POME: Post Optimization Model Edit via Muon-style Projection
This provides a practical, zero-cost enhancement for any fine-tuning pipeline, applicable from 7B to 72B models, though it is incremental as it builds on existing fine-tuning methods.
The paper tackles the problem of enhancing fine-tuned large language models without extra data or optimization by introducing POME, a post-processing algorithm that uses muon-style projection on weight differences, resulting in performance boosts of +2.5% on GSM8K and +1.0% on code generation.
We introduce Post-Optimization Model Edit (POME), a new algorithm that enhances the performance of fine-tuned large language models using only their pretrained and fine-tuned checkpoints, without requiring extra data or further optimization. The core idea is to apply a muon-style projection to $ΔW$, the difference between the fine-tuned and pretrained weights. This projection uses truncated singular value decomposition (SVD) to equalize the influence of dominant update directions and prune small singular values, which often represent noise. As a simple post-processing step, POME is completely decoupled from the training pipeline. It requires zero modifications and imposes no overhead, making it universally compatible with any optimizer or distributed framework. POME delivers consistent gains, boosting average performance by +2.5\% on GSM8K and +1.0\% on code generation. Its broad applicability -- from 7B foundation models to 72B RLHF-instructed models -- establishes it as a practical, zero-cost enhancement for any fine-tuning pipeline. Code is available at https://github.com/NUS-HPC-AI-Lab/POME.