Do Post-Training Algorithms Actually Differ? A Controlled Study Across Model Scales Uncovers Scale-Dependent Ranking Inversions

TL;DR

This study presents a large-scale, controlled comparison of 51 post-training algorithms across different model sizes, revealing that algorithm rankings are highly scale-dependent and that model scale has the most significant impact on performance.

Contribution

Introduces OXRL, a unified framework for fair comparison of post-training algorithms, and provides the first large-scale analysis showing scale-dependent ranking inversions and task-specific algorithm leverage.

Findings

Algorithm rankings invert with model scale.

Loss function modifications have negligible effects.

Algorithm leverage varies significantly across tasks.

Abstract

Post-training alignment has produced dozens of competing algorithms -- DPO, SimPO, KTO, GRPO, and others -- yet practitioners lack controlled comparisons to guide algorithm selection. We present OXRL, a unified framework implementing 51 post-training algorithms with identical infrastructure, enabling the first large-scale apples-to-apples evaluation. Our study spans 8 algorithms across 4 model scales (0.5B--7B), 3 evaluation domains, and a 20-variant DPO taxonomy (100 runs at 1.5B, 5 seeds each), totaling $\sim$240 training runs on H100 GPUs. Three headline findings emerge. (1)~Algorithm rankings are unstable across scale: at 1.5B, online RL (SGRPO) tops all methods at 58.0\%~$\pm$0.57 on GSM8K; by 7B, the worst small-scale method (SimPO) becomes the best (85.8\%), a complete ranking inversion driven by model scale rather than LoRA regularization (confirmed via 2$\times$2 factorial). (2)~Loss function modifications yield negligible gains: none of 20 DPO variants significantly outperform vanilla DPO after Bonferroni correction; the sole significant outlier, SimPO, is worse ($-$11.5~pp, $p < 10^{-4}$). (3)~Algorithm leverage is task-specific: the 19.3~pp GSM8K spread collapses to 0.54~pp on MATH ($36\times$) and 0.47~pp on general-domain benchmarks ($41\times$), confirming that algorithm choice matters primarily within the training distribution. These findings yield a hierarchy of leverage for practitioners: model scale (${\sim}$50~pp) $\gg$ training paradigm (${\sim}$10~pp) $\gg$ online vs.\ offline (${\sim}$9~pp) $\gg$ loss function (${\sim}$1~pp). We release all code, configs, and evaluation data as a living community benchmark.