--model¶
Why we added this flag: vectorization must use the same model registry as prepare-datasets to ensure metadata alignment and schema validation.
What it does¶
- Accepts a Python import path (
package.module:ATTRIBUTE) resolving to aMapping[str, ModelSpec]. - Drives per-model token policies, metadata enrichment, and schema signature validation during indexing.
- Required for every
idxr vectorizecommand.
Typical usage¶
idxr vectorize index \
--model kb.std.ecc_6_0_ehp_7.registry:MODEL_REGISTRY \
--partition-manifest workdir/partitions/manifest.json \
--partition-out-dir workdir/chroma_partitions
Tips¶
- Keep a shared environment variable (
IDXR_MODEL) so all automation references the same registry string. - If you maintain multiple registries, note them in your manifest to avoid mixing incompatible partitions.