Update README.md

README.md

@@ -280,6 +280,7 @@ To replicate the benchmark:
 from datasets import load_dataset
 ds = load_dataset("trl-lib/tldr", split="train").take(1000)
 ds.to_json("tldr_1000.json")
+```
 
 2. Start a vLLM server using your target model:
 ```bash