Topology-Enhanced Alignment for Large Language Models: Trajectory Topology Loss and Topological Preference Optimization
About
Alignment of large language models (LLMs) via SFT and RLHF/DPO typically ignores the global geometry of the representation space, relying instead on local token likelihoods or scalar scores. We view generation as tracing a semantic trajectory in hidden space and propose a topology-enhanced alignment framework that regularizes these trajectories using 0-dimensional persistent homology. First, for SFT, we introduce Trajectory Topology Loss (TTL). Treating prompt and gold-answer embeddings as a mixed point cloud, we use a 0D persistent homology algorithm to extract "prompt-answer bridges." TTL aligns the model's actual update direction with these topological bridges rather than arbitrary directions. Second, for DPO, we propose Topological Preference Optimization (TPO). TPO constructs topic-specific semantic preference vectors and aligns the improvement direction between rejected and chosen responses with these vectors in an intermediate hidden layer. We also introduce a dynamic weighting scheme to balance DPO and TPO losses. Evaluating on Qwen2.5-7B-Instruct using UltraChat and Anthropic HH-RLHF, our topology-enhanced objectives consistently outperform strong non-topological baselines (e.g., per-example, nearest-neighbor, random regularizers) on automatic preference metrics and LLM-judge evaluations, while maintaining or improving toxicity. Results show persistent homology and trajectory geometry offer a promising direction for controllable alignment.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Instruction Following | IFEval | IFEval Accuracy71.8 | 836 | |
| Instruction Following | AlpacaEval 2.0 | Win Rate55.6 | 722 | |
| Multi-turn Dialogue Evaluation | MT-Bench | Overall Score8.88 | 532 | |
| Instruction Following | AlpacaEval | Win Rate56.5 | 420 | |
| Reward Modeling | RewardBench | -- | 216 | |
| Multi-turn dialogue | MT-Bench | MT-Bench Score8.81 | 126 | |
| Harmlessness evaluation | HH-RLHF | Harmlessness Rate94.5 | 6 | |
| Reward Modeling Evaluation | RewardBench | R-Bench Score88.1 | 3 | |
| Alignment Evaluation | HH-RLHF (test) | Reward Model Score65.4 | 2 | |
| Instruction Following | UltraChat | RM Score67.8 | 2 |