ai-functions 2.1.3 → 2.4.0
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/.turbo/turbo-build.log +1 -1
- package/CHANGELOG.md +90 -1
- package/README.md +38 -0
- package/dist/ai-promise.d.ts +3 -3
- package/dist/ai-promise.d.ts.map +1 -1
- package/dist/ai-promise.js +135 -64
- package/dist/ai-promise.js.map +1 -1
- package/dist/ai-schemas.d.ts +56 -0
- package/dist/ai-schemas.d.ts.map +1 -0
- package/dist/ai-schemas.js +53 -0
- package/dist/ai-schemas.js.map +1 -0
- package/dist/ai.d.ts +16 -242
- package/dist/ai.d.ts.map +1 -1
- package/dist/ai.js +51 -858
- package/dist/ai.js.map +1 -1
- package/dist/batch/anthropic.d.ts +6 -4
- package/dist/batch/anthropic.d.ts.map +1 -1
- package/dist/batch/anthropic.js +83 -145
- package/dist/batch/anthropic.js.map +1 -1
- package/dist/batch/bedrock.d.ts +8 -30
- package/dist/batch/bedrock.d.ts.map +1 -1
- package/dist/batch/bedrock.js +155 -338
- package/dist/batch/bedrock.js.map +1 -1
- package/dist/batch/cloudflare.d.ts +8 -20
- package/dist/batch/cloudflare.d.ts.map +1 -1
- package/dist/batch/cloudflare.js +68 -189
- package/dist/batch/cloudflare.js.map +1 -1
- package/dist/batch/google.d.ts +6 -20
- package/dist/batch/google.d.ts.map +1 -1
- package/dist/batch/google.js +70 -238
- package/dist/batch/google.js.map +1 -1
- package/dist/batch/index.d.ts +4 -1
- package/dist/batch/index.d.ts.map +1 -1
- package/dist/batch/index.js +4 -1
- package/dist/batch/index.js.map +1 -1
- package/dist/batch/memory.d.ts +1 -1
- package/dist/batch/memory.d.ts.map +1 -1
- package/dist/batch/memory.js +14 -10
- package/dist/batch/memory.js.map +1 -1
- package/dist/batch/openai.d.ts +11 -14
- package/dist/batch/openai.d.ts.map +1 -1
- package/dist/batch/openai.js +52 -156
- package/dist/batch/openai.js.map +1 -1
- package/dist/batch/provider.d.ts +111 -0
- package/dist/batch/provider.d.ts.map +1 -0
- package/dist/batch/provider.js +233 -0
- package/dist/batch/provider.js.map +1 -0
- package/dist/batch-map.d.ts.map +1 -1
- package/dist/batch-map.js +23 -17
- package/dist/batch-map.js.map +1 -1
- package/dist/batch-queue.d.ts +65 -0
- package/dist/batch-queue.d.ts.map +1 -1
- package/dist/batch-queue.js +169 -14
- package/dist/batch-queue.js.map +1 -1
- package/dist/budget.d.ts.map +1 -1
- package/dist/budget.js +27 -14
- package/dist/budget.js.map +1 -1
- package/dist/cache.d.ts +23 -0
- package/dist/cache.d.ts.map +1 -1
- package/dist/cache.js +36 -15
- package/dist/cache.js.map +1 -1
- package/dist/context.d.ts +26 -8
- package/dist/context.d.ts.map +1 -1
- package/dist/context.js +64 -62
- package/dist/context.js.map +1 -1
- package/dist/digital-objects-registry.d.ts +229 -0
- package/dist/digital-objects-registry.d.ts.map +1 -0
- package/dist/digital-objects-registry.js +617 -0
- package/dist/digital-objects-registry.js.map +1 -0
- package/dist/embeddings.d.ts +2 -2
- package/dist/embeddings.d.ts.map +1 -1
- package/dist/errors.d.ts +22 -0
- package/dist/errors.d.ts.map +1 -0
- package/dist/errors.js +35 -0
- package/dist/errors.js.map +1 -0
- package/dist/eval/runner.d.ts +8 -0
- package/dist/eval/runner.d.ts.map +1 -1
- package/dist/eval/runner.js +41 -35
- package/dist/eval/runner.js.map +1 -1
- package/dist/eval-log/in-memory.d.ts +34 -0
- package/dist/eval-log/in-memory.d.ts.map +1 -0
- package/dist/eval-log/in-memory.js +84 -0
- package/dist/eval-log/in-memory.js.map +1 -0
- package/dist/eval-log/index.d.ts +29 -0
- package/dist/eval-log/index.d.ts.map +1 -0
- package/dist/eval-log/index.js +39 -0
- package/dist/eval-log/index.js.map +1 -0
- package/dist/eval-log/types.d.ts +101 -0
- package/dist/eval-log/types.d.ts.map +1 -0
- package/dist/eval-log/types.js +16 -0
- package/dist/eval-log/types.js.map +1 -0
- package/dist/function-registry.d.ts +176 -0
- package/dist/function-registry.d.ts.map +1 -0
- package/dist/function-registry.js +685 -0
- package/dist/function-registry.js.map +1 -0
- package/dist/generate.d.ts +9 -3
- package/dist/generate.d.ts.map +1 -1
- package/dist/generate.js +18 -18
- package/dist/generate.js.map +1 -1
- package/dist/index.d.ts +18 -11
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +35 -18
- package/dist/index.js.map +1 -1
- package/dist/logger.d.ts +118 -0
- package/dist/logger.d.ts.map +1 -0
- package/dist/logger.js +187 -0
- package/dist/logger.js.map +1 -0
- package/dist/middleware/budget.d.ts +84 -0
- package/dist/middleware/budget.d.ts.map +1 -0
- package/dist/middleware/budget.js +110 -0
- package/dist/middleware/budget.js.map +1 -0
- package/dist/middleware/cache.d.ts +103 -0
- package/dist/middleware/cache.d.ts.map +1 -0
- package/dist/middleware/cache.js +228 -0
- package/dist/middleware/cache.js.map +1 -0
- package/dist/middleware/embed-cache.d.ts +99 -0
- package/dist/middleware/embed-cache.d.ts.map +1 -0
- package/dist/middleware/embed-cache.js +128 -0
- package/dist/middleware/embed-cache.js.map +1 -0
- package/dist/middleware/index.d.ts +11 -0
- package/dist/middleware/index.d.ts.map +1 -0
- package/dist/middleware/index.js +11 -0
- package/dist/middleware/index.js.map +1 -0
- package/dist/middleware/trace.d.ts +103 -0
- package/dist/middleware/trace.d.ts.map +1 -0
- package/dist/middleware/trace.js +176 -0
- package/dist/middleware/trace.js.map +1 -0
- package/dist/primitives.d.ts +120 -1
- package/dist/primitives.d.ts.map +1 -1
- package/dist/primitives.js +398 -26
- package/dist/primitives.js.map +1 -1
- package/dist/retry.d.ts +66 -1
- package/dist/retry.d.ts.map +1 -1
- package/dist/retry.js +115 -8
- package/dist/retry.js.map +1 -1
- package/dist/sandbox.d.ts +36 -0
- package/dist/sandbox.d.ts.map +1 -0
- package/dist/sandbox.js +44 -0
- package/dist/sandbox.js.map +1 -0
- package/dist/schema.js +2 -2
- package/dist/schema.js.map +1 -1
- package/dist/telemetry.d.ts +128 -0
- package/dist/telemetry.d.ts.map +1 -0
- package/dist/telemetry.js +285 -0
- package/dist/telemetry.js.map +1 -0
- package/dist/template.d.ts.map +1 -1
- package/dist/template.js +6 -1
- package/dist/template.js.map +1 -1
- package/dist/tool-orchestration.d.ts +66 -4
- package/dist/tool-orchestration.d.ts.map +1 -1
- package/dist/tool-orchestration.js +123 -23
- package/dist/tool-orchestration.js.map +1 -1
- package/dist/type-guards.d.ts +28 -0
- package/dist/type-guards.d.ts.map +1 -0
- package/dist/type-guards.js +29 -0
- package/dist/type-guards.js.map +1 -0
- package/dist/types.d.ts +155 -19
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js +36 -1
- package/dist/types.js.map +1 -1
- package/dist/wrap-for-v3.d.ts +80 -0
- package/dist/wrap-for-v3.d.ts.map +1 -0
