ai-functions 0.3.0 → 0.4.0
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 +5 -0
- package/.turbo/turbo-test.log +105 -0
- package/README.md +190 -86
- package/TODO.md +138 -0
- package/dist/ai-promise.d.ts +219 -0
- package/dist/ai-promise.d.ts.map +1 -0
- package/dist/ai-promise.js +610 -0
- package/dist/ai-promise.js.map +1 -0
- package/dist/ai.d.ts +285 -0
- package/dist/ai.d.ts.map +1 -0
- package/dist/ai.js +842 -0
- package/dist/ai.js.map +1 -0
- package/dist/batch/anthropic.d.ts +23 -0
- package/dist/batch/anthropic.d.ts.map +1 -0
- package/dist/batch/anthropic.js +257 -0
- package/dist/batch/anthropic.js.map +1 -0
- package/dist/batch/bedrock.d.ts +64 -0
- package/dist/batch/bedrock.d.ts.map +1 -0
- package/dist/batch/bedrock.js +586 -0
- package/dist/batch/bedrock.js.map +1 -0
- package/dist/batch/cloudflare.d.ts +37 -0
- package/dist/batch/cloudflare.d.ts.map +1 -0
- package/dist/batch/cloudflare.js +289 -0
- package/dist/batch/cloudflare.js.map +1 -0
- package/dist/batch/google.d.ts +41 -0
- package/dist/batch/google.d.ts.map +1 -0
- package/dist/batch/google.js +360 -0
- package/dist/batch/google.js.map +1 -0
- package/dist/batch/index.d.ts +31 -0
- package/dist/batch/index.d.ts.map +1 -0
- package/dist/batch/index.js +31 -0
- package/dist/batch/index.js.map +1 -0
- package/dist/batch/memory.d.ts +44 -0
- package/dist/batch/memory.d.ts.map +1 -0
- package/dist/batch/memory.js +188 -0
- package/dist/batch/memory.js.map +1 -0
- package/dist/batch/openai.d.ts +37 -0
- package/dist/batch/openai.d.ts.map +1 -0
- package/dist/batch/openai.js +403 -0
- package/dist/batch/openai.js.map +1 -0
- package/dist/batch-map.d.ts +125 -0
- package/dist/batch-map.d.ts.map +1 -0
- package/dist/batch-map.js +406 -0
- package/dist/batch-map.js.map +1 -0
- package/dist/batch-queue.d.ts +273 -0
- package/dist/batch-queue.d.ts.map +1 -0
- package/dist/batch-queue.js +271 -0
- package/dist/batch-queue.js.map +1 -0
- package/dist/context.d.ts +133 -0
- package/dist/context.d.ts.map +1 -0
- package/dist/context.js +267 -0
- package/dist/context.js.map +1 -0
- package/dist/embeddings.d.ts +123 -0
- package/dist/embeddings.d.ts.map +1 -0
- package/dist/embeddings.js +170 -0
- package/dist/embeddings.js.map +1 -0
- package/dist/eval/index.d.ts +8 -0
- package/dist/eval/index.d.ts.map +1 -0
- package/dist/eval/index.js +8 -0
- package/dist/eval/index.js.map +1 -0
- package/dist/eval/models.d.ts +66 -0
- package/dist/eval/models.d.ts.map +1 -0
- package/dist/eval/models.js +120 -0
- package/dist/eval/models.js.map +1 -0
- package/dist/eval/runner.d.ts +64 -0
- package/dist/eval/runner.d.ts.map +1 -0
- package/dist/eval/runner.js +148 -0
- package/dist/eval/runner.js.map +1 -0
- package/dist/generate.d.ts +168 -0
- package/dist/generate.d.ts.map +1 -0
- package/dist/generate.js +174 -0
- package/dist/generate.js.map +1 -0
- package/dist/index.d.ts +29 -4
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +53 -52
- package/dist/index.js.map +1 -1
- package/dist/primitives.d.ts +292 -0
- package/dist/primitives.d.ts.map +1 -0
- package/dist/primitives.js +471 -0
- package/dist/primitives.js.map +1 -0
- package/dist/providers/cloudflare.d.ts +9 -0
- package/dist/providers/cloudflare.d.ts.map +1 -0
- package/dist/providers/cloudflare.js +9 -0
- package/dist/providers/cloudflare.js.map +1 -0
- package/dist/providers/index.d.ts +9 -0
- package/dist/providers/index.d.ts.map +1 -0
- package/dist/providers/index.js +9 -0
- package/dist/providers/index.js.map +1 -0
- package/dist/schema.d.ts +54 -0
- package/dist/schema.d.ts.map +1 -0
- package/dist/schema.js +109 -0
- package/dist/schema.js.map +1 -0
- package/dist/template.d.ts +73 -0
- package/dist/template.d.ts.map +1 -0
- package/dist/template.js +129 -0
- package/dist/template.js.map +1 -0
- package/dist/types.d.ts +474 -106
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js +4 -8
- package/dist/types.js.map +1 -1
- package/evalite.config.ts +19 -0
- package/evals/README.md +212 -0
- package/evals/classification.eval.ts +108 -0
- package/evals/marketing.eval.ts +370 -0
- package/evals/math.eval.ts +94 -0
- package/evals/run-evals.ts +166 -0
- package/evals/structured-output.eval.ts +143 -0
- package/evals/writing.eval.ts +117 -0
- package/examples/batch-blog-posts.ts +160 -0
- package/package.json +57 -57
- package/src/ai-promise.ts +784 -0
- package/src/ai.ts +1183 -0
- package/src/batch/anthropic.ts +375 -0
- package/src/batch/bedrock.ts +801 -0
- package/src/batch/cloudflare.ts +421 -0
- package/src/batch/google.ts +491 -0
- package/src/batch/index.ts +31 -0
- package/src/batch/memory.ts +253 -0
- package/src/batch/openai.ts +557 -0
- package/src/batch-map.ts +534 -0
- package/src/batch-queue.ts +493 -0
- package/src/context.ts +332 -0
- package/src/embeddings.ts +244 -0
- package/src/eval/index.ts +8 -0
- package/src/eval/models.ts +158 -0
- package/src/eval/runner.ts +217 -0
- package/src/generate.ts +245 -0
- package/src/index.ts +154 -0
- package/src/primitives.ts +612 -0
- package/src/providers/cloudflare.ts +15 -0
- package/src/providers/index.ts +14 -0
- package/src/schema.ts +147 -0
- package/src/template.ts +209 -0
- package/src/types.ts +540 -0
- package/test/README.md +105 -0
- package/test/ai-proxy.test.ts +192 -0
- package/test/async-iterators.test.ts +327 -0
- package/test/batch-background.test.ts +482 -0
- package/test/batch-blog-posts.test.ts +387 -0
- package/test/blog-generation.test.ts +510 -0
- package/test/browse-read.test.ts +611 -0
- package/test/core-functions.test.ts +694 -0
- package/test/decide.test.ts +393 -0
- package/test/define.test.ts +274 -0
- package/test/e2e-bedrock-manual.ts +163 -0
- package/test/e2e-bedrock.test.ts +191 -0
- package/test/e2e-flex-gateway.ts +157 -0
- package/test/e2e-flex-manual.ts +183 -0
- package/test/e2e-flex.test.ts +209 -0
- package/test/e2e-google-manual.ts +178 -0
- package/test/e2e-google.test.ts +216 -0
- package/test/embeddings.test.ts +284 -0
- package/test/evals/define-function.eval.test.ts +379 -0
- package/test/evals/primitives.eval.test.ts +384 -0
- package/test/function-types.test.ts +492 -0
- package/test/generate-core.test.ts +319 -0
- package/test/generate.test.ts +163 -0
- package/test/implicit-batch.test.ts +422 -0
- package/test/schema.test.ts +109 -0
- package/test/tagged-templates.test.ts +302 -0
- package/tsconfig.json +10 -0
