promptfoo 0.68.3 → 0.69.1
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/dist/drizzle/0002_tidy_juggernaut.sql +1 -0
- package/dist/drizzle/meta/0002_snapshot.json +244 -0
- package/dist/drizzle/meta/_journal.json +7 -0
- package/dist/package.json +18 -17
- package/dist/src/__mocks__/database.d.ts +0 -1
- package/dist/src/__mocks__/database.d.ts.map +1 -1
- package/dist/src/__mocks__/esm.js +2 -3
- package/dist/src/__mocks__/esm.js.map +1 -1
- package/dist/src/accounts.d.ts +2 -1
- package/dist/src/accounts.d.ts.map +1 -1
- package/dist/src/accounts.js +7 -4
- package/dist/src/accounts.js.map +1 -1
- package/dist/src/assertions/AssertionsResult.js +3 -3
- package/dist/src/assertions/AssertionsResult.js.map +1 -1
- package/dist/src/assertions/validateAssertions.js +2 -2
- package/dist/src/assertions/validateAssertions.js.map +1 -1
- package/dist/src/assertions.js +6 -6
- package/dist/src/assertions.js.map +1 -1
- package/dist/src/cache.d.ts.map +1 -1
- package/dist/src/cache.js +6 -7
- package/dist/src/cache.js.map +1 -1
- package/dist/src/commands/config.js +1 -2
- package/dist/src/commands/config.js.map +1 -1
- package/dist/src/commands/delete.js +1 -2
- package/dist/src/commands/delete.js.map +1 -1
- package/dist/src/commands/eval/filterFailingTests.js +1 -2
- package/dist/src/commands/eval/filterFailingTests.js.map +1 -1
- package/dist/src/commands/eval/filterProviders.d.ts +3 -0
- package/dist/src/commands/eval/filterProviders.d.ts.map +1 -0
- package/dist/src/commands/eval/filterProviders.js +15 -0
- package/dist/src/commands/eval/filterProviders.js.map +1 -0
- package/dist/src/commands/eval/filterTests.js +1 -2
- package/dist/src/commands/eval/filterTests.js.map +1 -1
- package/dist/src/commands/export.js +1 -2
- package/dist/src/commands/export.js.map +1 -1
- package/dist/src/commands/import.d.ts.map +1 -1
- package/dist/src/commands/import.js +2 -2
- package/dist/src/commands/import.js.map +1 -1
- package/dist/src/commands/list.js +1 -2
- package/dist/src/commands/list.js.map +1 -1
- package/dist/src/commands/show.js +3 -4
- package/dist/src/commands/show.js.map +1 -1
- package/dist/src/csv.js +2 -3
- package/dist/src/csv.js.map +1 -1
- package/dist/src/database.d.ts +12 -0
- package/dist/src/database.d.ts.map +1 -1
- package/dist/src/database.js +5 -4
- package/dist/src/database.js.map +1 -1
- package/dist/src/esm.js +2 -3
- package/dist/src/esm.js.map +1 -1
- package/dist/src/evaluator.d.ts.map +1 -1
- package/dist/src/evaluator.js +30 -17
- package/dist/src/evaluator.js.map +1 -1
- package/dist/src/feedback.js +2 -3
- package/dist/src/feedback.js.map +1 -1
- package/dist/src/fetch.d.ts +2 -3
- package/dist/src/fetch.d.ts.map +1 -1
- package/dist/src/fetch.js +7 -6
- package/dist/src/fetch.js.map +1 -1
- package/dist/src/globalConfig.js +5 -6
- package/dist/src/globalConfig.js.map +1 -1
- package/dist/src/googleSheets.js +4 -5
- package/dist/src/googleSheets.js.map +1 -1
- package/dist/src/index.js +2 -2
- package/dist/src/index.js.map +1 -1
- package/dist/src/integrations/langfuse.js +1 -2
- package/dist/src/integrations/langfuse.js.map +1 -1
- package/dist/src/integrations/portkey.js +1 -2
- package/dist/src/integrations/portkey.js.map +1 -1
- package/dist/src/logger.js +3 -3
- package/dist/src/logger.js.map +1 -1
- package/dist/src/main.js +23 -1
- package/dist/src/main.js.map +1 -1
- package/dist/src/matchers.d.ts.map +1 -1