- package/dist/wrap-for-v3.js +89 -0
- package/dist/wrap-for-v3.js.map +1 -0
- package/examples/00-quickstart.ts +232 -0
- package/examples/01-rag-chatbot.ts +212 -0
- package/examples/02-multi-agent-research.ts +290 -0
- package/examples/03-email-classification.ts +379 -0
- package/examples/04-content-moderation.ts +400 -0
- package/examples/05-document-extraction.ts +455 -0
- package/examples/06-streaming-chat-nextjs.ts +437 -0
- package/examples/07-cloudflare-worker.ts +483 -0
- package/examples/08-batch-processing.ts +491 -0
- package/examples/09-budget-constrained.ts +527 -0
- package/examples/10-tool-orchestration.ts +565 -0
- package/examples/11-retry-resilience.ts +403 -0
- package/examples/12-caching-strategies.ts +422 -0
- package/examples/README.md +145 -0
- package/package.json +29 -25
- package/src/ai-promise.ts +226 -140
- package/src/ai-schemas.ts +122 -0
- package/src/ai.ts +71 -1176
- package/src/batch/anthropic.ts +96 -161
- package/src/batch/bedrock.ts +203 -454
- package/src/batch/cloudflare.ts +99 -282
- package/src/batch/google.ts +91 -297
- package/src/batch/index.ts +4 -1
- package/src/batch/memory.ts +15 -10
- package/src/batch/openai.ts +65 -193
- package/src/batch/provider.ts +336 -0
- package/src/batch-map.ts +29 -24
- package/src/batch-queue.ts +200 -11
- package/src/budget.ts +31 -18
- package/src/cache.ts +45 -17
- package/src/context.ts +106 -77
- package/src/digital-objects-registry.ts +750 -0
- package/src/errors.ts +37 -0
- package/src/eval/runner.ts +60 -36
- package/src/eval-log/in-memory.ts +90 -0
- package/src/eval-log/index.ts +46 -0
- package/src/eval-log/types.ts +110 -0
- package/src/function-registry.ts +874 -0
- package/src/generate.ts +33 -28
- package/src/index.ts +122 -21
- package/src/logger.ts +232 -0
- package/src/middleware/budget.ts +171 -0
- package/src/middleware/cache.ts +299 -0
- package/src/middleware/embed-cache.ts +195 -0
- package/src/middleware/index.ts +23 -0
- package/src/middleware/trace.ts +248 -0
- package/src/primitives.ts +589 -62
- package/src/retry.ts +144 -18
- package/src/sandbox.ts +52 -0
- package/src/schema.ts +8 -8
- package/src/telemetry.ts +403 -0
- package/src/template.ts +8 -4
- package/src/tool-orchestration.ts +213 -48
- package/src/type-guards.ts +31 -0
- package/src/types.ts +186 -27
- package/src/wrap-for-v3.ts +105 -0
- package/test/ai-promise.test.ts +1080 -0
- package/test/ai-proxy.test.ts +1 -1
- package/test/batch-autosubmit-errors.test.ts +49 -37
- package/test/batch-blog-posts.test.ts +87 -129
- package/test/core-functions.test.ts +183 -579
- package/test/decide.test.ts +154 -322
- package/test/define.test.ts +211 -8
- package/test/digital-objects-registry.test.ts +760 -0
- package/test/embedding-cache-middleware.test.ts +140 -0
- package/test/fill-template.test.ts +89 -0
- package/test/generate-core.test.ts +140 -229
- package/test/implicit-batch.test.ts +22 -65
- package/test/retry-policy-integration.test.ts +117 -0
- package/test/sandbox-execution.test.ts +155 -0
- package/test/schema.test.ts +55 -19
- package/test/template.test.ts +1164 -0
- package/test/tool-orchestration.test.ts +270 -0
- package/test/wrap-for-v3.test.ts +612 -0
- package/vitest.config.js +6 -0
- package/vitest.config.ts +20 -0
- package/LICENSE +0 -21
- package/dist/rpc/auth.d.ts +0 -69
- package/dist/rpc/auth.d.ts.map +0 -1
- package/dist/rpc/auth.js +0 -136
- package/dist/rpc/auth.js.map +0 -1
- package/dist/rpc/client.d.ts +0 -62
- package/dist/rpc/client.d.ts.map +0 -1
- package/dist/rpc/client.js +0 -103
- package/dist/rpc/client.js.map +0 -1
- package/dist/rpc/deferred.d.ts +0 -60
- package/dist/rpc/deferred.d.ts.map +0 -1
- package/dist/rpc/deferred.js +0 -96
- package/dist/rpc/deferred.js.map +0 -1
- package/dist/rpc/index.d.ts +0 -22
- package/dist/rpc/index.d.ts.map +0 -1
- package/dist/rpc/index.js +0 -38
- package/dist/rpc/index.js.map +0 -1
- package/dist/rpc/local.d.ts +0 -42
- package/dist/rpc/local.d.ts.map +0 -1
- package/dist/rpc/local.js +0 -50
- package/dist/rpc/local.js.map +0 -1
- package/dist/rpc/server.d.ts +0 -165
- package/dist/rpc/server.d.ts.map +0 -1
- package/dist/rpc/server.js +0 -405
- package/dist/rpc/server.js.map +0 -1
- package/dist/rpc/session.d.ts +0 -32
- package/dist/rpc/session.d.ts.map +0 -1
- package/dist/rpc/session.js +0 -43
- package/dist/rpc/session.js.map +0 -1
- package/dist/rpc/transport.d.ts +0 -306
- package/dist/rpc/transport.d.ts.map +0 -1
- package/dist/rpc/transport.js +0 -731
- package/dist/rpc/transport.js.map +0 -1
- package/src/batch/anthropic.js +0 -256
- package/src/batch/bedrock.js +0 -584
- package/src/batch/cloudflare.js +0 -287
- package/src/batch/google.js +0 -359
- package/src/batch/index.js +0 -30
- package/src/batch/memory.js +0 -187
- package/src/batch/openai.js +0 -402
- package/src/eval/index.js +0 -7
- package/src/eval/models.js +0 -119
- package/src/eval/runner.js +0 -147
- package/test/schema.test.js +0 -96
package/dist/errors.d.ts
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Error classes for AI primitives
|
|
3
|
+
*/
|
|
4
|
+
/**
|
|
5
|
+
* Error thrown when a function is not yet implemented.
|
|
6
|
+
*
|
|
7
|
+
* This is used to clearly indicate at runtime that a function exists
|
|
8
|
+
* in the API but does not have a working implementation yet.
|
|
9
|
+
*
|
|
10
|
+
* @example
|
|
11
|
+
* ```ts
|
|
12
|
+
* throw new NotImplementedError('human', 'Human-in-the-loop functions require channel integrations')
|
|
13
|
+
* ```
|
|
14
|
+
*/
|
|
15
|
+
export declare class NotImplementedError extends Error {
|
|
16
|
+
/** The name of the function that is not implemented */
|
|
17
|
+
readonly functionName: string;
|
|
18
|
+
/** Additional details about why it's not implemented or what's needed */
|
|
19
|
+
readonly details?: string;
|
|
20
|
+
constructor(functionName: string, details?: string);
|
|
21
|
+
}
|
|
22
|
+
//# sourceMappingURL=errors.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"errors.d.ts","sourceRoot":"","sources":["../src/errors.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH;;;;;;;;;;GAUG;AACH,qBAAa,mBAAoB,SAAQ,KAAK;IAC5C,uDAAuD;IACvD,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;IAE7B,yEAAyE;IACzE,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAA;gBAEb,YAAY,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM;CAcnD"}
|
package/dist/errors.js
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Error classes for AI primitives
|
|
3
|
+
*/
|
|
4
|
+
/**
|
|
5
|
+
* Error thrown when a function is not yet implemented.
