What it teaches
A merge recipe is a candidate genome. It names parents, compatibility assumptions, operators, coefficients, evaluation scope, and intended niche. The sketchpad turns merging from an ad hoc experiment into a reusable lineage record.
Sample input
MERGE_RECIPE legal-code-helper-v1
parents: [base-instruct, legal-summary-adapter, code-explainer-adapter]
operator: adapter_weighted_average
coefficients: [0.60, 0.25, 0.15]
eval_scope: citation_explanation, code_snippet_explanation, latency_local
created_at_utc: 2026-06-29T00:00:00Z
ENDInterpretation
A recipe should be compared against its parents and against no-op. If it adds useful capability and remains maintainable, it can become a specialist or challenger. If it does not add enough value, archive it with notes so future searches learn from it.
Next step
Read Evolutionary Model Merging and Model Merging Operators.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.
Architecture and model mergingModel Breeder Architecture Deep DiveSource report · 76.7 KB
Model merging and adaptersModel Merging: Definitions, Taxonomy, Algorithms, Tooling, and Production Pipeline DesignEmerging practice · 35.7 KB
Model merging and adaptersTiny-LLM Adapters and LoRA: Breeding Operators, Compatibility, Evaluation, and PromotionEmerging implementation practice · 15.3 KB