import

To import model evaluation results, invoke the /import command with a model name to consolidate results from multiple folders into a single query-friendly evals.json file. This structured workflow ensures provenance, validation, and safe, non-destructive imports.

Consolidating evaluation data into cohorts requires importing new model variants, quantizations, and evaluation data into r12 or tables300 datasets. The tool validates and organizes results, generating per-variant import entries with precise globs for exact model/template/sampler combinations.

Yes, workflow automation is supported for discovering available results, determining their suites and quantizations, and selecting the correct cohort. This automated process safely guides users through structured steps to ensure non-destructive imports into the ReasonScape dataset structure.

Mapping exact combinations is handled by generating per-variant import entries with precise glob patterns. These globs accurately map to specific model/template/sampler combinations, ensuring correct consolidation and validation of results into the r12 or tables300 datasets.

No, importing new evaluation data is a safe, non-destructive process. The workflow ensures data provenance and validation, consolidating results into cohort-based evals without overwriting existing records in your ReasonScape datasets.

The supported datasets for importing model performance are the ReasonScape r12 and tables300 datasets. Importing into these structures enables centralized tracking, comparison, and governance of model performance across various evaluation suites.