|
|
6
|
+
*
|
|
7
|
+
* This is used to clearly indicate at runtime that a function exists
|
|
8
|
+
* in the API but does not have a working implementation yet.
|
|
9
|
+
*
|
|
10
|
+
* @example
|
|
11
|
+
* ```ts
|
|
12
|
+
* throw new NotImplementedError('human', 'Human-in-the-loop functions require channel integrations')
|
|
13
|
+
* ```
|
|
14
|
+
*/
|
|
15
|
+
export class NotImplementedError extends Error {
|
|
16
|
+
/** The name of the function that is not implemented */
|
|
17
|
+
functionName;
|
|
18
|
+
/** Additional details about why it's not implemented or what's needed */
|
|
19
|
+
details;
|
|
20
|
+
constructor(functionName, details) {
|
|
21
|
+
const message = details
|
|
22
|
+
? `Function '${functionName}' is not implemented: ${details}`
|
|
23
|
+
: `Function '${functionName}' is not implemented`;
|
|
24
|
+
super(message);
|
|
25
|
+
this.name = 'NotImplementedError';
|
|
26
|
+
this.functionName = functionName;
|
|
27
|
+
if (details !== undefined)
|
|
28
|
+
this.details = details;
|
|
29
|
+
// Maintain proper stack trace for where the error was thrown (V8 engines)
|
|
30
|
+
if (Error.captureStackTrace) {
|
|
31
|
+
Error.captureStackTrace(this, NotImplementedError);
|
|
32
|
+
}
|
|
33
|
+
}
|
|
34
|
+
}
|
|
35
|
+
//# sourceMappingURL=errors.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"errors.js","sourceRoot":"","sources":["../src/errors.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH;;;;;;;;;;GAUG;AACH,MAAM,OAAO,mBAAoB,SAAQ,KAAK;IAC5C,uDAAuD;IAC9C,YAAY,CAAQ;IAE7B,yEAAyE;IAChE,OAAO,CAAS;IAEzB,YAAY,YAAoB,EAAE,OAAgB;QAChD,MAAM,OAAO,GAAG,OAAO;YACrB,CAAC,CAAC,aAAa,YAAY,yBAAyB,OAAO,EAAE;YAC7D,CAAC,CAAC,aAAa,YAAY,sBAAsB,CAAA;QACnD,KAAK,CAAC,OAAO,CAAC,CAAA;QACd,IAAI,CAAC,IAAI,GAAG,qBAAqB,CAAA;QACjC,IAAI,CAAC,YAAY,GAAG,YAAY,CAAA;QAChC,IAAI,OAAO,KAAK,SAAS;YAAE,IAAI,CAAC,OAAO,GAAG,OAAO,CAAA;QAEjD,0EAA0E;QAC1E,IAAI,KAAK,CAAC,iBAAiB,EAAE,CAAC;YAC5B,KAAK,CAAC,iBAAiB,CAAC,IAAI,EAAE,mBAAmB,CAAC,CAAA;QACpD,CAAC;IACH,CAAC;CACF"}
|
package/dist/eval/runner.d.ts
CHANGED
|
@@ -7,6 +7,10 @@
|
|
|
7
7
|
import { generateObject, generateText } from '../generate.js';
|
|
8
8
|
import { schema } from '../schema.js';
|
|
9
9
|
import { type EvalModel, type ModelTier } from './models.js';
|
|
10
|
+
/**
|
|
11
|
+
* Output function type for eval progress reporting
|
|
12
|
+
*/
|
|
13
|
+
export type EvalOutputFn = (message: string) => void;
|
|
10
14
|
export interface EvalCase<TInput = unknown, TExpected = unknown> {
|
|
11
15
|
name: string;
|
|
12
16
|
input: TInput;
|
|
@@ -56,6 +60,10 @@ export interface RunEvalOptions<TInput, TOutput, TExpected> {
|
|
|
56
60
|
tiers?: ModelTier[];
|
|
57
61
|
providers?: string[];
|
|
58
62
|
concurrency?: number;
|
|
63
|
+
/** Custom output function for progress reporting (defaults to logger.info) */
|
|
64
|
+
output?: EvalOutputFn;
|
|
65
|
+
/** Whether to suppress progress output (defaults to false) */
|
|
66
|
+
quiet?: boolean;
|
|
59
67
|
}
|
|
60
68
|
/**
|
|
61
69
|
* Run an eval suite across models
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAC7D,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAA;AACrC,OAAO,EAAwC,KAAK,SAAS,EAAE,KAAK,SAAS,EAAE,MAAM,aAAa,CAAA;
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAC7D,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAA;AACrC,OAAO,EAAwC,KAAK,SAAS,EAAE,KAAK,SAAS,EAAE,MAAM,aAAa,CAAA;AAGlG;;GAEG;AACH,MAAM,MAAM,YAAY,GAAG,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAA;AAOpD,MAAM,WAAW,QAAQ,CAAC,MAAM,GAAG,OAAO,EAAE,SAAS,GAAG,OAAO;IAC7D,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,CAAC,EAAE,SAAS,CAAA;CACrB;AAED,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,QAAQ,CAAC,EAAE,OAAO,CAAA;CACnB;AAED,MAAM,WAAW,UAAU,CAAC,OAAO,GAAG,OAAO;IAC3C,KAAK,EAAE,SAAS,CAAA;IAChB,IAAI,EAAE,QAAQ,CAAA;IACd,mEAAmE;IACnE,MAAM,EAAE,OAAO,GAAG,IAAI,CAAA;IACtB,MAAM,EAAE,SAAS,EAAE,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,UAAU,EAAE,CAAA;IACrB,QAAQ,EAAE,MAAM,CAAA;IAChB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC5D,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS;IACxD,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,SAAS,CAAC,EAAE,CAAA;IACpC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC,OAAO,CAAC,CAAA;IAC3D,OAAO,EAAE,KAAK,CAAC;QACb,IAAI,EAAE,MAAM,CAAA;QACZ,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,MAAM,EAAE,CAAC,IAAI,EAAE;YACb,KAAK,EAAE,MAAM,CAAA;YACb,MAAM,EAAE,OAAO,CAAA;YACf,QAAQ,CAAC,EAAE,SAAS,CAAA;SACrB,KAAK,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;KAC/B,CAAC,CAAA;IACF,MAAM,CAAC,EAAE,SAAS,EAAE,CAAA;IACpB,KAAK,CAAC,EAAE,SAAS,EAAE,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;IACpB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,8EAA8E;IAC9E,MAAM,CAAC,EAAE,YAAY,CAAA;IACrB,8DAA8D;IAC9D,KAAK,CAAC,EAAE,OAAO,CAAA;CAChB;AAED;;GAEG;AACH,wBAAsB,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS,EACtD,OAAO,EAAE,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS,CAAC,GAClD,OAAO,CAAC,WAAW,CAAC,CA2JtB;AAGD,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,EAAE,CAAA"}
|
package/dist/eval/runner.js
CHANGED
|
@@ -7,22 +7,30 @@
|
|
|
7
7
|
import { generateObject, generateText } from '../generate.js';
|
|
8
8
|
import { schema } from '../schema.js';
|
|
9
9
|
import { createModelVariants, getModelPricing } from './models.js';
|
|
10
|
+
import { getLogger } from '../logger.js';
|
|
11
|
+
/**
|
|
12
|
+
* Default output function uses logger.info
|
|
13
|
+
*/
|
|
14
|
+
const defaultOutput = (message) => getLogger().info(message);
|
|
10
15
|
/**
|
|
11
16
|
* Run an eval suite across models
|
|
12
17
|
*/
|
|
13
18
|
export async function runEval(options) {
|
|
14
|
-
const { name, cases, task, scorers, concurrency = 3 } = options;
|
|
19
|
+
const { name, cases, task, scorers, concurrency = 3, quiet = false } = options;
|
|
20
|
+
const log = quiet ? () => { } : options.output ?? defaultOutput;
|
