Update README.md

README.md

@@ -33,6 +33,13 @@ tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True,
                                          )
 ```
 
+The model produces a structured thinking section before answering, indicated by these special tokens:
+
+```python
+thinking_start = '<|thinking|>'
+thinking_end = '<|/thinking|>'
+```
+
 ## Inference example
 
 ###  Simple inference: