Ladder Up, Memory Down: Low-Cost Fine-Tuning With Side Nets

TL;DR

This paper introduces Ladder Side Tuning (LST), a memory-efficient PEFT method for fine-tuning large language models that matches existing methods in accuracy while significantly reducing memory usage, enabling larger models to be fine-tuned on consumer hardware.

Contribution

The paper revisits and demonstrates the effectiveness of Ladder Side Tuning, a rarely explored PEFT technique, showing it reduces memory usage by 50% and scales similarly to QLoRA, with the addition of a new depth-extended variant called xLadder.

Findings

LST cuts peak memory by 50% compared to full fine-tuning.

LST achieves comparable accuracy to QLoRA across benchmarks.

xLadder enables deeper reasoning with no additional memory overhead.

Abstract

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.