|
15
21
|
// Get models to test
|
|
16
|
-
const
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
22
|
+
const variantOptions = {};
|
|
23
|
+
if (options.tiers !== undefined)
|
|
24
|
+
variantOptions.tiers = options.tiers;
|
|
25
|
+
if (options.providers !== undefined)
|
|
26
|
+
variantOptions.providers = options.providers;
|
|
27
|
+
const models = options.models ?? createModelVariants(variantOptions).map((v) => v.input);
|
|
20
28
|
const results = [];
|
|
21
29
|
const startTime = Date.now();
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
30
|
+
log(`\nRunning eval: ${name}`);
|
|
31
|
+
log(` Models: ${models.map((m) => m.name).join(', ')}`);
|
|
32
|
+
log(` Cases: ${cases.length}`);
|
|
33
|
+
log('');
|
|
26
34
|
// Run all model/case combinations
|
|
27
35
|
const jobs = [];
|
|
28
36
|
for (const model of models) {
|
|
@@ -37,7 +45,7 @@ export async function runEval(options) {
|
|
|
37
45
|
const caseStart = Date.now();
|
|
38
46
|
try {
|
|
39
47
|
// Run the task
|
|
40
|
-
const
|
|
48
|
+
const taskOutput = await task(job.case.input, job.model);
|
|
41
49
|
const latencyMs = Date.now() - caseStart;
|
|
42
50
|
// Run scorers
|
|
43
51
|
const scores = [];
|
|
@@ -45,20 +53,20 @@ export async function runEval(options) {
|
|
|
45
53
|
try {
|
|
46
54
|
const score = await s.scorer({
|
|
47
55
|
input: job.case.input,
|
|
48
|
-
output,
|
|
49
|
-
expected: job.case.expected,
|
|
56
|
+
output: taskOutput,
|
|
57
|
+
...(job.case.expected !== undefined && { expected: job.case.expected }),
|
|
50
58
|
});
|
|
51
59
|
scores.push({
|
|
52
60
|
name: s.name,
|
|
53
61
|
score: Math.max(0, Math.min(1, score)),
|
|
54
|
-
description: s.description,
|
|
62
|
+
...(s.description && { description: s.description }),
|
|
55
63
|
});
|
|
56
64
|
}
|
|
57
65
|
catch (err) {
|
|
58
66
|
scores.push({
|
|
59
67
|
name: s.name,
|
|
60
68
|
score: 0,
|
|
61
|
-
description: s.description,
|
|
69
|
+
...(s.description && { description: s.description }),
|
|
62
70
|
metadata: { error: String(err) },
|
|
63
71
|
});
|
|
64
72
|
}
|
|
@@ -69,29 +77,29 @@ export async function runEval(options) {
|
|
|
69
77
|
const estimatedPromptTokens = 100;
|
|
70
78
|
const estimatedCompletionTokens = 200;
|
|
71
79
|
const cost = pricing
|
|
72
|
-
? (estimatedPromptTokens * pricing.prompt +
|
|
73
|
-
|
|
74
|
-
|
|
75
|
-
? scores.reduce((sum, s) => sum + s.score, 0) / scores.length
|
|
80
|
+
? (estimatedPromptTokens * pricing.prompt +
|
|
81
|
+
estimatedCompletionTokens * pricing.completion) /
|
|
82
|
+
1_000_000
|
|
76
83
|
: 0;
|
|
77
|
-
const
|
|
78
|
-
|
|
84
|
+
const avgScore = scores.length > 0 ? scores.reduce((sum, s) => sum + s.score, 0) / scores.length : 0;
|
|
85
|
+
const symbol = avgScore >= 0.8 ? 'PASS' : avgScore >= 0.5 ? 'WARN' : 'FAIL';
|
|
86
|
+
log(` ${symbol} ${job.model.name} | ${job.case.name} | ${(avgScore * 100).toFixed(0)}% | ${latencyMs}ms`);
|
|
79
87
|
return {
|
|
80
88
|
model: job.model,
|
|
81
89
|
case: job.case,
|
|
82
|
-
output,
|
|
90
|
+
output: taskOutput,
|
|
83
91
|
scores,
|
|
84
92
|
latencyMs,
|
|
85
93
|
cost,
|
|
86
94
|
};
|
|
87
95
|
}
|
|
88
96
|
catch (err) {
|
|
89
|
-
|
|
97
|
+
log(` FAIL ${job.model.name} | ${job.case.name} | ERROR: ${err}`);
|
|
90
98
|
return {
|
|
91
99
|
model: job.model,
|
|
92
100
|
case: job.case,
|
|
93
101
|
output: null,
|
|
94
|
-
scores: scorers.map(s => ({ name: s.name, score: 0 })),
|
|
102
|
+
scores: scorers.map((s) => ({ name: s.name, score: 0 })),
|
|
95
103
|
latencyMs: Date.now() - caseStart,
|
|
96
104
|
cost: 0,
|
|
97
105
|
error: String(err),
|
|
@@ -103,10 +111,8 @@ export async function runEval(options) {
|
|
|
103
111
|
// Calculate summary
|
|
104
112
|
const totalTime = Date.now() - startTime;
|
|
105
113
|
const totalCost = results.reduce((sum, r) => sum + r.cost, 0);
|
|
106
|
-
const allScores = results.flatMap(r => r.scores.map(s => s.score));
|
|
107
|
-
const avgScore = allScores.length > 0
|
|
108
|
-
? allScores.reduce((a, b) => a + b, 0) / allScores.length
|
|
109
|
-
: 0;
|
|
114
|
+
const allScores = results.flatMap((r) => r.scores.map((s) => s.score));
|
|
115
|
+
const avgScore = allScores.length > 0 ? allScores.reduce((a, b) => a + b, 0) / allScores.length : 0;
|
|
110
116
|
// Group by model
|
|
111
117
|
const byModel = {};
|
|
112
118
|
for (const result of results) {
|
|
@@ -124,15 +130,15 @@ export async function runEval(options) {
|
|
|
124
130
|
entry.avgScore /= entry.count;
|
|
125
131
|
}
|
|
126
132
|
}
|
|
127
|
-
|
|
128
|
-
|
|
129
|
-
|
|
130
|
-
|
|
131
|
-
|
|
132
|
-
|
|
133
|
-
|
|
133
|
+
log('');
|
|
134
|
+
log(`Results:`);
|
|
135
|
+
log(` Overall: ${(avgScore * 100).toFixed(1)}%`);
|
|
136
|
+
log(` Time: ${(totalTime / 1000).toFixed(1)}s`);
|
|
137
|
+
log(` Cost: $${totalCost.toFixed(4)}`);
|
|
138
|
+
log('');
|
|
139
|
+
log(' By Model:');
|
|
134
140
|
for (const [modelId, stats] of Object.entries(byModel)) {
|
|
135
|
-
|
|
141
|
+
log(` - ${modelId}: ${(stats.avgScore * 100).toFixed(1)}%`);
|
|
136
142
|
}
|
|
137
143
|
return {
|
|
138
144
|
name,
|
package/dist/eval/runner.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAC7D,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAA;AACrC,OAAO,EAAE,mBAAmB,EAAE,eAAe,EAAkC,MAAM,aAAa,CAAA;
|
|
1
|
+
