braintrust 0.0.144 → 0.0.146
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/braintrust-0.0.141.tgz +0 -0
- package/dist/cli.js +25 -6
- package/dist/index.d.mts +1 -1
- package/dist/index.d.ts +1 -1
- package/dist/index.js +16 -2
- package/dist/index.mjs +16 -2
- package/package.json +2 -2
|
Binary file
|
package/dist/cli.js
CHANGED
|
@@ -1232,7 +1232,7 @@ var require_package = __commonJS({
|
|
|
1232
1232
|
"package.json"(exports2, module2) {
|
|
1233
1233
|
module2.exports = {
|
|
1234
1234
|
name: "braintrust",
|
|
1235
|
-
version: "0.0.
|
|
1235
|
+
version: "0.0.146",
|
|
1236
1236
|
description: "SDK for integrating Braintrust",
|
|
1237
1237
|
repository: {
|
|
1238
1238
|
type: "git",
|
|
@@ -1302,7 +1302,7 @@ var require_package = __commonJS({
|
|
|
1302
1302
|
},
|
|
1303
1303
|
dependencies: {
|
|
1304
1304
|
"@ai-sdk/provider": "^0.0.11",
|
|
1305
|
-
"@braintrust/core": "0.0.
|
|
1305
|
+
"@braintrust/core": "0.0.47",
|
|
1306
1306
|
"@next/env": "^14.2.3",
|
|
1307
1307
|
"@vercel/functions": "^1.0.2",
|
|
1308
1308
|
ai: "^3.2.16",
|
|
@@ -4872,6 +4872,17 @@ function initExperiment(state, options = {}) {
|
|
|
4872
4872
|
setCurrent: false
|
|
4873
4873
|
});
|
|
4874
4874
|
}
|
|
4875
|
+
function callEvaluatorData(data) {
|
|
4876
|
+
let dataResult = typeof data === "function" ? data() : data;
|
|
4877
|
+
let baseExperiment = void 0;
|
|
4878
|
+
if ("_type" in dataResult && dataResult._type === "BaseExperiment") {
|
|
4879
|
+
baseExperiment = dataResult.name;
|
|
4880
|
+
}
|
|
4881
|
+
return {
|
|
4882
|
+
data: dataResult,
|
|
4883
|
+
baseExperiment
|
|
4884
|
+
};
|
|
4885
|
+
}
|
|
4875
4886
|
globalThis._evals = {
|
|
4876
4887
|
evaluators: {},
|
|
4877
4888
|
reporters: {}
|
|
@@ -5705,10 +5716,11 @@ function evaluateBuildResults(inFile, buildResult) {
|
|
|
5705
5716
|
return { ...globalThis._evals };
|
|
5706
5717
|
});
|
|
5707
5718
|
}
|
|
5708
|
-
async function initLogger(projectName, experimentName, metadata) {
|
|
5719
|
+
async function initLogger(projectName, experimentName, metadata, baseExperiment) {
|
|
5709
5720
|
const logger = init(projectName, {
|
|
5710
5721
|
experiment: experimentName,
|
|
5711
|
-
metadata
|
|
5722
|
+
metadata,
|
|
5723
|
+
baseExperiment
|
|
5712
5724
|
});
|
|
5713
5725
|
const info = await logger.summarize({ summarizeScores: false });
|
|
5714
5726
|
console.error(
|
|
@@ -5952,15 +5964,22 @@ async function runOnce(handles, opts) {
|
|
|
5952
5964
|
}
|
|
5953
5965
|
const experimentIdToEvaluator = {};
|
|
5954
5966
|
const resultPromises = evaluators.evaluators.map(async (evaluator) => {
|
|
5967
|
+
const { data, baseExperiment } = callEvaluatorData(
|
|
5968
|
+
evaluator.evaluator.data
|
|
5969
|
+
);
|
|
5955
5970
|
const logger = opts.noSendLogs ? null : await initLogger(
|
|
5956
5971
|
evaluator.evaluator.projectName,
|
|
5957
5972
|
evaluator.evaluator.experimentName,
|
|
5958
|
-
evaluator.evaluator.metadata
|
|
5973
|
+
evaluator.evaluator.metadata,
|
|
5974
|
+
baseExperiment
|
|
5959
5975
|
);
|
|
5960
5976
|
try {
|
|
5961
5977
|
return await runEvaluator(
|
|
5962
5978
|
logger,
|
|
5963
|
-
|
|
5979
|
+
{
|
|
5980
|
+
...evaluator.evaluator,
|
|
5981
|
+
data
|
|
5982
|
+
},
|
|
5964
5983
|
opts.progressReporter,
|
|
5965
5984
|
opts.filters
|
|
5966
5985
|
);
|
package/dist/index.d.mts
CHANGED
|
@@ -1200,7 +1200,7 @@ type BaseExperiment<Input, Expected, Metadata extends BaseMetadata = DefaultMeta
|
|
|
1200
1200
|
declare function BaseExperiment<Input = unknown, Expected = unknown, Metadata extends BaseMetadata = DefaultMetadataType>(options?: {
|
|
1201
1201
|
name?: string;