- package/dist/src/matchers.js +16 -15
- package/dist/src/matchers.js.map +1 -1
- package/dist/src/migrate.js +1 -2
- package/dist/src/migrate.js.map +1 -1
- package/dist/src/onboarding.js +2 -2
- package/dist/src/onboarding.js.map +1 -1
- package/dist/src/prompts/grading.d.ts +1 -0
- package/dist/src/prompts/grading.d.ts.map +1 -1
- package/dist/src/prompts/grading.js +21 -1
- package/dist/src/prompts/grading.js.map +1 -1
- package/dist/src/prompts/index.js +3 -4
- package/dist/src/prompts/index.js.map +1 -1
- package/dist/src/prompts/processors/javascript.d.ts +5 -1
- package/dist/src/prompts/processors/javascript.d.ts.map +1 -1
- package/dist/src/prompts/processors/javascript.js +15 -3
- package/dist/src/prompts/processors/javascript.js.map +1 -1
- package/dist/src/prompts/processors/json.js +1 -2
- package/dist/src/prompts/processors/json.js.map +1 -1
- package/dist/src/prompts/processors/jsonl.js +1 -2
- package/dist/src/prompts/processors/jsonl.js.map +1 -1
- package/dist/src/prompts/processors/python.d.ts +1 -1
- package/dist/src/prompts/processors/python.d.ts.map +1 -1
- package/dist/src/prompts/processors/python.js +24 -18
- package/dist/src/prompts/processors/python.js.map +1 -1
- package/dist/src/prompts/processors/string.js +1 -2
- package/dist/src/prompts/processors/string.js.map +1 -1
- package/dist/src/prompts/processors/text.js +1 -2
- package/dist/src/prompts/processors/text.js.map +1 -1
- package/dist/src/prompts/processors/yaml.js +1 -2
- package/dist/src/prompts/processors/yaml.js.map +1 -1
- package/dist/src/prompts/utils.js +2 -3
- package/dist/src/prompts/utils.js.map +1 -1
- package/dist/src/providers/anthropic.d.ts +5 -0
- package/dist/src/providers/anthropic.d.ts.map +1 -1
- package/dist/src/providers/anthropic.js +58 -4
- package/dist/src/providers/anthropic.js.map +1 -1
- package/dist/src/providers/azureopenaiUtil.js +1 -2
- package/dist/src/providers/azureopenaiUtil.js.map +1 -1
- package/dist/src/providers/bedrock.d.ts +3 -3
- package/dist/src/providers/bedrock.d.ts.map +1 -1
- package/dist/src/providers/bedrock.js +3 -3
- package/dist/src/providers/bedrock.js.map +1 -1
- package/dist/src/providers/defaults.d.ts +3 -0
- package/dist/src/providers/defaults.d.ts.map +1 -1
- package/dist/src/providers/defaults.js +2 -2
- package/dist/src/providers/defaults.js.map +1 -1
- package/dist/src/providers/openai.d.ts.map +1 -1
- package/dist/src/providers/openai.js +1 -0
- package/dist/src/providers/openai.js.map +1 -1
- package/dist/src/providers/openaiUtil.js +1 -2
- package/dist/src/providers/openaiUtil.js.map +1 -1
- package/dist/src/providers/pythonCompletion.d.ts.map +1 -1
- package/dist/src/providers/scriptCompletion.d.ts.map +1 -1
- package/dist/src/providers/shared.js +3 -3
- package/dist/src/providers/shared.js.map +1 -1
- package/dist/src/providers/vertexUtil.js +3 -4
- package/dist/src/providers/vertexUtil.js.map +1 -1
- package/dist/src/providers.d.ts.map +1 -1
- package/dist/src/providers.js +5 -7
- package/dist/src/providers.js.map +1 -1
- package/dist/src/python/wrapper.d.ts +1 -1
- package/dist/src/python/wrapper.d.ts.map +1 -1
- package/dist/src/python/wrapper.js +2 -3
- package/dist/src/python/wrapper.js.map +1 -1
- package/dist/src/python/wrapper.py +9 -10
- package/dist/src/redteam/getCompetitorTests.js +1 -2
- package/dist/src/redteam/getCompetitorTests.js.map +1 -1
- package/dist/src/redteam/getHallucinationTests.js +1 -2
