Primitive of the day · 2026-10-26

flat-schema-for-small-local-models

Shape JSON schemas to the local model's size: flatten to <=2 nesting levels (at 3+ levels small models return empty arrays at depth), wrap nested lists in a BaseModel instead of List[List[str]], split schema dicts over ~2KB into sequential extraction calls, and reserve deep/complex schemas for >=12B models (flat 2-level verified fine on 9.4B glm4).

When this applies

Designing the Pydantic/JSON schema for a structured-output step targeting a local model, especially one under 12B parameters.

Preconditions

• Target model is local (Ollama) and its parameter size is known
• You control the schema shape

Do this

Flatten to 2 levels, replace List[List[...]] with List[SomeModel], keep the schema dict under ~2KB or split into sequential calls, and escalate to a qwen3:32b-class model for genuinely deep schemas.

What you should see

Reliable, fully populated structured output from a 9-12B model.

How it fails if ignored

Empty arrays at nesting depth, context window consumed by the schema itself, unreliable nested-list fields — all schema-valid, so silently wrong.

Do not use when

Large frontier models via API; schemas that are already flat and small.

Kind: calibration. Part of the skill Ollama structured output pipeline step. Free to reuse in your own agent skills.