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAC7D,OAAO,EAAE,MAAM,EAAE,MAAM,cAAc,CAAA;AACrC,OAAO,EAAE,mBAAmB,EAAE,eAAe,EAAkC,MAAM,aAAa,CAAA;AAClG,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAA;AAOxC;;GAEG;AACH,MAAM,aAAa,GAAiB,CAAC,OAAe,EAAE,EAAE,CAAC,SAAS,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAA;AA0DlF;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,OAAmD;IAEnD,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,WAAW,GAAG,CAAC,EAAE,KAAK,GAAG,KAAK,EAAE,GAAG,OAAO,CAAA;IAC9E,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,IAAI,aAAa,CAAA;IAE9D,qBAAqB;IACrB,MAAM,cAAc,GAAkD,EAAE,CAAA;IACxE,IAAI,OAAO,CAAC,KAAK,KAAK,SAAS;QAAE,cAAc,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAA;IACrE,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS;QAAE,cAAc,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,CAAA;IACjF,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,IAAI,mBAAmB,CAAC,cAAc,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAA;IAExF,MAAM,OAAO,GAA0B,EAAE,CAAA;IACzC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;IAE5B,GAAG,CAAC,mBAAmB,IAAI,EAAE,CAAC,CAAA;IAC9B,GAAG,CAAC,cAAc,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IACzD,GAAG,CAAC,aAAa,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;IAChC,GAAG,CAAC,EAAE,CAAC,CAAA;IAEP,kCAAkC;IAClC,MAAM,IAAI,GAAmE,EAAE,CAAA;IAC/E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,KAAK,MAAM,QAAQ,IAAI,KAAK,EAAE,CAAC;YAC7B,IAAI,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAA;QACtC,CAAC;IACH,CAAC;IAED,4CAA4C;IAC5C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,IAAI,WAAW,EAAE,CAAC;QAClD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,WAAW,CAAC,CAAA;QAE5C,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,GAAG,CACpC,KAAK,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;YACtB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;YAE5B,IAAI,CAAC;gBACH,eAAe;gBACf,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,CAAA;gBACxD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAA;gBAExC,cAAc;gBACd,MAAM,MAAM,GAAgB,EAAE,CAAA;gBAC9B,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;oBACxB,IAAI,CAAC;wBACH,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,MAAM,CAAC;4BAC3B,KAAK,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK;4BACrB,MAAM,EAAE,UAAU;4BAClB,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,KAAK,SAAS,IAAI,EAAE,QAAQ,EAAE,GAAG,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;yBACxE,CAAC,CAAA;wBACF,MAAM,CAAC,IAAI,CAAC;4BACV,IAAI,EAAE,CAAC,CAAC,IAAI;4BACZ,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;4BACtC,GAAG,CAAC,CAAC,CAAC,WAAW,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;yBACrD,CAAC,CAAA;oBACJ,CAAC;oBAAC,OAAO,GAAG,EAAE,CAAC;wBACb,MAAM,CAAC,IAAI,CAAC;4BACV,IAAI,EAAE,CAAC,CAAC,IAAI;4BACZ,KAAK,EAAE,CAAC;4BACR,GAAG,CAAC,CAAC,CAAC,WAAW,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;4BACpD,QAAQ,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE;yBACjC,CAAC,CAAA;oBACJ,CAAC;gBACH,CAAC;gBAED,iBAAiB;gBACjB,MAAM,OAAO,GAAG,eAAe,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;gBAC7C,wCAAwC;gBACxC,MAAM,qBAAqB,GAAG,GAAG,CAAA;gBACjC,MAAM,yBAAyB,GAAG,GAAG,CAAA;gBACrC,MAAM,IAAI,GAAG,OAAO;oBAClB,CAAC,CAAC,CAAC,qBAAqB,GAAG,OAAO,CAAC,MAAM;wBACrC,yBAAyB,GAAG,OAAO,CAAC,UAAU,CAAC;wBACjD,SAAS;oBACX,CAAC,CAAC,CAAC,CAAA;gBAEL,MAAM,QAAQ,GACZ,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAA;gBAErF,MAAM,MAAM,GAAG,QAAQ,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAA;gBAC3E,GAAG,CACD,MAAM,MAAM,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAC7E,CAAC,CACF,OAAO,SAAS,IAAI,CACtB,CAAA;gBAED,OAAO;oBACL,KAAK,EAAE,GAAG,CAAC,KAAK;oBAChB,IAAI,EAAE,GAAG,CAAC,IAAI;oBACd,MAAM,EAAE,UAAU;oBAClB,MAAM;oBACN,SAAS;oBACT,IAAI;iBACL,CAAA;YACH,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,GAAG,CAAC,WAAW,GAAG,CAAC,KAAK,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,aAAa,GAAG,EAAE,CAAC,CAAA;gBAEnE,OAAO;oBACL,KAAK,EAAE,GAAG,CAAC,KAAK;oBAChB,IAAI,EAAE,GAAG,CAAC,IAAI;oBACd,MAAM,EAAE,IAAI;oBACZ,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;oBACxD,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBACjC,IAAI,EAAE,CAAC;oBACP,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC;iBACnB,CAAA;YACH,CAAC;QACH,CAAC,CAAC,CACH,CAAA;QAED,OAAO,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,CAAA;IAC/B,CAAC;IAED,oBAAoB;IACpB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAA;IACxC,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAA;IAC7D,MAAM,SAAS,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAA;IACtE,MAAM,QAAQ,GACZ,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAA;IAEpF,iBAAiB;IACjB,MAAM,OAAO,GAAwD,EAAE,CAAA;IACvE,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,CAAA;QAChC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;YACvB,OAAO,CAAC,QAAQ,CAAC,GAAG,EAAE,QAAQ,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAA;QAC/C,CAAC;QACD,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAA;QAC3F,OAAO,CAAC,QAAQ,CAAC,CAAC,QAAQ,IAAI,SAAS,CAAA;QACvC,OAAO,CAAC,QAAQ,CAAC,CAAC,KAAK,EAAE,CAAA;IAC3B,CAAC;IACD,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,CAAA;QAC1B,IAAI,KAAK,EAAE,CAAC;YACV,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,KAAK,CAAA;QAC/B,CAAC;IACH,CAAC;IAED,GAAG,CAAC,EAAE,CAAC,CAAA;IACP,GAAG,CAAC,UAAU,CAAC,CAAA;IACf,GAAG,CAAC,eAAe,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IAClD,GAAG,CAAC,YAAY,CAAC,SAAS,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IACjD,GAAG,CAAC,aAAa,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IACxC,GAAG,CAAC,EAAE,CAAC,CAAA;IACP,GAAG,CAAC,cAAc,CAAC,CAAA;IACnB,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACvD,GAAG,CAAC,QAAQ,OAAO,KAAK,CAAC,KAAK,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IAC/D,CAAC;IAED,OAAO;QACL,IAAI;QACJ,OAAO;QACP,QAAQ;QACR,OAAO;QACP,SAAS;QACT,SAAS;KACV,CAAA;AACH,CAAC;AAED,oBAAoB;AACpB,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,EAAE,CAAA"}
|
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* InMemoryEvalLogStore — Map-backed default implementation of
|
|
3
|
+
* {@link EvalLogStore}.
|
|
4
|
+
*
|
|
5
|
+
* Matches Evalite v1's default backend: process-local Map keyed on `$id`,
|
|
6
|
+
* insertion-ordered for "most recent first" listing without sorting. Suitable
|
|
7
|
+
* for single-process tests, evals, and the cascade walker's in-flight log;
|
|
8
|
+
* not suitable for cross-process or multi-worker setups (use a disk/SQLite
|
|
9
|
+
* backend for those — same contract).
