Search papers, labs, and topics across Lattice.
This study investigates how supervised fine-tuning (SFT) influences instruction sensitivity in large language models, specifically analyzing Qwen3 models of varying scales (1.7B, 4B, and 8B) on the MS MARCO dataset. The findings reveal that while SFT significantly reduces instruction sensitivity by 54-71% for the smaller models, the effect diminishes at larger scales, with sensitivity changes at 8B being statistically indistinguishable from zero. Additionally, the robustness of SFT's impact varies across different models, highlighting the complexity of how fine-tuning interacts with task instruction variability.
SFT can dramatically reduce instruction sensitivity in smaller models, but its effectiveness diminishes in larger architectures, revealing a nuanced relationship between model size and fine-tuning outcomes.
Large language models can exhibit substantial performance variation across alternative formulations of the same task instruction, yet it remains unclear how conventional task-specific supervised fine-tuning (SFT) changes this instruction sensitivity. We study this question by evaluating fixed model checkpoints under multiple paraphrased instructions and defining instruction sensitivity as the standard deviation of task performance across them. We conduct a controlled scale analysis with Qwen3 models at 1.7B, 4B, and 8B on MS MARCO, together with targeted cross-family checks using Mistral-7B and Gemma-2-9B. Before SFT, instruction sensitivity decreases sharply with Qwen3 model scale. At 1.7B and 4B, SFT consistently reduces sensitivity across training instructions, with reductions of approximately 54--71%. At 8B, individual sensitivity changes are not statistically distinguishable from zero, but paired contrasts between training instructions are statistically reliable under query-level bootstrap analysis and have consistent directions across all three random seeds. Gemma-2-9B shows the same directional training-instruction contrast as Qwen3-8B, whereas Mistral-7B does not, suggesting that the strength of this effect also varies across models. Experiments on ESCI-English further show that free-generation and likelihood-based forced-choice evaluation can yield qualitatively different robustness conclusions even when valid-label generation is nearly perfect and average task performance is similar. Overall, SFT does not uniformly reduce instruction sensitivity: its robustness effect depends on the adaptation setting, while measured sensitivity can additionally depend on the prediction and scoring protocol.
