```python for token_index in range( start_budget + padding_length + recent_budget, seq_length ): ``` When merging with CAM, if seq_length is long, such as 8k, and start_budget_ratio=0.1, then the ```for``` loop will be very long and inefficient.