- package/dist/src/redteam/getHallucinationTests.js.map +1 -1
- package/dist/src/redteam/getHarmfulTests.d.ts +16 -16
- package/dist/src/redteam/getHarmfulTests.d.ts.map +1 -1
- package/dist/src/redteam/getHarmfulTests.js +6 -6
- package/dist/src/redteam/getHarmfulTests.js.map +1 -1
- package/dist/src/redteam/getHijackingTests.js +1 -2
- package/dist/src/redteam/getHijackingTests.js.map +1 -1
- package/dist/src/redteam/getOverconfidenceTests.js +1 -2
- package/dist/src/redteam/getOverconfidenceTests.js.map +1 -1
- package/dist/src/redteam/getPiiTests.js +1 -2
- package/dist/src/redteam/getPiiTests.js.map +1 -1
- package/dist/src/redteam/getPoliticalStatementsTests.js +1 -2
- package/dist/src/redteam/getPoliticalStatementsTests.js.map +1 -1
- package/dist/src/redteam/getUnderconfidenceTests.js +2 -3
- package/dist/src/redteam/getUnderconfidenceTests.js.map +1 -1
- package/dist/src/redteam/getUnintendedContractTests.js +1 -2
- package/dist/src/redteam/getUnintendedContractTests.js.map +1 -1
- package/dist/src/redteam/index.d.ts.map +1 -1
- package/dist/src/redteam/index.js +28 -9
- package/dist/src/redteam/index.js.map +1 -1
- package/dist/src/redteam/iterative.d.ts.map +1 -1
- package/dist/src/redteam/iterative.js +15 -2
- package/dist/src/redteam/iterative.js.map +1 -1
- package/dist/src/share.d.ts.map +1 -1
- package/dist/src/share.js +4 -3
- package/dist/src/share.js.map +1 -1
- package/dist/src/suggestions.js +1 -2
- package/dist/src/suggestions.js.map +1 -1
- package/dist/src/table.js +2 -3
- package/dist/src/table.js.map +1 -1
- package/dist/src/testCases.js +6 -7
- package/dist/src/testCases.js.map +1 -1
- package/dist/src/types.d.ts +951 -137
- package/dist/src/types.d.ts.map +1 -1
- package/dist/src/types.js +196 -4
- package/dist/src/types.js.map +1 -1
- package/dist/src/updates.js +2 -3
- package/dist/src/updates.js.map +1 -1
- package/dist/src/util.d.ts.map +1 -1
- package/dist/src/util.js +67 -56
- package/dist/src/util.js.map +1 -1
- package/dist/src/web/nextui/404/index.html +1 -1
- package/dist/src/web/nextui/404.html +1 -1
- package/dist/src/web/nextui/_next/static/chunks/2-671ad31c05d2c976.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/{540-378a20144435cec6.js → 540-c94eb8fa0d9db136.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/{166-157bfb431b68d949.js → 858-5d3a3678769b7e36.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/{897-1955b232a2148365.js → 998-d370cd9f4473ff31.js} +4 -4
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/{page-0a71ecaf651124c2.js → page-3c5a944373865122.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/{page-bf3968ed106645c6.js → page-73e894c39cc191f1.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/{layout-b7b366d291dbc9db.js → layout-2038906de6c19565.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/progress/page-73442c531d579c51.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/prompts/page-50e27c24c9e255bd.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/setup/{page-71b3060b06865254.js → page-26cb5d2478fdbd34.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/webpack-8a9bc9ee0defb756.js +1 -0
- package/dist/src/web/nextui/_next/static/css/106779eb64615639.css +1 -0
- package/dist/src/web/nextui/_next/static/css/35ab7ce8025438b0.css +1 -0
- package/dist/src/web/nextui/_next/static/css/94b771e23fc5d5f5.css +1 -0