|
|
10
|
+
*
|
|
11
|
+
* @packageDocumentation
|
|
12
|
+
*/
|
|
13
|
+
import type { EvalLogEntry, EvalLogListOptions, EvalLogStore } from './types.js';
|
|
14
|
+
/**
|
|
15
|
+
* In-memory implementation of {@link EvalLogStore}.
|
|
16
|
+
*/
|
|
17
|
+
export declare class InMemoryEvalLogStore implements EvalLogStore {
|
|
18
|
+
/**
|
|
19
|
+
* Map keyed on `$id`. Insertion order on a JS Map is preserved, so we
|
|
20
|
+
* walk it in reverse for "most recent first" listing.
|
|
21
|
+
*/
|
|
22
|
+
private readonly entries;
|
|
23
|
+
record(entry: Omit<EvalLogEntry, '$id' | 'createdAt'> & Partial<Pick<EvalLogEntry, '$id' | 'createdAt'>>): Promise<EvalLogEntry>;
|
|
24
|
+
get(id: string): Promise<EvalLogEntry | undefined>;
|
|
25
|
+
list(options?: EvalLogListOptions): Promise<EvalLogEntry[]>;
|
|
26
|
+
delete(id: string): Promise<boolean>;
|
|
27
|
+
/**
|
|
28
|
+
* Convenience for tests: drop every entry. Not on the public
|
|
29
|
+
* {@link EvalLogStore} interface because the disk/SQLite backends may not
|
|
30
|
+
* want to expose a one-shot wipe.
|
|
31
|
+
*/
|
|
32
|
+
clear(): void;
|
|
33
|
+
}
|
|
34
|
+
//# sourceMappingURL=in-memory.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"in-memory.d.ts","sourceRoot":"","sources":["../../src/eval-log/in-memory.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAEhF;;GAEG;AACH,qBAAa,oBAAqB,YAAW,YAAY;IACvD;;;OAGG;IACH,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAuC;IAEzD,MAAM,CACV,KAAK,EAAE,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,WAAW,CAAC,GAC5C,OAAO,CAAC,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,WAAW,CAAC,CAAC,GACjD,OAAO,CAAC,YAAY,CAAC;IAmBlB,GAAG,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,GAAG,SAAS,CAAC;IAIlD,IAAI,CAAC,OAAO,GAAE,kBAAuB,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;IAyB/D,MAAM,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAI1C;;;;OAIG;IACH,KAAK,IAAI,IAAI;CAGd"}
|
|
@@ -0,0 +1,84 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* InMemoryEvalLogStore — Map-backed default implementation of
|
|
3
|
+
* {@link EvalLogStore}.
|
|
4
|
+
*
|
|
5
|
+
* Matches Evalite v1's default backend: process-local Map keyed on `$id`,
|
|
6
|
+
* insertion-ordered for "most recent first" listing without sorting. Suitable
|
|
7
|
+
* for single-process tests, evals, and the cascade walker's in-flight log;
|
|
8
|
+
* not suitable for cross-process or multi-worker setups (use a disk/SQLite
|
|
9
|
+
* backend for those — same contract).
|
|
10
|
+
*
|
|
11
|
+
* @packageDocumentation
|
|
12
|
+
*/
|
|
13
|
+
import { randomUUID } from 'crypto';
|
|
14
|
+
/**
|
|
15
|
+
* In-memory implementation of {@link EvalLogStore}.
|
|
16
|
+
*/
|
|
17
|
+
export class InMemoryEvalLogStore {
|
|
18
|
+
/**
|
|
19
|
+
* Map keyed on `$id`. Insertion order on a JS Map is preserved, so we
|
|
20
|
+
* walk it in reverse for "most recent first" listing.
|
|
21
|
+
*/
|
|
22
|
+
entries = new Map();
|
|
23
|
+
async record(entry) {
|
|
24
|
+
const $id = entry.$id ?? randomUUID();
|
|
25
|
+
const createdAt = entry.createdAt ?? Date.now();
|
|
26
|
+
const stored = {
|
|
27
|
+
$id,
|
|
28
|
+
createdAt,
|
|
29
|
+
model: entry.model,
|
|
30
|
+
prompt: entry.prompt,
|
|
31
|
+
response: entry.response,
|
|
32
|
+
usage: entry.usage,
|
|
33
|
+
costUsd: entry.costUsd,
|
|
34
|
+
durationMs: entry.durationMs,
|
|
35
|
+
...(entry.traceId !== undefined ? { traceId: entry.traceId } : {}),
|
|
36
|
+
...(entry.tags !== undefined ? { tags: entry.tags } : {}),
|
|
37
|
+
};
|
|
38
|
+
this.entries.set($id, stored);
|
|
39
|
+
return stored;
|
|
40
|
+
}
|
|
41
|
+
async get(id) {
|
|
42
|
+
return this.entries.get(id);
|
|
43
|
+
}
|
|
44
|
+
async list(options = {}) {
|
|
45
|
+
const { traceId, model, tags, limit } = options;
|
|
46
|
+
const out = [];
|
|
47
|
+
// Iterate in reverse insertion order — Map preserves order; we walk
|
|
48
|
+
// values into an array, then reverse for most-recent-first.
|
|
49
|
+
const all = Array.from(this.entries.values()).reverse();
|
|
50
|
+
for (const entry of all) {
|
|
51
|
+
if (traceId !== undefined && entry.traceId !== traceId)
|
|
52
|
+
continue;
|
|
53
|
+
if (model !== undefined && entry.model !== model)
|
|
54
|
+
continue;
|
|
55
|
+
if (tags !== undefined) {
|
|
56
|
+
let matchesAll = true;
|
|
57
|
+
for (const k of Object.keys(tags)) {
|
|
58
|
+
if (entry.tags?.[k] !== tags[k]) {
|
|
59
|
+
matchesAll = false;
|
|
60
|
+
break;
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
if (!matchesAll)
|
|
64
|
+
continue;
|
|
65
|
+
}
|
|
66
|
+
out.push(entry);
|
|
67
|
+
if (limit !== undefined && out.length >= limit)
|
|
68
|
+
break;
|
|
69
|
+
}
|
|
70
|
+
return out;
|
|
71
|
+
}
|
|
72
|
+
async delete(id) {
|
|
73
|
+
return this.entries.delete(id);
|
|
74
|
+
}
|
|
75
|
+
/**
|
|
76
|
+
* Convenience for tests: drop every entry. Not on the public
|
|
77
|
+
* {@link EvalLogStore} interface because the disk/SQLite backends may not
|
|
78
|
+
* want to expose a one-shot wipe.
