Update README.md

README.md

@@ -4,8 +4,8 @@ license_name: other
 license_link: https://ai.meta.com/llama/license
 ---
 Repairing the chat template for the model.
-There is a slight problem with the eorigianl llam 3.2 chat template. If you train a model with that current chat template and if the training script builds the prompts
-frmm a json file with the chat temaplte the model starts to output as its first token <|eot_id|> and naturally the script will then halt generation. 
+There is a slight problem with the original llama 3.1 3.2 chat template. If you train a model with that current chat template and if the training script builds the prompts
+from a json file with the chat tempalte the model starts to output as its first token <|eot_id|><|start_header_id|>assistant<|end_header_id|> and naturally the script will then halt generation. 
 the model learns to see this:
 
 <|begin_of_text|><|start_header_id|>user<|end_header_id|>