|
|
1202
1202
|
}): BaseExperiment<Input, Expected, Metadata>;
|
|
1203
|
-
type EvalData<Input, Expected, Metadata extends BaseMetadata = DefaultMetadataType> = EvalCase<Input, Expected, Metadata>[] | (() => EvalCase<Input, Expected, Metadata>[]) | (() => Promise<EvalCase<Input, Expected, Metadata>[]>) | AsyncGenerator<EvalCase<Input, Expected, Metadata>> | AsyncIterable<EvalCase<Input, Expected, Metadata>> | BaseExperiment<Input, Expected, Metadata> | (() => BaseExperiment<Input, Expected, Metadata>);
|
|
1203
|
+
type EvalData<Input, Expected, Metadata extends BaseMetadata = DefaultMetadataType> = EvalCase<Input, Expected, Metadata>[] | (() => EvalCase<Input, Expected, Metadata>[]) | Promise<EvalCase<Input, Expected, Metadata>[]> | (() => Promise<EvalCase<Input, Expected, Metadata>[]>) | AsyncGenerator<EvalCase<Input, Expected, Metadata>> | AsyncIterable<EvalCase<Input, Expected, Metadata>> | BaseExperiment<Input, Expected, Metadata> | (() => BaseExperiment<Input, Expected, Metadata>);
|
|
1204
1204
|
type EvalTask<Input, Output> = ((input: Input, hooks: EvalHooks) => Promise<Output>) | ((input: Input, hooks: EvalHooks) => Output);
|
|
1205
1205
|
interface EvalHooks {
|
|
1206
1206
|
meta: (info: Record<string, unknown>) => void;
|
package/dist/index.d.ts
CHANGED
|
@@ -1200,7 +1200,7 @@ type BaseExperiment<Input, Expected, Metadata extends BaseMetadata = DefaultMeta
|
|
|
1200
1200
|
declare function BaseExperiment<Input = unknown, Expected = unknown, Metadata extends BaseMetadata = DefaultMetadataType>(options?: {
|
|
1201
1201
|
name?: string;
|
|
1202
1202
|
}): BaseExperiment<Input, Expected, Metadata>;
|
|
1203
|
-
type EvalData<Input, Expected, Metadata extends BaseMetadata = DefaultMetadataType> = EvalCase<Input, Expected, Metadata>[] | (() => EvalCase<Input, Expected, Metadata>[]) | (() => Promise<EvalCase<Input, Expected, Metadata>[]>) | AsyncGenerator<EvalCase<Input, Expected, Metadata>> | AsyncIterable<EvalCase<Input, Expected, Metadata>> | BaseExperiment<Input, Expected, Metadata> | (() => BaseExperiment<Input, Expected, Metadata>);
|
|
1203
|
+
type EvalData<Input, Expected, Metadata extends BaseMetadata = DefaultMetadataType> = EvalCase<Input, Expected, Metadata>[] | (() => EvalCase<Input, Expected, Metadata>[]) | Promise<EvalCase<Input, Expected, Metadata>[]> | (() => Promise<EvalCase<Input, Expected, Metadata>[]>) | AsyncGenerator<EvalCase<Input, Expected, Metadata>> | AsyncIterable<EvalCase<Input, Expected, Metadata>> | BaseExperiment<Input, Expected, Metadata> | (() => BaseExperiment<Input, Expected, Metadata>);
|
|
1204
1204
|
type EvalTask<Input, Output> = ((input: Input, hooks: EvalHooks) => Promise<Output>) | ((input: Input, hooks: EvalHooks) => Output);
|
|
1205
1205
|
interface EvalHooks {
|
|
1206
1206
|
meta: (info: Record<string, unknown>) => void;
|
package/dist/index.js
CHANGED
|
@@ -4311,6 +4311,17 @@ function initExperiment2(state, options = {}) {
|
|
|
4311
4311
|
setCurrent: false
|
|
4312
4312
|
});
|
|
4313
4313
|
}
|
|
4314
|
+
function callEvaluatorData(data) {
|
|
4315
|
+
let dataResult = typeof data === "function" ? data() : data;
|
|
4316
|
+
let baseExperiment = void 0;
|
|
4317
|
+
if ("_type" in dataResult && dataResult._type === "BaseExperiment") {
|
|
4318
|
+
baseExperiment = dataResult.name;
|
|
4319
|
+
}
|
|
4320
|
+
return {
|
|
4321
|
+
data: dataResult,
|
|
4322
|
+
baseExperiment
|
|
4323
|
+
};
|
|
4324
|
+
}
|
|
4314
4325
|
globalThis._evals = {
|
|
4315
4326
|
evaluators: {},
|
|
4316
4327
|
reporters: {}
|
|
@@ -4345,12 +4356,14 @@ async function Eval(name, evaluator, reporterOrOpts) {
|
|
|
4345
4356