- package/vitest.config.ts +42 -0
- package/LICENSE +0 -21
- package/bin/cli.js +0 -5
- package/dist/cli/index.d.ts +0 -10
- package/dist/cli/index.d.ts.map +0 -1
- package/dist/cli/index.js +0 -38
- package/dist/cli/index.js.map +0 -1
- package/dist/cli/index.test.d.ts +0 -2
- package/dist/cli/index.test.d.ts.map +0 -1
- package/dist/cli/index.test.js +0 -35
- package/dist/cli/index.test.js.map +0 -1
- package/dist/constants/models.d.ts +0 -10
- package/dist/constants/models.d.ts.map +0 -1
- package/dist/constants/models.js +0 -12
- package/dist/constants/models.js.map +0 -1
- package/dist/converters/index.d.ts +0 -3
- package/dist/converters/index.d.ts.map +0 -1
- package/dist/converters/index.js +0 -3
- package/dist/converters/index.js.map +0 -1
- package/dist/converters/model.d.ts +0 -4
- package/dist/converters/model.d.ts.map +0 -1
- package/dist/converters/model.js +0 -19
- package/dist/converters/model.js.map +0 -1
- package/dist/converters/schema.d.ts +0 -4
- package/dist/converters/schema.d.ts.map +0 -1
- package/dist/converters/schema.js +0 -25
- package/dist/converters/schema.js.map +0 -1
- package/dist/core/responses.d.ts +0 -5
- package/dist/core/responses.d.ts.map +0 -1
- package/dist/core/responses.js +0 -16
- package/dist/core/responses.js.map +0 -1
- package/dist/core/responses.test.d.ts +0 -2
- package/dist/core/responses.test.d.ts.map +0 -1
- package/dist/core/responses.test.js +0 -31
- package/dist/core/responses.test.js.map +0 -1
- package/dist/errors.d.ts +0 -6
- package/dist/errors.d.ts.map +0 -1
- package/dist/errors.js +0 -9
- package/dist/errors.js.map +0 -1
- package/dist/examples/streaming.test.d.ts +0 -2
- package/dist/examples/streaming.test.d.ts.map +0 -1
- package/dist/examples/streaming.test.js +0 -176
- package/dist/examples/streaming.test.js.map +0 -1
- package/dist/factory/__tests__/index.test.d.ts +0 -2
- package/dist/factory/__tests__/index.test.d.ts.map +0 -1
- package/dist/factory/__tests__/index.test.js +0 -430
- package/dist/factory/__tests__/index.test.js.map +0 -1
- package/dist/factory/__tests__/list.test.d.ts +0 -2
- package/dist/factory/__tests__/list.test.d.ts.map +0 -1
- package/dist/factory/__tests__/list.test.js +0 -92
- package/dist/factory/__tests__/list.test.js.map +0 -1
- package/dist/factory/index.d.ts +0 -20
- package/dist/factory/index.d.ts.map +0 -1
- package/dist/factory/index.js +0 -287
- package/dist/factory/index.js.map +0 -1
- package/dist/factory/index.test.d.ts +0 -2
- package/dist/factory/index.test.d.ts.map +0 -1
- package/dist/factory/index.test.js +0 -287
- package/dist/factory/index.test.js.map +0 -1
- package/dist/factory/list.d.ts +0 -3
- package/dist/factory/list.d.ts.map +0 -1
- package/dist/factory/list.js +0 -221
- package/dist/factory/list.js.map +0 -1
- package/dist/factory/list.test.d.ts +0 -2
- package/dist/factory/list.test.d.ts.map +0 -1
- package/dist/factory/list.test.js +0 -84
- package/dist/factory/list.test.js.map +0 -1
- package/dist/generate/index.d.ts +0 -5
- package/dist/generate/index.d.ts.map +0 -1
- package/dist/generate/index.js +0 -17
- package/dist/generate/index.js.map +0 -1
- package/dist/index.test.d.ts +0 -2
- package/dist/index.test.d.ts.map +0 -1
- package/dist/index.test.js +0 -59
- package/dist/index.test.js.map +0 -1
- package/dist/list/await.d.ts +0 -3
- package/dist/list/await.d.ts.map +0 -1
- package/dist/list/await.js +0 -28
- package/dist/list/await.js.map +0 -1
- package/dist/list/constants.d.ts +0 -4
- package/dist/list/constants.d.ts.map +0 -1
- package/dist/list/constants.js +0 -5
- package/dist/list/constants.js.map +0 -1
- package/dist/list/create-function.d.ts +0 -3
- package/dist/list/create-function.d.ts.map +0 -1
- package/dist/list/create-function.js +0 -11
- package/dist/list/create-function.js.map +0 -1
- package/dist/list/index.d.ts +0 -4
- package/dist/list/index.d.ts.map +0 -1
- package/dist/list/index.js +0 -5
- package/dist/list/index.js.map +0 -1
- package/dist/list/prompt.d.ts +0 -3
- package/dist/list/prompt.d.ts.map +0 -1
- package/dist/list/prompt.js +0 -6
- package/dist/list/prompt.js.map +0 -1
- package/dist/list/schemas.d.ts +0 -4
- package/dist/list/schemas.d.ts.map +0 -1
- package/dist/list/schemas.js +0 -8
- package/dist/list/schemas.js.map +0 -1
- package/dist/list/stream.d.ts +0 -3
- package/dist/list/stream.d.ts.map +0 -1
- package/dist/list/stream.js +0 -33
- package/dist/list/stream.js.map +0 -1
- package/dist/list/types.d.ts +0 -11
- package/dist/list/types.d.ts.map +0 -1
- package/dist/list/types.js +0 -2
- package/dist/list/types.js.map +0 -1
- package/dist/list/validation.d.ts +0 -3
- package/dist/list/validation.d.ts.map +0 -1
- package/dist/list/validation.js +0 -12
- package/dist/list/validation.js.map +0 -1
- package/dist/providers/config.d.ts +0 -4
- package/dist/providers/config.d.ts.map +0 -1
- package/dist/providers/config.js +0 -21
- package/dist/providers/config.js.map +0 -1
- package/dist/providers/config.test.d.ts +0 -2
- package/dist/providers/config.test.d.ts.map +0 -1
- package/dist/providers/config.test.js +0 -37
- package/dist/providers/config.test.js.map +0 -1
- package/dist/proxy/constants.d.ts +0 -4
- package/dist/proxy/constants.d.ts.map +0 -1
- package/dist/proxy/constants.js +0 -5
- package/dist/proxy/constants.js.map +0 -1
- package/dist/proxy/create-function.d.ts +0 -4
- package/dist/proxy/create-function.d.ts.map +0 -1
- package/dist/proxy/create-function.js +0 -24
- package/dist/proxy/create-function.js.map +0 -1
- package/dist/proxy/create-proxy.d.ts +0 -2
- package/dist/proxy/create-proxy.d.ts.map +0 -1
- package/dist/proxy/create-proxy.js +0 -11
- package/dist/proxy/create-proxy.js.map +0 -1
- package/dist/proxy/function-generator.d.ts +0 -9
- package/dist/proxy/function-generator.d.ts.map +0 -1
- package/dist/proxy/function-generator.js +0 -29
- package/dist/proxy/function-generator.js.map +0 -1
- package/dist/proxy/index.d.ts +0 -4
- package/dist/proxy/index.d.ts.map +0 -1
- package/dist/proxy/index.js +0 -4
- package/dist/proxy/index.js.map +0 -1
- package/dist/proxy/prompt.d.ts +0 -2
- package/dist/proxy/prompt.d.ts.map +0 -1
- package/dist/proxy/prompt.js +0 -6
- package/dist/proxy/prompt.js.map +0 -1
- package/dist/proxy/types.d.ts +0 -7
- package/dist/proxy/types.d.ts.map +0 -1
- package/dist/proxy/types.js +0 -2
- package/dist/proxy/types.js.map +0 -1
- package/dist/queue/manager.d.ts +0 -5
