Spectral Rewiring for Exploration, Purification, and Model Merging
About
Reinforcement learning has become a standard post-training recipe for large language models, but dense full-parameter updates create two deployment-relevant bottlenecks: suppressed reasoning performance, often reflected by premature saturation of test-time scaling, and interference when consolidating multiple capabilities through multi-domain training or model merging. We show that the reasoning-effective component of these updates is largely concentrated in the base model's spectral space, motivating Subspace-Aligned Rewiring (SAR), a post-hoc editing method that retains this spectral core while removing orthogonal components. SAR therefore preserves reasoning gains and filters residual update directions that suppress performance or amplify cross-domain interference. Across several model families and scales, SAR extracts compact reasoning cores using as little as approximately 0.58% of total parameters: it preserves over 99% of post-training performance and improves high-k exploration in mathematical reasoning, and generalizes to agentic coding by improving six of seven open benchmarks on an in-house model. SAR also purifies mixed-domain training updates by releasing suppressed coding capability while maintaining math reasoning and instruction following. It further enables model merging across experts, yielding cross-domain generalization that surpasses previous merging baselines and even the best single-domain experts. Overall, SAR shows that extracting reasoning-effective updates from parameter geometry can serve as a training-free mechanism to improve reasoning and multi-domain performance.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Math | AIME 24 | AVG@3274.38 | 16 | |
| Code | LCB | AVG@863.4 | 16 | |
| Mathematical Reasoning | AIME 2024 | AVG@3278.88 | 12 | |
| Agentic Coding | SWE-Bench | Score59 | 4 | |
| Agentic Coding | MSWE-Bench | Score31.3 | 4 | |
| Agentic Coding | SWE-Bench Pro | Score32.7 | 2 | |
| Agentic Coding | TerminalBench 2.0 | Score0.243 | 2 | |
| Agentic Coding | TerminalBench 1.0 | Score0.344 | 2 |