Update README.md

README.md

@@ -28,7 +28,7 @@ DPO lets us directly train the model to score preferred responses higher than le
 - **Optimizer**: AdamW (learning rate = `2e-6`, weight decay = `0`)  
 - **Precision**: bf16  
 - **Batch size**: 2 (gradient accumulation = 4)  
-- **Scheduler**: Linear with 7% warmup  
+- **Scheduler**: cosine with 1% warmup  
 - **DPO Beta**: 0.1  
 - **Eval & Checkpointing**: Every epoch  
 - **Monitoring**: Weights & Biases (WandB)