- package/dist/queue/manager.d.ts.map +0 -1
- package/dist/queue/manager.js +0 -37
- package/dist/queue/manager.js.map +0 -1
- package/dist/queue/manager.test.d.ts +0 -2
- package/dist/queue/manager.test.d.ts.map +0 -1
- package/dist/queue/manager.test.js +0 -52
- package/dist/queue/manager.test.js.map +0 -1
- package/dist/schema-converter.d.ts +0 -4
- package/dist/schema-converter.d.ts.map +0 -1
- package/dist/schema-converter.js +0 -30
- package/dist/schema-converter.js.map +0 -1
- package/dist/stream/index.d.ts +0 -7
- package/dist/stream/index.d.ts.map +0 -1
- package/dist/stream/index.js +0 -23
- package/dist/stream/index.js.map +0 -1
- package/dist/streaming/utils.d.ts +0 -4
- package/dist/streaming/utils.d.ts.map +0 -1
- package/dist/streaming/utils.js +0 -131
- package/dist/streaming/utils.js.map +0 -1
- package/dist/streaming/utils.test.d.ts +0 -2
- package/dist/streaming/utils.test.d.ts.map +0 -1
- package/dist/streaming/utils.test.js +0 -84
- package/dist/streaming/utils.test.js.map +0 -1
- package/dist/templates/result.d.ts +0 -7
- package/dist/templates/result.d.ts.map +0 -1
- package/dist/templates/result.js +0 -40
- package/dist/templates/result.js.map +0 -1
- package/dist/templates/result.test.d.ts +0 -2
- package/dist/templates/result.test.d.ts.map +0 -1
- package/dist/templates/result.test.js +0 -75
- package/dist/templates/result.test.js.map +0 -1
- package/dist/test/setup.d.ts +0 -2
- package/dist/test/setup.d.ts.map +0 -1
- package/dist/test/setup.js +0 -21
- package/dist/test/setup.js.map +0 -1
- package/dist/test-types.d.ts +0 -13
- package/dist/test-types.d.ts.map +0 -1
- package/dist/test-types.js +0 -55
- package/dist/test-types.js.map +0 -1
- package/dist/types/index.d.ts +0 -4
- package/dist/types/index.d.ts.map +0 -1
- package/dist/types/index.js +0 -4
- package/dist/types/index.js.map +0 -1
- package/dist/types/list.d.ts +0 -10
- package/dist/types/list.d.ts.map +0 -1
- package/dist/types/list.js +0 -2
- package/dist/types/list.js.map +0 -1
- package/dist/types/model.d.ts +0 -7
- package/dist/types/model.d.ts.map +0 -1
- package/dist/types/model.js +0 -2
- package/dist/types/model.js.map +0 -1
- package/dist/types/options.d.ts +0 -25
- package/dist/types/options.d.ts.map +0 -1
- package/dist/types/options.js +0 -2
- package/dist/types/options.js.map +0 -1
- package/dist/types/schema.d.ts +0 -5
- package/dist/types/schema.d.ts.map +0 -1
- package/dist/types/schema.js +0 -2
- package/dist/types/schema.js.map +0 -1
- package/dist/utils/__tests__/request-handler.test.d.ts +0 -2
- package/dist/utils/__tests__/request-handler.test.d.ts.map +0 -1
- package/dist/utils/__tests__/request-handler.test.js +0 -134
- package/dist/utils/__tests__/request-handler.test.js.map +0 -1
- package/dist/utils/__tests__/schema.test.d.ts +0 -2
- package/dist/utils/__tests__/schema.test.d.ts.map +0 -1
- package/dist/utils/__tests__/schema.test.js +0 -49
- package/dist/utils/__tests__/schema.test.js.map +0 -1
- package/dist/utils/__tests__/stream-progress.test.d.ts +0 -2
- package/dist/utils/__tests__/stream-progress.test.d.ts.map +0 -1
- package/dist/utils/__tests__/stream-progress.test.js +0 -85
- package/dist/utils/__tests__/stream-progress.test.js.map +0 -1
- package/dist/utils/index.d.ts +0 -2
- package/dist/utils/index.d.ts.map +0 -1
- package/dist/utils/index.js +0 -2
- package/dist/utils/index.js.map +0 -1
- package/dist/utils/request-handler.d.ts +0 -17
- package/dist/utils/request-handler.d.ts.map +0 -1
- package/dist/utils/request-handler.js +0 -105
- package/dist/utils/request-handler.js.map +0 -1
- package/dist/utils/schema.d.ts +0 -11
- package/dist/utils/schema.d.ts.map +0 -1
- package/dist/utils/schema.js +0 -51
- package/dist/utils/schema.js.map +0 -1
- package/dist/utils/stream-progress.d.ts +0 -17
- package/dist/utils/stream-progress.d.ts.map +0 -1
- package/dist/utils/stream-progress.js +0 -86
- package/dist/utils/stream-progress.js.map +0 -1
- package/dist/utils/validation.d.ts +0 -3
- package/dist/utils/validation.d.ts.map +0 -1
- package/dist/utils/validation.js +0 -30
- package/dist/utils/validation.js.map +0 -1
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../../src/eval/models.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAsB,KAAK,SAAS,EAAE,MAAM,iBAAiB,CAAA;AAEpE,MAAM,MAAM,SAAS,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAA;AAEjD,MAAM,WAAW,SAAS;IACxB,6BAA6B;IAC7B,EAAE,EAAE,MAAM,CAAA;IACV,0BAA0B;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,oBAAoB;IACpB,QAAQ,EAAE,MAAM,CAAA;IAChB,sBAAsB;IACtB,IAAI,EAAE,SAAS,CAAA;IACf,qBAAqB;IACrB,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAMD;;;;;;;;GAQG;AACH,eAAO,MAAM,WAAW,EAAE,SAAS,EA2ClC,CAAA;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,SAAS,GAAG,SAAS,EAAE,CAE5D;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,EAAE,CAEjE;AAED;;GAEG;AACH,wBAAgB,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS,CAG9D;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,EAAE,EAAE,MAAM,GAAG;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAA;CAAE,GAAG,SAAS,CAQ9F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,CAAC,EAAE;IACzC,KAAK,CAAC,EAAE,SAAS,EAAE,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAA;CACrB,GAAG,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,SAAS,CAAA;CAAE,CAAC,CAe5C;AAED;;GAEG;AACH,wBAAgB,uBAAuB,CAAC,IAAI,EAAE,SAAS,GAAG,SAAS,EAAE,CAYpE"}
|
|
@@ -0,0 +1,120 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Model Registry for AI Functions Eval Suite
|
|
3
|
+
*
|
|
4
|
+
* Simple model list for running evals across providers.
|
|
5
|
+
* Uses ai-providers/language-models for resolution and pricing.
|
|
6
|
+
*
|
|
7
|
+
* @packageDocumentation
|
|
8
|
+
*/
|
|
9
|
+
import { resolve, get, list } from 'language-models';
|
|
10
|
+
// ============================================================================
|
|
11
|
+
// Models to evaluate - using aliases from language-models
|
|
12
|
+
// ============================================================================
|
|
13
|
+
/**
|
|
14
|
+
* Core models to test - one per tier per major provider
|
|
15
|
+
* These resolve via ai-providers to OpenRouter or direct SDKs
|
|
16
|
+
*
|
|
17
|
+
* Updated: December 2025
|
|
18
|
+
*
|
|
19
|
+
* Note: Some models use OpenRouter format (provider/model) to avoid
|
|
20
|
+
* resolution issues with provider_model_id mismatches.
