| Model Soups (Wortsman et al. 2022) | MergeEngine | Weight averaging is the baseline merge operation |
| Task Arithmetic (Ilharco et al. 2022) | MergeEngine, math/task-arithmetic | Task vector addition and subtraction are core merge primitives |
| TIES-Merging (Yadav et al. 2023) | MergeEngine, math/ties-math | Three-step trimming, sign election, and disjoint merge reduce interference |
| DARE (Yu et al. 2023) | MergeEngine, math/dare-math | Random dropping and rescaling control merge capacity |
| SLERP (Shoemake 1985) | MergeEngine, math/slerp-math | Spherical interpolation preserves angular relationships between checkpoints |
| Evolutionary Model Merge (Akiba et al. 2024) | EvolutionEngine, CandidateGenerator | Genome encoding of merge recipes and NSGA-II search justify the evolution subsystem |
| NSGA-II (Deb et al. 2002) | multi-objective-optimization | Default Pareto-based selection for 2-4 fitness objectives |
| NSGA-III (Deb & Jain 2014) | multi-objective-optimization | Reference point selection for experiments with 5+ objectives |
| CMA-ES (Hansen 2001) | EvolutionEngine | Continuous hyperparameter search for merge ratios and scaling coefficients |
| LoRA (Hu et al. 2021) | lora, optional-finetuning | Low-rank adaptation produces task vectors for subsequent merging |
| QLoRA (Dettmers et al. 2023) | qlora | Resource-constrained fine-tuning enables task vector production on limited hardware |
| PEFT survey (Han et al. 2024) | peft | Strategy selection guidance for optional fine-tuning pipeline |
| Catastrophic Forgetting (McCloskey & Cohen 1989) | catastrophic-forgetting-controls | Fundamental problem that merge-time regression tests must detect |
| EWC (Kirkpatrick et al. 2017) | catastrophic-forgetting-controls | Parameter importance weighting as a candidate forgetting mitigation |
| LM Evaluation Harness (Gao et al. 2023) | EvaluationEngine | Task loading, prompt formatting, and metric computation patterns |
| HELM (Liang et al. 2022) | benchmark-specification | Multi-metric, multi-scenario evaluation design |
| Benchmark contamination (Sainz et al. 2023) | benchmark-integrity | Justifies strict split isolation and contamination detection |