Dive into the computational demands of bio-sequence LLMs. We emphasize that while pre-training requires the most resources, both fine-tuning and prediction still necessitate significant computational power. Explores cost-effective solutions for accessing powerful GPUs.