|
|
21
|
+
*/
|
|
22
|
+
export const EVAL_MODELS = [
|
|
23
|
+
// Anthropic Claude 4.5 - via AWS Bedrock (uses AWS credits with bearer token auth)
|
|
24
|
+
// All Claude models should be 4.5 - older versions are deprecated
|
|
25
|
+
{ id: 'bedrock:us.anthropic.claude-opus-4-5-20251101-v1:0', name: 'Claude Opus 4.5', provider: 'anthropic', tier: 'best', notes: 'Bedrock' },
|
|
26
|
+
{ id: 'bedrock:us.anthropic.claude-sonnet-4-5-20250929-v1:0', name: 'Claude Sonnet 4.5', provider: 'anthropic', tier: 'fast', notes: 'Bedrock' },
|
|
27
|
+
{ id: 'bedrock:us.anthropic.claude-haiku-4-5-20251001-v1:0', name: 'Claude Haiku 4.5', provider: 'anthropic', tier: 'cheap', notes: 'Bedrock' },
|
|
28
|
+
// OpenAI - GPT-5.1 variants + GPT-oss (open source)
|
|
29
|
+
{ id: 'openai/o3', name: 'o3', provider: 'openai', tier: 'best' },
|
|
30
|
+
{ id: 'openai/gpt-5.1', name: 'GPT-5.1', provider: 'openai', tier: 'best' },
|
|
31
|
+
{ id: 'openai/gpt-5-mini', name: 'GPT-5 Mini', provider: 'openai', tier: 'fast' },
|
|
32
|
+
{ id: 'openai/gpt-5-nano', name: 'GPT-5 Nano', provider: 'openai', tier: 'cheap' },
|
|
33
|
+
// GPT-oss 120B removed - times out frequently
|
|
34
|
+
{ id: 'openai/gpt-oss-20b', name: 'GPT-oss 20B', provider: 'openai', tier: 'fast', notes: 'Open source' },
|
|
35
|
+
// Google - Gemini 3 (November 2025)
|
|
36
|
+
{ id: 'google/gemini-3-pro-preview', name: 'Gemini 3 Pro', provider: 'google', tier: 'best', notes: '1M context, #1 LMArena' },
|
|
37
|
+
// Gemini 2.5 Pro removed - times out frequently
|
|
38
|
+
{ id: 'flash', name: 'Gemini 2.5 Flash', provider: 'google', tier: 'fast' },
|
|
39
|
+
// Meta (via OpenRouter)
|
|
40
|
+
{ id: 'meta-llama/llama-4-maverick', name: 'Llama 4 Maverick', provider: 'meta-llama', tier: 'best' },
|
|
41
|
+
{ id: 'meta-llama/llama-3.3-70b-instruct', name: 'Llama 3.3 70B', provider: 'meta-llama', tier: 'fast' },
|
|
42
|
+
// DeepSeek - V3.2 (December 2025)
|
|
43
|
+
{ id: 'deepseek/deepseek-v3.2', name: 'DeepSeek V3.2', provider: 'deepseek', tier: 'best', notes: 'GPT-5 class reasoning' },
|
|
44
|
+
// DeepSeek V3.2 Speciale removed - no tool use support on OpenRouter
|
|
45
|
+
{ id: 'deepseek/deepseek-chat', name: 'DeepSeek Chat', provider: 'deepseek', tier: 'fast' },
|
|
46
|
+
// Mistral - Mistral 3 family (December 2025)
|
|
47
|
+
{ id: 'mistralai/mistral-large-2512', name: 'Mistral Large 3', provider: 'mistralai', tier: 'best', notes: '675B MoE, 41B active' },
|
|
48
|
+
{ id: 'mistralai/mistral-medium-3.1', name: 'Mistral Medium 3.1', provider: 'mistralai', tier: 'fast' },
|
|
49
|
+
// Ministral 3 14B removed - often fails structured output
|
|
50
|
+
// Qwen - Qwen3 family (2025)
|
|
51
|
+
{ id: 'qwen/qwen3-coder', name: 'Qwen3 Coder 480B', provider: 'qwen', tier: 'best', notes: 'Agentic coding' },
|
|
52
|
+
{ id: 'qwen/qwen3-30b-a3b', name: 'Qwen3 30B', provider: 'qwen', tier: 'fast', notes: 'MoE 30B/3B active' },
|
|
53
|
+
{ id: 'qwen/qwen3-next-80b-a3b-instruct', name: 'Qwen3 Next 80B', provider: 'qwen', tier: 'best', notes: 'Ultra-long context' },
|
|
54
|
+
// xAI - Grok 4 family (December 2025)
|
|
55
|
+
{ id: 'x-ai/grok-4', name: 'Grok 4', provider: 'x-ai', tier: 'best', notes: '256K context, reasoning' },
|
|
56
|
+
{ id: 'x-ai/grok-4.1-fast', name: 'Grok 4.1 Fast', provider: 'x-ai', tier: 'fast', notes: '2M context, agentic' },
|
|
57
|
+
{ id: 'x-ai/grok-4-fast', name: 'Grok 4 Fast', provider: 'x-ai', tier: 'fast', notes: '2M context' },
|
|
58
|
+
];
|
|
59
|
+
/**
|
|
60
|
+
* Get models by tier
|
|
61
|
+
*/
|
|
62
|
+
export function getModelsByTier(tier) {
|
|
63
|
+
return EVAL_MODELS.filter(m => m.tier === tier);
|
|
64
|
+
}
|
|
65
|
+
/**
|
|
66
|
+
* Get models by provider
|
|
67
|
+
*/
|
|
68
|
+
export function getModelsByProvider(provider) {
|
|
69
|
+
return EVAL_MODELS.filter(m => m.provider === provider);
|
|
70
|
+
}
|
|
71
|
+
/**
|
|
72
|
+
* Get model info from language-models package (includes pricing)
|
|
73
|
+
*/
|
|
74
|
+
export function getModelInfo(id) {
|
|
75
|
+
const resolved = resolve(id);
|
|
76
|
+
return get(resolved);
|
|
77
|
+
}
|
|
78
|
+
/**
|
|
79
|
+
* Get pricing for a model (from OpenRouter data)
|
|
80
|
+
*/
|
|
81
|
+
export function getModelPricing(id) {
|
|
82
|
+
const info = getModelInfo(id);
|
|
83
|
+
if (!info?.pricing)
|
|
84
|
+
return undefined;
|
|
85
|
+
return {
|
|
86
|
+
prompt: parseFloat(info.pricing.prompt) * 1_000_000, // Convert to per-million
|
|
87
|
+
completion: parseFloat(info.pricing.completion) * 1_000_000,
|
|
88
|
+
};
|
|
89
|
+
}
|
|
90
|
+
/**
|
|
91
|
+
* Create evalite variants for model testing
|
|
92
|
+
*/
|
|
93
|
+
export function createModelVariants(opts) {
|
|
94
|
+
let models = EVAL_MODELS;
|
|
95
|
+
if (opts?.tiers) {
|
|
96
|
+
models = models.filter(m => opts.tiers.includes(m.tier));
|
|
97
|
+
}
|
|
98
|
+
if (opts?.providers) {
|
|
99
|
+
models = models.filter(m => opts.providers.includes(m.provider));
|
|
100
|
+
}
|
|
101
|
+
return models.map(model => ({
|
|
102
|
+
name: `${model.provider}/${model.name}`,
|
|
103
|
+
input: model,
|
|
104
|
+
}));
|
|
105
|
+
}
|
|
106
|
+
/**
|
|
107
|
+