|
|
79
|
+
*/
|
|
80
|
+
clear() {
|
|
81
|
+
this.entries.clear();
|
|
82
|
+
}
|
|
83
|
+
}
|
|
84
|
+
//# sourceMappingURL=in-memory.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"in-memory.js","sourceRoot":"","sources":["../../src/eval-log/in-memory.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAA;AAGnC;;GAEG;AACH,MAAM,OAAO,oBAAoB;IAC/B;;;OAGG;IACc,OAAO,GAA8B,IAAI,GAAG,EAAE,CAAA;IAE/D,KAAK,CAAC,MAAM,CACV,KACkD;QAElD,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,IAAI,UAAU,EAAE,CAAA;QACrC,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,IAAI,IAAI,CAAC,GAAG,EAAE,CAAA;QAC/C,MAAM,MAAM,GAAiB;YAC3B,GAAG;YACH,SAAS;YACT,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,GAAG,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAClE,GAAG,CAAC,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC1D,CAAA;QACD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAA;QAC7B,OAAO,MAAM,CAAA;IACf,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,EAAU;QAClB,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAA;IAC7B,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,UAA8B,EAAE;QACzC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,OAAO,CAAA;QAC/C,MAAM,GAAG,GAAmB,EAAE,CAAA;QAC9B,oEAAoE;QACpE,4DAA4D;QAC5D,MAAM,GAAG,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAA;QACvD,KAAK,MAAM,KAAK,IAAI,GAAG,EAAE,CAAC;YACxB,IAAI,OAAO,KAAK,SAAS,IAAI,KAAK,CAAC,OAAO,KAAK,OAAO;gBAAE,SAAQ;YAChE,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK;gBAAE,SAAQ;YAC1D,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;gBACvB,IAAI,UAAU,GAAG,IAAI,CAAA;gBACrB,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;oBAClC,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;wBAChC,UAAU,GAAG,KAAK,CAAA;wBAClB,MAAK;oBACP,CAAC;gBACH,CAAC;gBACD,IAAI,CAAC,UAAU;oBAAE,SAAQ;YAC3B,CAAC;YACD,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;YACf,IAAI,KAAK,KAAK,SAAS,IAAI,GAAG,CAAC,MAAM,IAAI,KAAK;gBAAE,MAAK;QACvD,CAAC;QACD,OAAO,GAAG,CAAA;IACZ,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,EAAU;QACrB,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC,CAAA;IAChC,CAAC;IAED;;;;OAIG;IACH,KAAK;QACH,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAA;IACtB,CAAC;CACF"}
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* EvalLogStore — pluggable persistence primitive for trace/eval entries.
|
|
3
|
+
*
|
|
4
|
+
* Exports the {@link EvalLogStore} contract, the
|
|
5
|
+
* {@link InMemoryEvalLogStore} default implementation, and a global
|
|
6
|
+
* accessor pair (`getEvalLogStore` / `configureEvalLogStore`) mirroring the
|
|
7
|
+
* marketplace persistence pattern from round 9.
|
|
8
|
+
*
|
|
9
|
+
* @packageDocumentation
|
|
10
|
+
*/
|
|
11
|
+
import type { EvalLogStore } from './types.js';
|
|
12
|
+
export type { EvalLogEntry, EvalLogListOptions, EvalLogStore } from './types.js';
|
|
13
|
+
export { InMemoryEvalLogStore } from './in-memory.js';
|
|
14
|
+
/**
|
|
15
|
+
* Get the global {@link EvalLogStore}. Lazily constructs an
|
|
16
|
+
* {@link InMemoryEvalLogStore} on first call when no store has been
|
|
17
|
+
* configured.
|
|
18
|
+
*
|
|
19
|
+
* Match the round-9 marketplace persistence accessor: callers that don't
|
|
20
|
+
* care about isolation read the global; callers that do (tests, multi-tenant
|
|
21
|
+
* apps) install their own via {@link configureEvalLogStore}.
|
|
22
|
+
*/
|
|
23
|
+
export declare function getEvalLogStore(): EvalLogStore;
|
|
24
|
+
/**
|
|
25
|
+
* Install a global {@link EvalLogStore}. Pass `null` to reset to the lazy
|
|
26
|
+
* in-memory default (useful in test teardown).
|
|
27
|
+
*/
|
|
28
|
+
export declare function configureEvalLogStore(store: EvalLogStore | null): void;
|
|
29
|
+
//# sourceMappingURL=index.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/eval-log/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAE9C,YAAY,EAAE,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAChF,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAA;AAQrD;;;;;;;;GAQG;AACH,wBAAgB,eAAe,IAAI,YAAY,CAK9C;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,YAAY,GAAG,IAAI,GAAG,IAAI,CAEtE"}
|
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* EvalLogStore — pluggable persistence primitive for trace/eval entries.
|
|
3
|
+
*
|
|
4
|
+
* Exports the {@link EvalLogStore} contract, the
|
|
5
|
+
* {@link InMemoryEvalLogStore} default implementation, and a global
|
|
6
|
+
* accessor pair (`getEvalLogStore` / `configureEvalLogStore`) mirroring the
|
|
7
|
+
* marketplace persistence pattern from round 9.
|
|
8
|
+
*
|
|
9
|
+
* @packageDocumentation
|
|
10
|
+
*/
|
|
11
|
+
import { InMemoryEvalLogStore } from './in-memory.js';
|
|
12
|
+
export { InMemoryEvalLogStore } from './in-memory.js';
|
|
13
|
+
// ============================================================================
|
|
14
|
+
// Global accessor (lazy default + override)
|
|
15
|
+
// ============================================================================
|
|
16
|
+
let _store = null;
|
|
17
|
+
/**
|
|
18
|
+
* Get the global {@link EvalLogStore}. Lazily constructs an
|
|
19
|
+
* {@link InMemoryEvalLogStore} on first call when no store has been
|
|
20
|
+
* configured.
|
|
21
|
+
*
|
|
22
|
+
* Match the round-9 marketplace persistence accessor: callers that don't
|
|
23
|
+
* care about isolation read the global; callers that do (tests, multi-tenant
|
|
24
|
+
* apps) install their own via {@link configureEvalLogStore}.
|
|
25
|
+
*/
|
|
26
|
+
export function getEvalLogStore() {
|
|
27
|
+
if (_store === null) {
|
|
28
|
+
_store = new InMemoryEvalLogStore();
|
|
29
|
+
}
|
|
30
|
+
return _store;
|
|
31
|
+
}
|
|
32
|
+
/**
|
|
33
|
+
* Install a global {@link EvalLogStore}. Pass `null` to reset to the lazy
|
|
34
|
+
* in-memory default (useful in test teardown).
|
|
35
|
+
*/
|
|
36
|
+
export function configureEvalLogStore(store) {
|
|
37
|
+
_store = store;
|
|
38
|
+
}
|
|
39
|
+
//# sourceMappingURL=index.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/eval-log/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAA;AAIrD,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAA;AAErD,+EAA+E;AAC/E,4CAA4C;AAC5C,+EAA+E;AAE/E,IAAI,MAAM,GAAwB,IAAI,CAAA;AAEtC;;;;;;;;GAQG;AACH,MAAM,UAAU,eAAe;IAC7B,IAAI,MAAM,KAAK,IAAI,EAAE,CAAC;QACpB,MAAM,GAAG,IAAI,oBAAoB,EAAE,CAAA;IACrC,CAAC;IACD,OAAO,MAAM,CAAA;AACf,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAA0B;IAC9D,MAAM,GAAG,KAAK,CAAA;AAChB,CAAC"}
|
|
@@ -0,0 +1,101 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* EvalLogStore — pluggable persistence primitive for trace/eval entries.
|
|
3
|
+
*
|
|
4
|
+
* Forward-looking primitive matching Evalite v1's EvalLogStore pattern:
|
|
5
|
+
* the in-memory default ships today; the disk/SQLite/durable backends can
|
|
6
|
+
* land later without breaking the trace middleware contract.
|
|
7
|
+
*
|
|
8
|
+
* Used downstream by `traceMiddleware` (in `../middleware/trace.ts`) as the
|
|
9
|
+
* sink for per-call prompt+response+usage records. The cascade-walker in
|
|
10
|
+
* services-as-software will consume `list()` / `get()` to populate the
|
|
11
|
+
* InvocationEvent stream once round 16+ adds the `'persona-trace'` variant.
|
|
12
|
+
*
|
|
13
|
+
* @packageDocumentation
|
|
14
|
+
*/
|
|
15
|
+
/**
|
|
16
|
+
* A single entry in the eval log — one LLM call with its full payload.
|
|
17
|
+
*
|
|
18
|
+
* Shape mirrors what `traceMiddleware` emits, with optional `tags` for
|
|
19
|
+
* caller-supplied dimensions (persona name, evaluator role, cascade depth).
