Update README.md

README.md

@@ -15,7 +15,7 @@ Doge uses `wsd_scheduler` as the training scheduler, which divides the learning
 
 Here are the initial learning rates required to continue training at each checkpoint:
 
-- **[Doge-40M](https://huggingface.co/SmallDoge/Doge-40M-checkpoint)**: 8e-3
+- **[Doge-40M](https://huggingface.co/SmallDoge/Doge-40M-checkpoint): 8e-3**
 - [Doge-40M-MoE](https://huggingface.co/SmallDoge/Doge-40M-MoE-checkpoint): 8e-3