* Get a representative model from each provider for a given tier
|
|
108
|
+
*/
|
|
109
|
+
export function getRepresentativeModels(tier) {
|
|
110
|
+
const seen = new Set();
|
|
111
|
+
const result = [];
|
|
112
|
+
for (const model of EVAL_MODELS) {
|
|
113
|
+
if (model.tier === tier && !seen.has(model.provider)) {
|
|
114
|
+
seen.add(model.provider);
|
|
115
|
+
result.push(model);
|
|
116
|
+
}
|
|
117
|
+
}
|
|
118
|
+
return result;
|
|
119
|
+
}
|
|
120
|
+
//# sourceMappingURL=models.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"models.js","sourceRoot":"","sources":["../../src/eval/models.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,IAAI,EAAkB,MAAM,iBAAiB,CAAA;AAiBpE,+EAA+E;AAC/E,0DAA0D;AAC1D,+EAA+E;AAE/E;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,WAAW,GAAgB;IACtC,mFAAmF;IACnF,kEAAkE;IAClE,EAAE,EAAE,EAAE,oDAAoD,EAAE,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE;IAC5I,EAAE,EAAE,EAAE,sDAAsD,EAAE,IAAI,EAAE,mBAAmB,EAAE,QAAQ,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE;IAChJ,EAAE,EAAE,EAAE,qDAAqD,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE;IAE/I,oDAAoD;IACpD,EAAE,EAAE,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE;IACjE,EAAE,EAAE,EAAE,gBAAgB,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE;IAC3E,EAAE,EAAE,EAAE,mBAAmB,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE;IACjF,EAAE,EAAE,EAAE,mBAAmB,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE;IAClF,8CAA8C;IAC9C,EAAE,EAAE,EAAE,oBAAoB,EAAE,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,EAAE;IAEzG,oCAAoC;IACpC,EAAE,EAAE,EAAE,6BAA6B,EAAE,IAAI,EAAE,cAAc,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,wBAAwB,EAAE;IAC9H,gDAAgD;IAChD,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE;IAE3E,wBAAwB;IACxB,EAAE,EAAE,EAAE,6BAA6B,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,YAAY,EAAE,IAAI,EAAE,MAAM,EAAE;IACrG,EAAE,EAAE,EAAE,mCAAmC,EAAE,IAAI,EAAE,eAAe,EAAE,QAAQ,EAAE,YAAY,EAAE,IAAI,EAAE,MAAM,EAAE;IAExG,kCAAkC;IAClC,EAAE,EAAE,EAAE,wBAAwB,EAAE,IAAI,EAAE,eAAe,EAAE,QAAQ,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,uBAAuB,EAAE;IAC3H,qEAAqE;IACrE,EAAE,EAAE,EAAE,wBAAwB,EAAE,IAAI,EAAE,eAAe,EAAE,QAAQ,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,EAAE;IAE3F,6CAA6C;IAC7C,EAAE,EAAE,EAAE,8BAA8B,EAAE,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,sBAAsB,EAAE;IACnI,EAAE,EAAE,EAAE,8BAA8B,EAAE,IAAI,EAAE,oBAAoB,EAAE,QAAQ,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE;IACvG,0DAA0D;IAE1D,6BAA6B;IAC7B,EAAE,EAAE,EAAE,kBAAkB,EAAE,IAAI,EAAE,kBAAkB,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE;IAC7G,EAAE,EAAE,EAAE,oBAAoB,EAAE,IAAI,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,mBAAmB,EAAE;IAC3G,EAAE,EAAE,EAAE,kCAAkC,EAAE,IAAI,EAAE,gBAAgB,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,oBAAoB,EAAE;IAE/H,sCAAsC;IACtC,EAAE,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,yBAAyB,EAAE;IACvG,EAAE,EAAE,EAAE,oBAAoB,EAAE,IAAI,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,qBAAqB,EAAE;IACjH,EAAE,EAAE,EAAE,kBAAkB,EAAE,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,YAAY,EAAE;CACrG,CAAA;AAED;;GAEG;AACH,MAAM,UAAU,eAAe,CAAC,IAAe;IAC7C,OAAO,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAA;AACjD,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,CAAC,QAAgB;IAClD,OAAO,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAA;AACzD,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,YAAY,CAAC,EAAU;IACrC,MAAM,QAAQ,GAAG,OAAO,CAAC,EAAE,CAAC,CAAA;IAC5B,OAAO,GAAG,CAAC,QAAQ,CAAC,CAAA;AACtB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,eAAe,CAAC,EAAU;IACxC,MAAM,IAAI,GAAG,YAAY,CAAC,EAAE,CAAC,CAAA;IAC7B,IAAI,CAAC,IAAI,EAAE,OAAO;QAAE,OAAO,SAAS,CAAA;IAEpC,OAAO;QACL,MAAM,EAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,SAAS,EAAE,yBAAyB;QAC9E,UAAU,EAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,SAAS;KAC5D,CAAA;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAGnC;IACC,IAAI,MAAM,GAAG,WAAW,CAAA;IAExB,IAAI,IAAI,EAAE,KAAK,EAAE,CAAC;QAChB,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,KAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;IAC3D,CAAC;IAED,IAAI,IAAI,EAAE,SAAS,EAAE,CAAC;QACpB,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,SAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAA;IACnE,CAAC;IAED,OAAO,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC1B,IAAI,EAAE,GAAG,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE;QACvC,KAAK,EAAE,KAAK;KACb,CAAC,CAAC,CAAA;AACL,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,uBAAuB,CAAC,IAAe;IACrD,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAC9B,MAAM,MAAM,GAAgB,EAAE,CAAA;IAE9B,KAAK,MAAM,KAAK,IAAI,WAAW,EAAE,CAAC;QAChC,IAAI,KAAK,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACrD,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAA;YACxB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;QACpB,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAA;AACf,CAAC"}
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Simple eval runner for AI Functions
|
|
3
|
+
*
|
|
4
|
+
* Runs evals across multiple models and collects results.
|
|
5
|
+
* Does not depend on evalite - uses our own infrastructure.
