Ladder Up, Memory Down: Low-Cost Fine-Tuning With Side Nets
This work addresses the memory constraints for researchers and practitioners fine-tuning LLMs on commodity hardware, offering a more efficient alternative to existing PEFT methods.
The paper tackles the memory bottleneck in fine-tuning large language models by revisiting Ladder Side Tuning (LST), which adds a lightweight side network to reduce peak memory usage by 50% compared to QLoRA while maintaining competitive performance across various benchmarks. It enables fine-tuning 7B-parameter models on a single 12 GB GPU with 2k-token contexts without gradient checkpointing, where QLoRA fails.
Fine-tuning large language models (LLMs) is often limited by the memory available on commodity GPUs. Parameter-efficient fine-tuning (PEFT) methods such as QLoRA reduce the number of trainable parameters, yet still incur high memory usage induced by the backward pass in the full model. We revisit Ladder Side Tuning (LST), a rarely explored PEFT technique that adds a lightweight side network, and show that it matches QLoRA's compute scaling slope while cutting peak memory by 50\%. Across different downstream benchmarks spanning natural language understanding, mathematical and LLM-critic tasks, LST has competitive performance with QLoRA's accuracy on average while being much more memory-efficient. This efficiency enables fine-tuning of 7B-parameter models on a single 12 GB consumer GPU with 2k-token contexts, requiring no gradient checkpointing\textemdash conditions under which QLoRA exhausts memory. Beyond memory efficiency, we also establish scaling laws showing that LST scales similarly to QLoRA. We exploit Ladder's architectural flexibility by introducing xLadder, a depth-extended variant that increases effective depth via cross-connections and shortens chain-of-thought (CoT) at fixed parameter count. Ladder is strong when memory is the bottleneck; xLadder builds on this by enabling deeper reasoning without additional memory overhead.