|
|
20
|
+
*/
|
|
21
|
+
export interface EvalLogEntry {
|
|
22
|
+
/** MDXLD identity — typically a UUID generated at insert time. */
|
|
23
|
+
$id: string;
|
|
24
|
+
/**
|
|
25
|
+
* Optional caller-supplied trace correlation ID. When the cascade walker
|
|
26
|
+
* spans multiple LLM calls under one user request, all entries share the
|
|
27
|
+
* same `traceId` so `list({ traceId })` rolls them up.
|
|
28
|
+
*/
|
|
29
|
+
traceId?: string;
|
|
30
|
+
/** Model identifier (e.g. `'anthropic/claude-sonnet-4.5'` or `'sonnet'`). */
|
|
31
|
+
model: string;
|
|
32
|
+
/**
|
|
33
|
+
* Stringified prompt as submitted to the model. We don't store the
|
|
34
|
+
* structured `LanguageModelV3Prompt` shape because (a) it's bulky and (b)
|
|
35
|
+
* downstream consumers (replay, fixture diff) only need the text payload.
|
|
36
|
+
*/
|
|
37
|
+
prompt: string;
|
|
38
|
+
/** The model's text response. Tool calls/files are not stored here. */
|
|
39
|
+
response: string;
|
|
40
|
+
/** Token usage as reported by the AI SDK. */
|
|
41
|
+
usage: {
|
|
42
|
+
inputTokens: number;
|
|
43
|
+
outputTokens: number;
|
|
44
|
+
};
|
|
45
|
+
/** Computed USD cost (caller-supplied via the `pricing` overlay). */
|
|
46
|
+
costUsd: number;
|
|
47
|
+
/** Wall-clock duration of the underlying `doGenerate` / `doStream` call. */
|
|
48
|
+
durationMs: number;
|
|
49
|
+
/** Caller-supplied dimensions (persona, evaluator role, cascade step). */
|
|
50
|
+
tags?: Record<string, string>;
|
|
51
|
+
/** Insert timestamp (epoch ms). */
|
|
52
|
+
createdAt: number;
|
|
53
|
+
}
|
|
54
|
+
/**
|
|
55
|
+
* Options accepted by `EvalLogStore.list`. All fields are AND-combined.
|
|
56
|
+
*/
|
|
57
|
+
export interface EvalLogListOptions {
|
|
58
|
+
/** Filter to entries with this trace correlation ID. */
|
|
59
|
+
traceId?: string;
|
|
60
|
+
/** Filter to entries for a specific model. */
|
|
61
|
+
model?: string;
|
|
62
|
+
/**
|
|
63
|
+
* Filter to entries whose `tags` are a *superset* of the supplied object.
|
|
64
|
+
* (E.g. `{ persona: 'cfo' }` matches entries tagged
|
|
65
|
+
* `{ persona: 'cfo', step: '3' }` but not entries tagged
|
|
66
|
+
* `{ persona: 'cto' }`.)
|
|
67
|
+
*/
|
|
68
|
+
tags?: Record<string, string>;
|
|
69
|
+
/** Maximum number of entries to return (most recent first). */
|
|
70
|
+
limit?: number;
|
|
71
|
+
}
|
|
72
|
+
/**
|
|
73
|
+
* Pluggable persistence interface for eval log entries.
|
|
74
|
+
*
|
|
75
|
+
* Modeled after the Evalite v1 EvalLogStore contract: in-memory default,
|
|
76
|
+
* disk JSON / SQLite / durable backends supplied via
|
|
77
|
+
* `configureEvalLogStore`.
|
|
78
|
+
*
|
|
79
|
+
* All methods are async to keep the contract uniform across backends — the
|
|
80
|
+
* in-memory implementation resolves synchronously under the hood.
|
|
81
|
+
*/
|
|
82
|
+
export interface EvalLogStore {
|
|
83
|
+
/**
|
|
84
|
+
* Persist a new entry. Returns the stored entry (with `$id` and
|
|
85
|
+
* `createdAt` filled in if the caller omitted them).
|
|
86
|
+
*/
|
|
87
|
+
record(entry: Omit<EvalLogEntry, '$id' | 'createdAt'> & Partial<Pick<EvalLogEntry, '$id' | 'createdAt'>>): Promise<EvalLogEntry>;
|
|
88
|
+
/**
|
|
89
|
+
* Read an entry by `$id`. Returns `undefined` when not found.
|
|
90
|
+
*/
|
|
91
|
+
get(id: string): Promise<EvalLogEntry | undefined>;
|
|
92
|
+
/**
|
|
93
|
+
* List entries matching the supplied filter. Returns most recent first.
|
|
94
|
+
*/
|
|
95
|
+
list(options?: EvalLogListOptions): Promise<EvalLogEntry[]>;
|
|
96
|
+
/**
|
|
97
|
+
* Delete an entry. Returns `true` if an entry was actually removed.
|
|
98
|
+
*/
|
|
99
|
+
delete(id: string): Promise<boolean>;
|
|
100
|
+
}
|
|
101
|
+
//# sourceMappingURL=types.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/eval-log/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAMH;;;;;GAKG;AACH,MAAM,WAAW,YAAY;IAC3B,kEAAkE;IAClE,GAAG,EAAE,MAAM,CAAA;IACX;;;;OAIG;IACH,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,6EAA6E;IAC7E,KAAK,EAAE,MAAM,CAAA;IACb;;;;OAIG;IACH,MAAM,EAAE,MAAM,CAAA;IACd,uEAAuE;IACvE,QAAQ,EAAE,MAAM,CAAA;IAChB,6CAA6C;IAC7C,KAAK,EAAE;QACL,WAAW,EAAE,MAAM,CAAA;QACnB,YAAY,EAAE,MAAM,CAAA;KACrB,CAAA;IACD,qEAAqE;IACrE,OAAO,EAAE,MAAM,CAAA;IACf,4EAA4E;IAC5E,UAAU,EAAE,MAAM,CAAA;IAClB,0EAA0E;IAC1E,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC7B,mCAAmC;IACnC,SAAS,EAAE,MAAM,CAAA;CAClB;AAED;;GAEG;AACH,MAAM,WAAW,kBAAkB;IACjC,wDAAwD;IACxD,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,8CAA8C;IAC9C,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;;;;OAKG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC7B,+DAA+D;IAC/D,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED;;;;;;;;;GASG;AACH,MAAM,WAAW,YAAY;IAC3B;;;OAGG;IACH,MAAM,CACJ,KAAK,EAAE,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,WAAW,CAAC,GAC5C,OAAO,CAAC,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,WAAW,CAAC,CAAC,GACjD,OAAO,CAAC,YAAY,CAAC,CAAA;IACxB;;OAEG;IACH,GAAG,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,GAAG,SAAS,CAAC,CAAA;IAClD;;OAEG;IACH,IAAI,CAAC,OAAO,CAAC,EAAE,kBAAkB,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAAA;IAC3D;;OAEG;IACH,MAAM,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAA;CACrC"}
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* EvalLogStore — pluggable persistence primitive for trace/eval entries.
|
|
3
|
+
*
|
|
4
|
+
* Forward-looking primitive matching Evalite v1's EvalLogStore pattern:
|
|
5
|
+
* the in-memory default ships today; the disk/SQLite/durable backends can
|
|
6
|
+
* land later without breaking the trace middleware contract.
|
|
7
|
+
*
|
|
8
|
+
* Used downstream by `traceMiddleware` (in `../middleware/trace.ts`) as the
|
|
9
|
+
* sink for per-call prompt+response+usage records. The cascade-walker in
|
|
10
|
+
* services-as-software will consume `list()` / `get()` to populate the
|
|
11
|
+
* InvocationEvent stream once round 16+ adds the `'persona-trace'` variant.
|
|
12
|
+
*
|
|
13
|
+
* @packageDocumentation
|
|
14
|
+
*/
|
|
15
|
+
export {};
|
|
16
|
+
//# sourceMappingURL=types.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/eval-log/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG"}
|