- package/dist/src/web/nextui/_next/static/media/05a31a2ca4975f99-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/513657b02c5c193f-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/51ed15f9841b9f9d-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/c9a5bc6a7c948fb0-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/d6b16ce4a6175f26-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/ec159349637c90ad-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/fd4db3eb5472fc27-s.woff2 +0 -0
- package/dist/src/web/nextui/auth/login/index.html +1 -1
- package/dist/src/web/nextui/auth/login/index.txt +10 -15
- package/dist/src/web/nextui/auth/signup/index.html +1 -1
- package/dist/src/web/nextui/auth/signup/index.txt +10 -15
- package/dist/src/web/nextui/datasets/index.html +1 -1
- package/dist/src/web/nextui/datasets/index.txt +10 -15
- package/dist/src/web/nextui/eval/index.html +1 -1
- package/dist/src/web/nextui/eval/index.txt +12 -17
- package/dist/src/web/nextui/index.html +1 -1
- package/dist/src/web/nextui/index.txt +11 -16
- package/dist/src/web/nextui/progress/index.html +1 -1
- package/dist/src/web/nextui/progress/index.txt +10 -15
- package/dist/src/web/nextui/prompts/index.html +1 -1
- package/dist/src/web/nextui/prompts/index.txt +10 -15
- package/dist/src/web/nextui/report/index.html +1 -1
- package/dist/src/web/nextui/report/index.txt +12 -17
- package/dist/src/web/nextui/setup/index.html +2 -2
- package/dist/src/web/nextui/setup/index.txt +12 -17
- package/dist/src/web/server.js +2 -2
- package/dist/src/web/server.js.map +1 -1
- package/package.json +18 -17
- package/dist/src/web/nextui/_next/static/chunks/2-d534f7b3af3e0505.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/progress/page-52079a97d0c2e2ea.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/prompts/page-7da4796409bc2636.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/webpack-ac7a3125017b92ed.js +0 -1
- package/dist/src/web/nextui/_next/static/css/036bf4af64e53e86.css +0 -1
- package/dist/src/web/nextui/_next/static/css/b76f1739b5a14901.css +0 -1
- package/dist/src/web/nextui/_next/static/css/e141e895af3747c6.css +0 -1
- package/dist/src/web/nextui/_next/static/media/03b685511c0eaac3-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/04fe87c30c4f76ea-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/0e4fe491bf84089c-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/101c7b68f2d8b610-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/13259ce60de2aa3e-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/1c57ca6f5208a29b-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/1e8103c5d17beb1d-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/388fb79562c9fc54-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/3a04115668d8070d-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/3a18fc8da1cdcd01-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/3dbd163d3bb09d47-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/42d52f46a26971a3-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/5647e4c23315a2d2-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/5f4dfe83faee04f1-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/627622453ef56b0d-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/67e05ba3467b9f4d-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/699512af39861afa-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/79015435f1a55eef-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/7be645d133f3ee22-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/7c53f7419436e04b-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/8fb72f69fba4e3d2-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/912a9cfe43c928d9-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/91a88e0c5dd21dfa-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/934c4b7cb736f2a3-s.p.