Member Junction
    Preparing search index...

    What the engine sends to the sidecar's POST /trainer: one Training Run over one immutable Training Snapshot.

    interface TrainerRequest {
        candidates: readonly CandidateSpec[];
        incumbent: { artifact: PreAuthDownload; modelId: string } | null;
        maxRetries: number;
        protocol: EvaluationProtocolSpec;
        seed: number;
        snapshot: {
            contentHash: string;
            featureSchemaHash: string;
            id: string;
            matrix: PreAuthDownload;
            rowManifest: PreAuthDownload;
        };
        task: TaskType;
        threads: number;
        timeoutSeconds: number;
        trainingRunId: string;
    }
    Index

    Properties

    candidates: readonly CandidateSpec[]
    incumbent: { artifact: PreAuthDownload; modelId: string } | null
    maxRetries: number
    seed: number
    snapshot: {
        contentHash: string;
        featureSchemaHash: string;
        id: string;
        matrix: PreAuthDownload;
        rowManifest: PreAuthDownload;
    }

    Type Declaration

    • ReadonlycontentHash: string
    • ReadonlyfeatureSchemaHash: string

      The MJ: ML Training Snapshots row's FeatureSchemaHash; the sidecar copies it onto every returned artifact instead of recomputing it.

    • Readonlyid: string
    • Readonlymatrix: PreAuthDownload
    • ReadonlyrowManifest: PreAuthDownload
    task: TaskType
    threads: number
    timeoutSeconds: number
    trainingRunId: string