Fusion Check: a zero-data law that predicts whether merging two fine-tunes preserves behavior

说明与用途

Fusion Check 能在不到一秒的时间内预测两份微调 LLM 是否可以安全融合，避免性能下降。

问题

Blind weight averaging (soup) can dilute model performance when fine‑tuned LLMs are trained on different tasks.

结果

Fusion Check accurately predicts fusion outcomes with 0.81 LOOCV correlation and 0.072 MAE across 13 task pairs, enabling safe model merging and avoiding performance dilution.

工作流

• 1. Install Fusion Check via pip.
• 2. Run fusion-check verificar <base> <m1> <m2> --acc1 <acc1> --acc2 <acc2> to compute cosine of weight deltas and effective dimensionality.
• 3. Read the decision output (MERGE/NO_MERGE/PRIORITIZED/REFUSE/VERIFY) and the optimal mixing coefficient α*.
• 4. If MERGE, apply the recommended α* to fuse the models; otherwise, follow the suggested action.

证据

Fusion Check — a law that tells you, before merging, whether two fine‑tuned LLMs are compatible.

来源

Proven three independent ways: Analytic derivation, Formal verification with z3 (28 theorems), Empirical validation (147 combinations, max deviation 1.13%, zero violations).

来源

Real result: Two weak, diverse estimators of the same task (0.762/0.756) → merged soup 0.822 (+6–7 points).

来源