|
|
6
|
+
*/
|
|
7
|
+
import { generateObject, generateText } from '../generate.js';
|
|
8
|
+
import { schema } from '../schema.js';
|
|
9
|
+
import { type EvalModel, type ModelTier } from './models.js';
|
|
10
|
+
export interface EvalCase<TInput = unknown, TExpected = unknown> {
|
|
11
|
+
name: string;
|
|
12
|
+
input: TInput;
|
|
13
|
+
expected?: TExpected;
|
|
14
|
+
}
|
|
15
|
+
export interface EvalScore {
|
|
16
|
+
name: string;
|
|
17
|
+
score: number;
|
|
18
|
+
description?: string;
|
|
19
|
+
metadata?: unknown;
|
|
20
|
+
}
|
|
21
|
+
export interface EvalResult<TOutput = unknown> {
|
|
22
|
+
model: EvalModel;
|
|
23
|
+
case: EvalCase;
|
|
24
|
+
output: TOutput;
|
|
25
|
+
scores: EvalScore[];
|
|
26
|
+
latencyMs: number;
|
|
27
|
+
cost: number;
|
|
28
|
+
error?: string;
|
|
29
|
+
}
|
|
30
|
+
export interface EvalSummary {
|
|
31
|
+
name: string;
|
|
32
|
+
results: EvalResult[];
|
|
33
|
+
avgScore: number;
|
|
34
|
+
byModel: Record<string, {
|
|
35
|
+
avgScore: number;
|
|
36
|
+
count: number;
|
|
37
|
+
}>;
|
|
38
|
+
totalCost: number;
|
|
39
|
+
totalTime: number;
|
|
40
|
+
}
|
|
41
|
+
export interface RunEvalOptions<TInput, TOutput, TExpected> {
|
|
42
|
+
name: string;
|
|
43
|
+
cases: EvalCase<TInput, TExpected>[];
|
|
44
|
+
task: (input: TInput, model: EvalModel) => Promise<TOutput>;
|
|
45
|
+
scorers: Array<{
|
|
46
|
+
name: string;
|
|
47
|
+
description?: string;
|
|
48
|
+
scorer: (args: {
|
|
49
|
+
input: TInput;
|
|
50
|
+
output: TOutput;
|
|
51
|
+
expected?: TExpected;
|
|
52
|
+
}) => number | Promise<number>;
|
|
53
|
+
}>;
|
|
54
|
+
models?: EvalModel[];
|
|
55
|
+
tiers?: ModelTier[];
|
|
56
|
+
providers?: string[];
|
|
57
|
+
concurrency?: number;
|
|
58
|
+
}
|
|
59
|
+
/**
|
|
60
|
+
* Run an eval suite across models
|
|
61
|
+
*/
|
|
62
|
+
export declare function runEval<TInput, TOutput, TExpected>(options: RunEvalOptions<TInput, TOutput, TExpected>): Promise<EvalSummary>;
|
|
63
|
+
export { generateObject, generateText, schema };
|
|
64
|
+
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -0,0 +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;AAElG,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,MAAM,EAAE,OAAO,CAAA;IACf,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;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,MAAM,EAAE,OAAO,CAAC;YAAC,QAAQ,CAAC,EAAE,SAAS,CAAA;SAAE,KAAK,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;KACrG,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;CACrB;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,CAsJtB;AAGD,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,EAAE,CAAA"}
|
|
@@ -0,0 +1,148 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Simple eval runner for AI Functions
|
|
3
|
+
*
|
|
4
|
+
* Runs evals across multiple models and collects results.
|
|
5
|
+
* Does not depend on evalite - uses our own infrastructure.
|
|
6
|
+
*/
|
|
7
|
+
import { generateObject, generateText } from '../generate.js';
|
|
8
|
+
import { schema } from '../schema.js';
|
|
9
|
+
import { createModelVariants, getModelPricing } from './models.js';
|
|
10
|
+
/**
|
|
11
|
+
* Run an eval suite across models
|
|
12
|
+
*/
|
|
13
|
+
export async function runEval(options) {
|
|
14
|
+
const { name, cases, task, scorers, concurrency = 3 } = options;
|
|
15
|
+
// Get models to test
|
|
16
|
+
const models = options.models ?? createModelVariants({
|
|
17
|
+
tiers: options.tiers,
|
|
18
|
+
providers: options.providers,
|
|
19
|
+
}).map(v => v.input);
|
|
20
|
+
const results = [];
|
|
21
|
+
const startTime = Date.now();
|
|
22
|
+
console.log(`\n🧪 Running eval: ${name}`);
|
|
23
|
+
console.log(` Models: ${models.map(m => m.name).join(', ')}`);
|
|
24
|
+
console.log(` Cases: ${cases.length}`);
|
|
25
|
+
console.log('');
|
|
26
|
+
// Run all model/case combinations
|
|
27
|
+
const jobs = [];
|
|
28
|
+
for (const model of models) {
|
|
29
|
+
for (const evalCase of cases) {
|
|
30
|
+
jobs.push({ model, case: evalCase });
|
|
31
|
+
}
|
|
32
|
+
}
|
|
33
|
+
// Process in batches with concurrency limit
|
|
34
|
+
for (let i = 0; i < jobs.length; i += concurrency) {
|
|
35
|
+
const batch = jobs.slice(i, i + concurrency);
|
|
36
|
+
const batchResults = await Promise.all(batch.map(async (job) => {
|
|
37
|
+
const caseStart = Date.now();
|
|
38
|
+
try {
|
|
39
|
+
// Run the task
|
|
40
|
+
const output = await task(job.case.input, job.model);
|
|
41
|
+
const latencyMs = Date.now() - caseStart;
|
|
42
|
+
// Run scorers
|
|
43
|
+
const scores = [];
|
|
44
|
+
for (const s of scorers) {
|
|
45
|
+
try {
|
|
46
|
+
const score = await s.scorer({
|
|
47
|
+
input: job.case.input,
|
|
48
|
+
output,
|
|
49
|
+
expected: job.case.expected,
|
|
50
|
+
});
|
|
51
|
+
scores.push({
|
|
52
|
+
name: s.name,
|
|
53
|
+
score: Math.max(0, Math.min(1, score)),
|
|
54
|
+
description: s.description,
|
|
55
|
+
});
|
|
56
|
+
}
|
|
57
|
+
catch (err) {
|
|
58
|
+
scores.push({
|
|
59
|
+
name: s.name,
|
|
60
|
+
score: 0,
|
|
61
|
+
description: s.description,
|
|
62
|
+
metadata: { error: String(err) },
|
|
63
|
+
});
|
|
64
|
+
}
|
|
65
|
+
}
|
|
66
|
+
// Calculate cost
|
|
67
|
+
const pricing = getModelPricing(job.model.id);
|
|
68
|
+
// Estimate tokens - rough approximation
|
|
69
|
+
const estimatedPromptTokens = 100;
|
|
70
|
+
const estimatedCompletionTokens = 200;
|
|
71
|
+
const cost = pricing
|
|
72
|
+
? (estimatedPromptTokens * pricing.prompt + estimatedCompletionTokens * pricing.completion) / 1_000_000
|
|
73
|
+
: 0;
|
|
74
|
+
const avgScore = scores.length > 0
|
|
75
|
+
? scores.reduce((sum, s) => sum + s.score, 0) / scores.length
|
|
76
|
+
: 0;
|
|
77
|
+
const symbol = avgScore >= 0.8 ? '✓' : avgScore >= 0.5 ? '~' : '✗';
|
|
78
|
+
console.log(` ${symbol} ${job.model.name} | ${job.case.name} | ${(avgScore * 100).toFixed(0)}% | ${latencyMs}ms`);
|
|
79
|
+
return {
|
|
80
|
+
model: job.model,
|
|
81
|
+
case: job.case,
|
|
82
|
+
output,
|
|
83
|
+
scores,
|
|
84
|
+
latencyMs,
|
|
85
|
+
cost,
|
|
86
|
+
};
|
|
87
|
+
}
|
|
88
|
+
catch (err) {
|
|
89
|
+
console.log(` ✗ ${job.model.name} | ${job.case.name} | ERROR: ${err}`);
|
|
90
|
+
return {
|
|
91
|
+
model: job.model,
|
|
92
|
+
case: job.case,
|
|
93
|
+
output: null,
|
|
94
|
+
scores: scorers.map(s => ({ name: s.name, score: 0 })),
|
|
95
|
+
latencyMs: Date.now() - caseStart,
|
|
96
|
+
cost: 0,
|
|
97
|
+
error: String(err),
|
|
98
|
+
};
|
|
99
|
+
}
|
|
100
|
+
}));
|
|
101
|
+
results.push(...batchResults);
|
|
102
|
+
}
|
|
103
|
+
// Calculate summary
|
|
104
|
+
const totalTime = Date.now() - startTime;
|
|
105
|
+
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;
|
|
110
|
+
// Group by model
|
|
111
|
+
const byModel = {};
|
|
112
|
+
for (const result of results) {
|
|
113
|
+
const modelKey = result.model.id;
|
|
114
|
+
if (!byModel[modelKey]) {
|
|
115
|
+
byModel[modelKey] = { avgScore: 0, count: 0 };
|
|
116
|
+
}
|
|
117
|
+
const resultAvg = result.scores.reduce((sum, s) => sum + s.score, 0) / result.scores.length;
