@fabricorg/experiments-evals
v0.2.0
Published
LLM and code evaluators for Fabric Experiments: judge library, ModelClient contract, run aggregation.
Readme
@fabricorg/experiments-evals
Provider-neutral TypeScript evaluation contracts, built-in LLM judges, code evaluators, bounded-concurrency execution, hard judge-token reservations, and score aggregation for Fabric Experiments.
import { executeEvalRun, relevanceJudge } from '@fabricorg/experiments-evals';
import { DatabricksModelClient } from '@fabricorg/experiments-evals/databricks';Databricks is an explicit adapter subpath; it is not exported from the portable
package root. The subpath also exports a read-only DatabricksMlflowClient for
linking native managed-MLflow run metrics into Fabric release policy. MLflow
artifacts, datasets, traces, scorers, feedback, and lineage remain in
Databricks.