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/9b67ab375515cd6f-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/9cf7d128be063d32-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/a3c201c07e8eb753-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/a5b77b63ef20339c-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/a6d330d7873e7320-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/b6db722c6886c2cd-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/baf12dd90520ae41-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/bbdb6f0234009aba-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/cff529cd86cc0276-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/d117eea74e01de14-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/daa8cca6b52a70fa-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/dfa8b99978df7bbc-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/e195dd2ded485df3-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/e25729ca87cc7df9-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/e35c7314ac518ddc-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/eb52b768f62eeeb4-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/ee7e17a7bdd8636b-s.woff2 +0 -0
- package/dist/src/web/nextui/_next/static/media/f06116e890b3dadb-s.woff2 +0 -0
- /package/dist/src/web/nextui/_next/static/{4W0Qng9llk0Tg-jZCYupr → 1r10QuQFZj3AAU-Bx1DRP}/_buildManifest.js +0 -0
- /package/dist/src/web/nextui/_next/static/{4W0Qng9llk0Tg-jZCYupr → 1r10QuQFZj3AAU-Bx1DRP}/_ssgManifest.js +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{215-7d83203d614bcd6a.js → 215-7bd08e1ce7028897.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{322-92c9551f32057648.js → 322-ac32773931b7a8aa.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{378-4b0cc74c685f98f9.js → 378-c23e2864527ac3a0.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{770-577cd7d7d380b3ac.js → 770-28c6d673e9540d80.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{873-8bd18cf80e6e64e1.js → 873-e1436a5f57769f17.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{915-cd96102a51e75c9d.js → 915-b1361d86b62c2cfe.js} +0 -0
- /package/dist/src/web/nextui/_next/static/chunks/{995-c9a656a43545902a.js → 995-81cea2f3d08297f2.js} +0 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
ALTER TABLE evals ADD `author` text;
|
|
@@ -0,0 +1,244 @@
|
|
|
1
|
+
{
|
|
2
|
+
"version": "5",
|
|
3
|
+
"dialect": "sqlite",
|
|
4
|
+
"id": "6325e41a-40cd-4db1-9e99-df139a264a14",
|
|
5
|
+
"prevId": "476c887f-edcb-4b29-b054-50a8383cc6c2",
|
|
6
|
+
"tables": {
|
|
7
|
+
"datasets": {
|
|
8
|
+
"name": "datasets",
|
|
9
|
+
"columns": {
|
|
10
|
+
"id": {
|
|
11
|
+
"name": "id",
|
|
12
|
+
"type": "text",
|
|
13
|
+
"primaryKey": true,
|
|
14
|
+
"notNull": true,
|
|
15
|
+
"autoincrement": false
|
|
16
|
+
},
|
|
17
|
+
"tests": {
|
|
18
|
+
"name": "tests",
|
|
19
|
+
"type": "text",
|
|
20
|
+
"primaryKey": false,
|
|
21
|
+
"notNull": false,
|
|
22
|
+
"autoincrement": false