|
|
118
|
+
byModel[modelKey].avgScore += resultAvg;
|
|
119
|
+
byModel[modelKey].count++;
|
|
120
|
+
}
|
|
121
|
+
for (const key of Object.keys(byModel)) {
|
|
122
|
+
const entry = byModel[key];
|
|
123
|
+
if (entry) {
|
|
124
|
+
entry.avgScore /= entry.count;
|
|
125
|
+
}
|
|
126
|
+
}
|
|
127
|
+
console.log('');
|
|
128
|
+
console.log(`📊 Results:`);
|
|
129
|
+
console.log(` Overall: ${(avgScore * 100).toFixed(1)}%`);
|
|
130
|
+
console.log(` Time: ${(totalTime / 1000).toFixed(1)}s`);
|
|
131
|
+
console.log(` Cost: $${totalCost.toFixed(4)}`);
|
|
132
|
+
console.log('');
|
|
133
|
+
console.log(' By Model:');
|
|
134
|
+
for (const [modelId, stats] of Object.entries(byModel)) {
|
|
135
|
+
console.log(` - ${modelId}: ${(stats.avgScore * 100).toFixed(1)}%`);
|
|
136
|
+
}
|
|
137
|
+
return {
|
|
138
|
+
name,
|
|
139
|
+
results,
|
|
140
|
+
avgScore,
|
|
141
|
+
byModel,
|
|
142
|
+
totalCost,
|
|
143
|
+
totalTime,
|
|
144
|
+
};
|
|
145
|
+
}
|
|
146
|
+
// Re-export helpers
|
|
147
|
+
export { generateObject, generateText, schema };
|
|
148
|
+
//# sourceMappingURL=runner.js.map
|
|
@@ -0,0 +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;AAiDlG;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,OAAmD;IAEnD,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,WAAW,GAAG,CAAC,EAAE,GAAG,OAAO,CAAA;IAE/D,qBAAqB;IACrB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,IAAI,mBAAmB,CAAC;QACnD,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,SAAS,EAAE,OAAO,CAAC,SAAS;KAC7B,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAA;IAEpB,MAAM,OAAO,GAA0B,EAAE,CAAA;IACzC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;IAE5B,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,EAAE,CAAC,CAAA;IACzC,OAAO,CAAC,GAAG,CAAC,cAAc,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IAC/D,OAAO,CAAC,GAAG,CAAC,aAAa,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;IACxC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAA;IAEf,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,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,CAAA;gBACpD,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;4BACN,QAAQ,EAAE,GAAG,CAAC,IAAI,CAAC,QAAQ;yBAC5B,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,WAAW,EAAE,CAAC,CAAC,WAAW;yBAC3B,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,WAAW,EAAE,CAAC,CAAC,WAAW;4BAC1B,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,GAAG,yBAAyB,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG,SAAS;oBACvG,CAAC,CAAC,CAAC,CAAA;gBAEL,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC;oBAChC,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;oBAC7D,CAAC,CAAC,CAAC,CAAA;gBAEL,MAAM,MAAM,GAAG,QAAQ,IAAI,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,IAAI,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAA;gBAClE,OAAO,CAAC,GAAG,CAAC,MAAM,MAAM,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,SAAS,IAAI,CAAC,CAAA;gBAEnH,OAAO;oBACL,KAAK,EAAE,GAAG,CAAC,KAAK;oBAChB,IAAI,EAAE,GAAG,CAAC,IAAI;oBACd,MAAM;oBACN,MAAM;oBACN,SAAS;oBACT,IAAI;iBACL,CAAA;YACH,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO,CAAC,GAAG,CAAC,QAAQ,GAAG,CAAC,KAAK,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,aAAa,GAAG,EAAE,CAAC,CAAA;gBAExE,OAAO;oBACL,KAAK,EAAE,GAAG,CAAC,KAAK;oBAChB,IAAI,EAAE,GAAG,CAAC,IAAI;oBACd,MAAM,EAAE,IAA0B;oBAClC,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;oBACtD,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,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAA;IAClE,MAAM,QAAQ,GAAG,SAAS,CAAC,MAAM,GAAG,CAAC;QACnC,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM;QACzD,CAAC,CAAC,CAAC,CAAA;IAEL,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,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAA;IACf,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAA;IAC1B,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IAC1D,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,SAAS,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IACzD,OAAO,CAAC,GAAG,CAAC,aAAa,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IAChD,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAA;IACf,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAA;IAC3B,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACvD,OAAO,CAAC,GAAG,CAAC,QAAQ,OAAO,KAAK,CAAC,KAAK,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IACvE,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 +1,168 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* AI Generation functions with automatic model resolution and routing
|
|
3
|
+
*
|
|
4
|
+
* Wraps AI SDK generateObject and generateText with smart model routing:
|
|
5
|
+
* - Simple aliases: 'opus', 'sonnet', 'gpt-4o'
|
|
6
|
+
* - Full IDs: 'anthropic/claude-sonnet-4.5'
|
|
7
|
+
* - Auto-routes to native SDKs for openai/anthropic/google
|
|
8
|
+
*
|
|
9
|
+
* @packageDocumentation
|
|
10
|
+
*/
|
|
11
|
+
import { generateText as sdkGenerateText, streamText as sdkStreamText, type GenerateObjectResult, type StreamObjectResult, type LanguageModel } from 'ai';
|
|
12
|
+
type ModelArg = string | LanguageModel;
|
|
13
|
+
interface GenerateObjectOptions<T> {
|
|
14
|
+
model: ModelArg;
|
|
15
|
+
schema: T;
|
|
16
|
+
prompt?: string;
|
|
17
|
+
messages?: Array<{
|
|
18
|
+
role: 'user' | 'assistant' | 'system';
|
|
19
|
+
content: string;
|
|
20
|
+
}>;
|
|
21
|
+
system?: string;
|
|
22
|
+
mode?: 'auto' | 'json' | 'tool';
|
|
23
|
+
maxTokens?: number;
|
|
24
|
+
temperature?: number;
|
|
25
|
+
topP?: number;
|
|
26
|
+
topK?: number;
|
|
27
|
+
presencePenalty?: number;
|
|
28
|
+
frequencyPenalty?: number;
|
|
29
|
+
seed?: number;
|
|
30
|
+
maxRetries?: number;
|
|
31
|
+
abortSignal?: AbortSignal;
|
|
32
|
+
headers?: Record<string, string>;
|
|
33
|
+
experimental_telemetry?: {
|
|
34
|
+
isEnabled?: boolean;
|
|
35
|
+
functionId?: string;
|
|
36
|
+
metadata?: Record<string, string>;
|
|
37
|
+
};
|
|
38
|
+
}
|
|
39
|
+
interface GenerateTextOptions {
|
|
40
|
+
model: ModelArg;
|
|
41
|
+
prompt?: string;
|
|
42
|
+
messages?: Array<{
|
|
43
|
+
role: 'user' | 'assistant' | 'system';
|
|
44
|
+
content: string;
|
|
45
|
+
}>;
|
|
46
|
+
system?: string;
|
|
47
|
+
maxTokens?: number;
|
|
48
|
+
temperature?: number;
|
|
49
|
+
topP?: number;
|
|
50
|
+
topK?: number;
|
|
51
|
+
presencePenalty?: number;
|
|
52
|
+
frequencyPenalty?: number;
|
|
53
|
+
seed?: number;
|
|
54
|
+
maxRetries?: number;
|
|
55
|
+
abortSignal?: AbortSignal;
|
|
56
|
+
headers?: Record<string, string>;
|
|
57
|
+
tools?: Record<string, unknown>;
|
|
58
|
+
toolChoice?: 'auto' | 'none' | 'required' | {
|
|
59
|
+
type: 'tool';
|
|
60
|
+
toolName: string;
|
|
61
|
+
};
|
|
62
|
+
maxSteps?: number;
|
|
63
|
+
experimental_telemetry?: {
|
|
64
|
+
isEnabled?: boolean;
|
|
65
|
+
functionId?: string;
|
|
66
|
+
metadata?: Record<string, string>;
|
|
67
|
+
};
|
|
68
|
+
}
|
|
69
|
+
/**
|
|
70
|
+
* Generate a typed object from a prompt using AI
|
|
71
|
+
*
|
|
72
|
+
* Automatically resolves model aliases and routes to the best provider.
