0.5
-    top_k: 5
+    top_k: 4
+    repetition_penalty: 1.105
 ---
 
 # A Llama Chat Model of 101M Parameters
@@ -90,8 +90,6 @@ inference:
 
 ## Recommended Prompt Format
 
-The recommended prompt format is as follows:
-
 ```
 <|im_start|>system
 {system_message}<|im_end|>
@@ -102,10 +100,8 @@ The recommended prompt format is as follows:
 
 ## Recommended Inference Parameters
 
-To get the best results, add special tokens and use [contrastive search](https://huggingface.co/docs/transformers/main/en/generation_strategies#contrastive-search):
-
 ```yml
-add_special_tokens: true
 penalty_alpha: 0.5
-top_k: 5
+top_k: 4
+repetition_penalty: 1.105
 ```