|
|
23
|
+
},
|
|
24
|
+
"created_at": {
|
|
25
|
+
"name": "created_at",
|
|
26
|
+
"type": "integer",
|
|
27
|
+
"primaryKey": false,
|
|
28
|
+
"notNull": true,
|
|
29
|
+
"autoincrement": false,
|
|
30
|
+
"default": "CURRENT_TIMESTAMP"
|
|
31
|
+
}
|
|
32
|
+
},
|
|
33
|
+
"indexes": {},
|
|
34
|
+
"foreignKeys": {},
|
|
35
|
+
"compositePrimaryKeys": {},
|
|
36
|
+
"uniqueConstraints": {}
|
|
37
|
+
},
|
|
38
|
+
"evals": {
|
|
39
|
+
"name": "evals",
|
|
40
|
+
"columns": {
|
|
41
|
+
"id": {
|
|
42
|
+
"name": "id",
|
|
43
|
+
"type": "text",
|
|
44
|
+
"primaryKey": true,
|
|
45
|
+
"notNull": true,
|
|
46
|
+
"autoincrement": false
|
|
47
|
+
},
|
|
48
|
+
"created_at": {
|
|
49
|
+
"name": "created_at",
|
|
50
|
+
"type": "integer",
|
|
51
|
+
"primaryKey": false,
|
|
52
|
+
"notNull": true,
|
|
53
|
+
"autoincrement": false,
|
|
54
|
+
"default": "CURRENT_TIMESTAMP"
|
|
55
|
+
},
|
|
56
|
+
"author": {
|
|
57
|
+
"name": "author",
|
|
58
|
+
"type": "text",
|
|
59
|
+
"primaryKey": false,
|
|
60
|
+
"notNull": false,
|
|
61
|
+
"autoincrement": false
|
|
62
|
+
},
|
|
63
|
+
"description": {
|
|
64
|
+
"name": "description",
|
|
65
|
+
"type": "text",
|
|
66
|
+
"primaryKey": false,
|
|
67
|
+
"notNull": false,
|
|
68
|
+
"autoincrement": false
|
|
69
|
+
},
|
|
70
|
+
"results": {
|
|
71
|
+
"name": "results",
|
|
72
|
+
"type": "text",
|
|
73
|
+
"primaryKey": false,
|
|
74
|
+
"notNull": true,
|
|
75
|
+
"autoincrement": false
|
|
76
|
+
},
|
|
77
|
+
"config": {
|
|
78
|
+
"name": "config",
|
|
79
|
+
"type": "text",
|
|
80
|
+
"primaryKey": false,
|
|
81
|
+
"notNull": true,
|
|
82
|
+
"autoincrement": false
|
|
83
|
+
}
|
|
84
|
+
},
|
|
85
|
+
"indexes": {},
|
|
86
|
+
"foreignKeys": {},
|
|
87
|
+
"compositePrimaryKeys": {},
|
|
88
|
+
"uniqueConstraints": {}
|
|
89
|
+
},
|
|
90
|
+
"evals_to_datasets": {
|
|
91
|
+
"name": "evals_to_datasets",
|
|
92
|
+
"columns": {
|
|
93
|
+
"eval_id": {
|
|
94
|
+
"name": "eval_id",
|
|
95
|
+
"type": "text",
|
|
96
|
+
"primaryKey": false,
|
|
97
|
+
"notNull": true,
|
|
98
|
+
"autoincrement": false
|
|
99
|
+
},
|
|
100
|
+
"dataset_id": {
|
|
101
|
+
"name": "dataset_id",
|
|
102
|
+
"type": "text",
|
|
103
|
+
"primaryKey": false,
|
|
104
|
+
"notNull": true,
|
|
105
|
+
"autoincrement": false
|
|
106
|
+
}
|
|
107
|
+
},
|
|
108
|
+
"indexes": {},
|
|
109
|
+
"foreignKeys": {
|
|
110
|
+
"evals_to_datasets_eval_id_evals_id_fk": {
|
|
111
|
+
"name": "evals_to_datasets_eval_id_evals_id_fk",
|
|
112
|
+
"tableFrom": "evals_to_datasets",
|
|
113
|
+
"tableTo": "evals",
|
|
114
|
+
"columnsFrom": [
|
|
115
|
+
"eval_id"
|
|
116
|
+
],
|
|
117
|
+
"columnsTo": [
|
|
118
|
+
"id"
|
|
119
|
+
],
|
|
120
|
+
"onDelete": "no action",
|
|
121
|
+
"onUpdate": "no action"
|
|
122
|
+
},
|
|
123
|
+
"evals_to_datasets_dataset_id_datasets_id_fk": {
|
|
124
|
+
"name": "evals_to_datasets_dataset_id_datasets_id_fk",
|
|
125
|
+
"tableFrom": "evals_to_datasets",
|
|
126
|
+
"tableTo": "datasets",
|
|
127
|
+
"columnsFrom": [
|
|
128
|
+
"dataset_id"
|
|
129
|
+
],
|
|
130
|
+
"columnsTo": [
|
|
131
|
+
"id"
|
|
132
|
+
],
|
|
133
|
+
"onDelete": "no action",
|
|
134
|
+
"onUpdate": "no action"
|
|
135
|
+
}
|
|
136
|
+
},
|
|
137
|
+
"compositePrimaryKeys": {
|
|
138
|
+
"evals_to_datasets_eval_id_dataset_id_pk": {
|
|
139
|
+
"columns": [
|
|
140
|
+
"dataset_id",
|
|
141
|
+
"eval_id"
|
|
142
|
+
