@@ -59,6 +59,9 @@ export class Benchmarks extends APIResource {
5959 * A benchmark resource for evaluating model performance.
6060 */
6161export interface Benchmark {
62+ /**
63+ * Identifier of the dataset to use for the benchmark evaluation.
64+ */
6265 dataset_id : string ;
6366
6467 /**
@@ -71,10 +74,13 @@ export interface Benchmark {
7174 */
7275 provider_id : string ;
7376
77+ /**
78+ * List of scoring function identifiers to apply during evaluation.
79+ */
7480 scoring_functions : Array < string > ;
7581
7682 /**
77- * Metadata for this evaluation task
83+ * Metadata for this evaluation task.
7884 */
7985 metadata ?: { [ key : string ] : unknown } ;
8086
@@ -83,26 +89,56 @@ export interface Benchmark {
8389 */
8490 provider_resource_id ?: string | null ;
8591
92+ /**
93+ * The resource type, always benchmark.
94+ */
8695 type ?: 'benchmark' ;
8796}
8897
98+ /**
99+ * Response containing a list of benchmark objects.
100+ */
89101export interface ListBenchmarksResponse {
102+ /**
103+ * List of benchmark objects.
104+ */
90105 data : BenchmarkListResponse ;
91106}
92107
108+ /**
109+ * List of benchmark objects.
110+ */
93111export type BenchmarkListResponse = Array < Benchmark > ;
94112
95113export interface BenchmarkRegisterParams {
114+ /**
115+ * The ID of the benchmark to register.
116+ */
96117 benchmark_id : string ;
97118
119+ /**
120+ * The ID of the dataset to use for the benchmark.
121+ */
98122 dataset_id : string ;
99123
124+ /**
125+ * The scoring functions to use for the benchmark.
126+ */
100127 scoring_functions : Array < string > ;
101128
129+ /**
130+ * The metadata to use for the benchmark.
131+ */
102132 metadata ?: { [ key : string ] : unknown } | null ;
103133
134+ /**
135+ * The ID of the provider benchmark to use for the benchmark.
136+ */
104137 provider_benchmark_id ?: string | null ;
105138
139+ /**
140+ * The ID of the provider to use for the benchmark.
141+ */
106142 provider_id ?: string | null ;
107143}
108144
0 commit comments