|
}
|
|
4346
4357
|
const resolvedReporter = options.reporter || defaultReporter;
|
|
4347
4358
|
try {
|
|
4359
|
+
const { data, baseExperiment } = callEvaluatorData(evaluator.data);
|
|
4348
4360
|
const experiment = initExperiment2(evaluator.state, {
|
|
4349
4361
|
...evaluator.projectId ? { projectId: evaluator.projectId } : { project: name },
|
|
4350
4362
|
experiment: evaluator.experimentName,
|
|
4351
4363
|
metadata: evaluator.metadata,
|
|
4352
4364
|
isPublic: evaluator.isPublic,
|
|
4353
|
-
update: evaluator.update
|
|
4365
|
+
update: evaluator.update,
|
|
4366
|
+
baseExperiment
|
|
4354
4367
|
});
|
|
4355
4368
|
if (options.onStart) {
|
|
4356
4369
|
experiment.summarize({ summarizeScores: false }).then(options.onStart);
|
|
@@ -4359,7 +4372,8 @@ async function Eval(name, evaluator, reporterOrOpts) {
|
|
|
4359
4372
|
const evalDef = {
|
|
4360
4373
|
evalName,
|
|
4361
4374
|
projectName: name,
|
|
4362
|
-
...evaluator
|
|
4375
|
+
...evaluator,
|
|
4376
|
+
data
|
|
4363
4377
|
};
|
|
4364
4378
|
const ret = await runEvaluator(experiment, evalDef, progressReporter, []);
|
|
4365
4379
|
progressReporter.stop();
|
package/dist/index.mjs
CHANGED
|
@@ -4254,6 +4254,17 @@ function initExperiment2(state, options = {}) {
|
|
|
4254
4254
|
setCurrent: false
|
|
4255
4255
|
});
|
|
4256
4256
|
}
|
|
4257
|
+
function callEvaluatorData(data) {
|
|
4258
|
+
let dataResult = typeof data === "function" ? data() : data;
|
|
4259
|
+
let baseExperiment = void 0;
|
|
4260
|
+
if ("_type" in dataResult && dataResult._type === "BaseExperiment") {
|
|
4261
|
+
baseExperiment = dataResult.name;
|
|
4262
|
+
}
|
|
4263
|
+
return {
|
|
4264
|
+
data: dataResult,
|
|
4265
|
+
baseExperiment
|
|
4266
|
+
};
|
|
4267
|
+
}
|
|
4257
4268
|
globalThis._evals = {
|
|
4258
4269
|
evaluators: {},
|
|
4259
4270
|
reporters: {}
|
|
@@ -4288,12 +4299,14 @@ async function Eval(name, evaluator, reporterOrOpts) {
|
|
|
4288
4299
|
}
|
|
4289
4300
|
const resolvedReporter = options.reporter || defaultReporter;
|
|
4290
4301
|
try {
|
|
4302
|
+
const { data, baseExperiment } = callEvaluatorData(evaluator.data);
|
|
4291
4303
|
const experiment = initExperiment2(evaluator.state, {
|
|
4292
4304
|
...evaluator.projectId ? { projectId: evaluator.projectId } : { project: name },
|
|
4293
4305
|
experiment: evaluator.experimentName,
|
|
4294
4306
|
metadata: evaluator.metadata,
|
|
4295
4307
|
isPublic: evaluator.isPublic,
|
|
4296
|
-
update: evaluator.update
|
|
4308
|
+
update: evaluator.update,
|
|
4309
|
+
baseExperiment
|
|
4297
4310
|
});
|
|
4298
4311
|
if (options.onStart) {
|
|
4299
4312
|
experiment.summarize({ summarizeScores: false }).then(options.onStart);
|
|
@@ -4302,7 +4315,8 @@ async function Eval(name, evaluator, reporterOrOpts) {
|
|
|
4302
4315
|
const evalDef = {
|
|
4303
4316
|
evalName,
|
|
4304
4317
|
projectName: name,
|
|
4305
|
-
...evaluator
|
|
4318
|
+
...evaluator,
|
|
4319
|
+
data
|
|
4306
4320
|
};
|
|
4307
4321
|
const ret = await runEvaluator(experiment, evalDef, progressReporter, []);
|
|
4308
4322
|
progressReporter.stop();
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "braintrust",
|
|
3
|
-
"version": "0.0.
|
|
3
|
+
"version": "0.0.146",
|
|
4
4
|
"description": "SDK for integrating Braintrust",
|
|
5
5
|
"repository": {
|
|
6
6
|
"type": "git",
|
|
@@ -70,7 +70,7 @@
|
|
|
70
70
|
},
|
|
71
71
|
"dependencies": {
|
|
72
72
|
"@ai-sdk/provider": "^0.0.11",
|
|
73
|
-
"@braintrust/core": "0.0.
|
|
73
|
+
"@braintrust/core": "0.0.47",
|
|
74
74
|
"@next/env": "^14.2.3",
|
|
75
75
|
"@vercel/functions": "^1.0.2",
|
|
76
76
|
"ai": "^3.2.16",
|