|
|
73
|
+
* Supports both Zod schemas and simplified schema syntax.
|
|
74
|
+
*
|
|
75
|
+
* @example
|
|
76
|
+
* ```ts
|
|
77
|
+
* import { generateObject } from 'ai-functions'
|
|
78
|
+
*
|
|
79
|
+
* // Simplified schema syntax
|
|
80
|
+
* const { object } = await generateObject({
|
|
81
|
+
* model: 'sonnet',
|
|
82
|
+
* schema: {
|
|
83
|
+
* recipe: {
|
|
84
|
+
* name: 'What is the recipe name?',
|
|
85
|
+
* type: 'food | drink | dessert',
|
|
86
|
+
* ingredients: ['List all ingredients'],
|
|
87
|
+
* steps: ['List all cooking steps'],
|
|
88
|
+
* },
|
|
89
|
+
* },
|
|
90
|
+
* prompt: 'Generate a lasagna recipe.',
|
|
91
|
+
* })
|
|
92
|
+
*
|
|
93
|
+
* // Zod schema also works
|
|
94
|
+
* import { z } from 'zod'
|
|
95
|
+
* const { object } = await generateObject({
|
|
96
|
+
* model: 'sonnet',
|
|
97
|
+
* schema: z.object({
|
|
98
|
+
* name: z.string(),
|
|
99
|
+
* ingredients: z.array(z.string()),
|
|
100
|
+
* }),
|
|
101
|
+
* prompt: 'Generate a lasagna recipe.',
|
|
102
|
+
* })
|
|
103
|
+
* ```
|
|
104
|
+
*/
|
|
105
|
+
export declare function generateObject<T>(options: GenerateObjectOptions<T>): Promise<GenerateObjectResult<T>>;
|
|
106
|
+
/**
|
|
107
|
+
* Generate text from a prompt using AI
|
|
108
|
+
*
|
|
109
|
+
* Automatically resolves model aliases and routes to the best provider.
|
|
110
|
+
*
|
|
111
|
+
* @example
|
|
112
|
+
* ```ts
|
|
113
|
+
* import { generateText } from 'ai-functions'
|
|
114
|
+
*
|
|
115
|
+
* const { text } = await generateText({
|
|
116
|
+
* model: 'opus', // → anthropic/claude-opus-4.5
|
|
117
|
+
* prompt: 'Write a haiku about programming.',
|
|
118
|
+
* })
|
|
119
|
+
*
|
|
120
|
+
* // With tools
|
|
121
|
+
* const { text, toolResults } = await generateText({
|
|
122
|
+
* model: 'gpt-4o', // → openai/gpt-4o
|
|
123
|
+
* prompt: 'What is the weather in San Francisco?',
|
|
124
|
+
* tools: { ... },
|
|
125
|
+
* maxSteps: 5,
|
|
126
|
+
* })
|
|
127
|
+
* ```
|
|
128
|
+
*/
|
|
129
|
+
export declare function generateText(options: GenerateTextOptions): Promise<Awaited<ReturnType<typeof sdkGenerateText>>>;
|
|
130
|
+
/**
|
|
131
|
+
* Stream a typed object from a prompt using AI
|
|
132
|
+
*
|
|
133
|
+
* @example
|
|
134
|
+
* ```ts
|
|
135
|
+
* import { streamObject } from 'ai-functions'
|
|
136
|
+
*
|
|
137
|
+
* const { partialObjectStream } = streamObject({
|
|
138
|
+
* model: 'sonnet',
|
|
139
|
+
* schema: { story: 'Write a creative story' },
|
|
140
|
+
* prompt: 'Write a short story.',
|
|
141
|
+
* })
|
|
142
|
+
*
|
|
143
|
+
* for await (const partial of partialObjectStream) {
|
|
144
|
+
* console.log(partial.story)
|
|
145
|
+
* }
|
|
146
|
+
* ```
|
|
147
|
+
*/
|
|
148
|
+
export declare function streamObject<T>(options: GenerateObjectOptions<T>): Promise<StreamObjectResult<T, T, never>>;
|
|
149
|
+
/**
|
|
150
|
+
* Stream text from a prompt using AI
|
|
151
|
+
*
|
|
152
|
+
* @example
|
|
153
|
+
* ```ts
|
|
154
|
+
* import { streamText } from 'ai-functions'
|
|
155
|
+
*
|
|
156
|
+
* const { textStream } = streamText({
|
|
157
|
+
* model: 'gemini', // → google/gemini-2.5-flash
|
|
158
|
+
* prompt: 'Explain quantum computing.',
|
|
159
|
+
* })
|
|
160
|
+
*
|
|
161
|
+
* for await (const chunk of textStream) {
|
|
162
|
+
* process.stdout.write(chunk)
|
|
163
|
+
* }
|
|
164
|
+
* ```
|
|
165
|
+
*/
|
|
166
|
+
export declare function streamText(options: GenerateTextOptions): Promise<ReturnType<typeof sdkStreamText>>;
|
|
167
|
+
export {};
|
|
168
|
+
//# sourceMappingURL=generate.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"generate.d.ts","sourceRoot":"","sources":["../src/generate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,EAEL,YAAY,IAAI,eAAe,EAE/B,UAAU,IAAI,aAAa,EAC3B,KAAK,oBAAoB,EAEzB,KAAK,kBAAkB,EAEvB,KAAK,aAAa,EACnB,MAAM,IAAI,CAAA;AAIX,KAAK,QAAQ,GAAG,MAAM,GAAG,aAAa,CAAA;AAGtC,UAAU,qBAAqB,CAAC,CAAC;IAC/B,KAAK,EAAE,QAAQ,CAAA;IACf,MAAM,EAAE,CAAC,CAAA;IACT,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC5E,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,IAAI,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAA;IAC/B,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,CAAC,EAAE,WAAW,CAAA;IACzB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,sBAAsB,CAAC,EAAE;QAAE,SAAS,CAAC,EAAE,OAAO,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,CAAA;CACzG;AAED,UAAU,mBAAmB;IAC3B,KAAK,EAAE,QAAQ,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC5E,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,CAAC,EAAE,WAAW,CAAA;IACzB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC/B,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,UAAU,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAA;IAC9E,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,sBAAsB,CAAC,EAAE;QAAE,SAAS,CAAC,EAAE,OAAO,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,CAAA;CACzG;AAqCD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AACH,wBAAsB,cAAc,CAAC,CAAC,EACpC,OAAO,EAAE,qBAAqB,CAAC,CAAC,CAAC,GAChC,OAAO,CAAC,oBAAoB,CAAC,CAAC,CAAC,CAAC,CAUlC;AAED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAsB,YAAY,CAChC,OAAO,EAAE,mBAAmB,GAC3B,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,OAAO,eAAe,CAAC,CAAC,CAAC,CAMtD;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,YAAY,CAAC,CAAC,EAClC,OAAO,EAAE,qBAAqB,CAAC,CAAC,CAAC,GAChC,OAAO,CAAC,kBAAkB,CAAC,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,CAU1C;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAsB,UAAU,CAC9B,OAAO,EAAE,mBAAmB,GAC3B,OAAO,CAAC,UAAU,CAAC,OAAO,aAAa,CAAC,CAAC,CAM3C"}
|