A manifesto for model merging
Bibliographic record
Abstract
If a modeling task is distributed, it will frequently be necessary to merge models developed by different team members. Existing approaches to model merging make assumptions about the types of model to be merged, and the nature of the relationship between them. This makes it hard to compare approaches. In this paper, we present a manifesto for research on model merging. We propose a framework for comparing different approaches to merging, by treating merge as an algebraic operator over models and model relationships. We specify the algebraic properties of an idealized merge operator, as well as related operators such as match, diff, split, and slice. We then show how our framework can be used to compare existing approaches by applying it to two of our own research projects on model merging. We show how this analysis permits a detailed comparison of approaches, reveals the key features of each, and identifies weaknesses that require further research. Most importantly, the framework emphasizes the need to make explicit all assumptions about the relationships between models, and indeed to treat model relationships as first class objects.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.058 | 0.048 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.007 | 0.017 |
| Scholarly communication | 0.015 | 0.035 |
| Open science | 0.004 | 0.015 |
| Research integrity | 0.005 | 0.009 |
| Insufficient payload (model declined to judge) | 0.006 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".