@@ -58,6 +58,9 @@ export class Benchmarks extends APIResource {
5858 * A benchmark resource for evaluating model performance.
5959 */
6060export interface Benchmark {
61+ /**
62+ * Identifier of the dataset to use for the benchmark evaluation.
63+ */
6164 dataset_id : string ;
6265
6366 /**
@@ -70,10 +73,13 @@ export interface Benchmark {
7073 */
7174 provider_id : string ;
7275
76+ /**
77+ * List of scoring function identifiers to apply during evaluation.
78+ */
7379 scoring_functions : Array < string > ;
7480
7581 /**
76- * Metadata for this evaluation task
82+ * Metadata for this evaluation task.
7783 */
7884 metadata ?: { [ key : string ] : unknown } ;
7985
@@ -82,26 +88,56 @@ export interface Benchmark {
8288 */
8389 provider_resource_id ?: string | null ;
8490
91+ /**
92+ * The resource type, always benchmark.
93+ */
8594 type ?: 'benchmark' ;
8695}
8796
97+ /**
98+ * Response containing a list of benchmark objects.
99+ */
88100export interface ListBenchmarksResponse {
101+ /**
102+ * List of benchmark objects.
103+ */
89104 data : BenchmarkListResponse ;
90105}
91106
107+ /**
108+ * List of benchmark objects.
109+ */
92110export type BenchmarkListResponse = Array < Benchmark > ;
93111
94112export interface BenchmarkRegisterParams {
113+ /**
114+ * The ID of the benchmark to register.
115+ */
95116 benchmark_id : string ;
96117
118+ /**
119+ * The ID of the dataset to use for the benchmark.
120+ */
97121 dataset_id : string ;
98122
123+ /**
124+ * The scoring functions to use for the benchmark.
125+ */
99126 scoring_functions : Array < string > ;
100127
128+ /**
129+ * The metadata to use for the benchmark.
130+ */
101131 metadata ?: { [ key : string ] : unknown } | null ;
102132
133+ /**
134+ * The ID of the provider benchmark to use for the benchmark.
135+ */
103136 provider_benchmark_id ?: string | null ;
104137
138+ /**
139+ * The ID of the provider to use for the benchmark.
140+ */
105141 provider_id ?: string | null ;
106142}
107143
0 commit comments