],
|
|
143
|
+
"name": "evals_to_datasets_eval_id_dataset_id_pk"
|
|
144
|
+
}
|
|
145
|
+
},
|
|
146
|
+
"uniqueConstraints": {}
|
|
147
|
+
},
|
|
148
|
+
"evals_to_prompts": {
|
|
149
|
+
"name": "evals_to_prompts",
|
|
150
|
+
"columns": {
|
|
151
|
+
"eval_id": {
|
|
152
|
+
"name": "eval_id",
|
|
153
|
+
"type": "text",
|
|
154
|
+
"primaryKey": false,
|
|
155
|
+
"notNull": true,
|
|
156
|
+
"autoincrement": false
|
|
157
|
+
},
|
|
158
|
+
"prompt_id": {
|
|
159
|
+
"name": "prompt_id",
|
|
160
|
+
"type": "text",
|
|
161
|
+
"primaryKey": false,
|
|
162
|
+
"notNull": true,
|
|
163
|
+
"autoincrement": false
|
|
164
|
+
}
|
|
165
|
+
},
|
|
166
|
+
"indexes": {},
|
|
167
|
+
"foreignKeys": {
|
|
168
|
+
"evals_to_prompts_eval_id_evals_id_fk": {
|
|
169
|
+
"name": "evals_to_prompts_eval_id_evals_id_fk",
|
|
170
|
+
"tableFrom": "evals_to_prompts",
|
|
171
|
+
"tableTo": "evals",
|
|
172
|
+
"columnsFrom": [
|
|
173
|
+
"eval_id"
|
|
174
|
+
],
|
|
175
|
+
"columnsTo": [
|
|
176
|
+
"id"
|
|
177
|
+
],
|
|
178
|
+
"onDelete": "no action",
|
|
179
|
+
"onUpdate": "no action"
|
|
180
|
+
},
|
|
181
|
+
"evals_to_prompts_prompt_id_prompts_id_fk": {
|
|
182
|
+
"name": "evals_to_prompts_prompt_id_prompts_id_fk",
|
|
183
|
+
"tableFrom": "evals_to_prompts",
|
|
184
|
+
"tableTo": "prompts",
|
|
185
|
+
"columnsFrom": [
|
|
186
|
+
"prompt_id"
|
|
187
|
+
],
|
|
188
|
+
"columnsTo": [
|
|
189
|
+
"id"
|
|
190
|
+
],
|
|
191
|
+
"onDelete": "no action",
|
|
192
|
+
"onUpdate": "no action"
|
|
193
|
+
}
|
|
194
|
+
},
|
|
195
|
+
"compositePrimaryKeys": {
|
|
196
|
+
"evals_to_prompts_eval_id_prompt_id_pk": {
|
|
197
|
+
"columns": [
|
|
198
|
+
"eval_id",
|
|
199
|
+
"prompt_id"
|
|
200
|
+
],
|
|
201
|
+
"name": "evals_to_prompts_eval_id_prompt_id_pk"
|
|
202
|
+
}
|
|
203
|
+
},
|
|
204
|
+
"uniqueConstraints": {}
|
|
205
|
+
},
|
|
206
|
+
"prompts": {
|
|
207
|
+
"name": "prompts",
|
|
208
|
+
"columns": {
|
|
209
|
+
"id": {
|
|
210
|
+
"name": "id",
|
|
211
|
+
"type": "text",
|
|
212
|
+
"primaryKey": true,
|
|
213
|
+
"notNull": true,
|
|
214
|
+
"autoincrement": false
|
|
215
|
+
},
|
|
216
|
+
"created_at": {
|
|
217
|
+
"name": "created_at",
|
|
218
|
+
"type": "integer",
|
|
219
|
+
"primaryKey": false,
|
|
220
|
+
"notNull": true,
|
|
221
|
+
"autoincrement": false,
|
|
222
|
+
"default": "CURRENT_TIMESTAMP"
|
|
223
|
+
},
|
|
224
|
+
"prompt": {
|
|
225
|
+
"name": "prompt",
|
|
226
|
+
"type": "text",
|
|
227
|
+
"primaryKey": false,
|
|
228
|
+
"notNull": true,
|
|
229
|
+
"autoincrement": false
|
|
230
|
+
}
|
|
231
|
+
},
|
|
232
|
+
"indexes": {},
|
|
233
|
+
"foreignKeys": {},
|
|
234
|
+
"compositePrimaryKeys": {},
|
|
235
|
+
"uniqueConstraints": {}
|
|
236
|
+
}
|
|
237
|
+
},
|
|
238
|
+
"enums": {},
|
|
239
|
+
"_meta": {
|
|
240
|
+
"schemas": {},
|
|
241
|
+
"tables": {},
|
|
242
|
+
"columns": {}
|
|
243
|
+
}
|
|
244
|
+
}
|
package/dist/package.json
CHANGED
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
"name": "promptfoo",
|
|
3
3
|
"description": "LLM eval & testing toolkit",
|
|
4
4
|
"author": "Ian Webster",
|
|
5
|
-
"version": "0.
|
|
5
|
+
"version": "0.69.1",
|
|
6
6
|
"license": "MIT",
|
|
7
7
|
"type": "commonjs",
|
|
8
8
|
"repository": {
|
|
@@ -62,18 +62,18 @@
|
|
|
62
62
|
"node-sql-parser": "^5.2.0"
|
|
63
63
|
},
|
|
64
64
|
"devDependencies": {
|
|
65
|
-
"@aws-sdk/client-bedrock-runtime": "^3.
|
|
65
|
+
"@aws-sdk/client-bedrock-runtime": "^3.609.0",
|
|
66
66
|
"@azure/identity": "^4.3.0",
|
|
67
|
-
"@eslint/js": "^9.
|
|
68
|
-
"@swc/cli": "^0.
|
|
69
|
-
"@swc/core": "^1.6.
|
|
67
|
+
"@eslint/js": "^9.6.0",
|
|
68
|
+
"@swc/cli": "^0.4.0",
|
|
69
|
+
"@swc/core": "^1.6.13",
|
|
70
70
|
"@swc/jest": "^0.2.36",
|
|
71
71
|
"@trivago/prettier-plugin-sort-imports": "^4.3.0",
|
|
72
72
|
"@types/async": "^3.2.24",
|
|
73
|
-
"@types/better-sqlite3": "^7.6.
|
|
73
|
+
"@types/better-sqlite3": "^7.6.11",
|
|
74
74
|
"@types/cache-manager": "^4.0.6",
|
|
75
75
|
"@types/cache-manager-fs-hash": "^0.0.4",
|
|
76
|
-
"@types/cli-progress": "^3.11.
|
|
76
|
+
"@types/cli-progress": "^3.11.6",
|
|
77
77
|
"@types/compression": "^1.7.5",
|
|
78
78
|
"@types/cors": "^2.8.17",
|
|
79
79
|
"@types/debounce": "^1.2.4",
|
|
@@ -87,7 +87,7 @@
|
|
|
87
87
|
"@types/nunjucks": "^3.2.6",
|
|
88
88
|
"@types/opener": "^1.4.3",
|
|
89
89
|
"@types/semver": "^7.5.8",
|
|
90
|
-
"@types/uuid": "^
|
|
90
|
+
"@types/uuid": "^10.0.0",
|
|
91
91
|
"check-dependency-version-consistency": "^4.1.0",
|
|
92
92
|
"cloudflare": "^3.4.0",
|
|
93
93
|
"drizzle-kit": "^0.20.13",
|
|
@@ -101,17 +101,17 @@
|
|
|
101
101
|
"prettier": "^3.3.2",
|
|
102
102
|
"shx": "^0.3.4",
|
|
103
103
|
"ts-node": "^10.9.2",
|
|
104
|
-
"typescript": "^5.
|
|
105
|
-
"typescript-eslint": "^7.
|
|
104
|
+
"typescript": "^5.5.3",
|
|
105
|
+
"typescript-eslint": "^7.15.0"
|
|
106
106
|
},
|
|
107
107
|
"dependencies": {
|
|
108
|
-
"@anthropic-ai/sdk": "^0.24.
|
|
108
|
+
"@anthropic-ai/sdk": "^0.24.3",
|
|
109
109
|
"@apidevtools/json-schema-ref-parser": "^11.6.4",
|
|
110
|
-
"@googleapis/sheets": "^
|
|
110
|
+
"@googleapis/sheets": "^9.0.0",
|
|
111
111
|
"ajv": "^8.16.0",
|
|
112
112
|
"ajv-formats": "^2.1.1",
|
|
113
113
|
"async": "^3.2.5",
|
|
114
|
-
"better-sqlite3": "^11.
|
|
114
|
+
"better-sqlite3": "^11.1.2",
|
|
115
115
|
"cache-manager": "^4.1.0",
|
|
116
116
|
"cache-manager-fs-hash": "^1.0.0",
|
|
117
117
|
"chalk": "^4.1.2",
|
|
@@ -130,23 +130,24 @@
|
|
|
130
130
|
"express": "^4.19.2",
|
|
131
131
|
"fast-deep-equal": "^3.1.3",
|
|
132
132
|
"fastest-levenshtein": "^1.0.16",
|
|
133
|
-
"glob": "^10.4.
|
|
133
|
+
"glob": "^10.4.3",
|
|
134
134
|
"inquirer": "^8.2.6",
|
|
135
135
|
"js-yaml": "^4.1.0",
|
|
136
|
-
"mathjs": "^13.0.
|
|
136
|
+
"mathjs": "^13.0.2",
|
|
137
137
|
"node-fetch": "^2.6.7",
|
|
138
138
|
"nunjucks": "^3.2.4",
|
|
139
139
|
"openai": "^4.52.3",
|
|
140
140
|
"opener": "^1.5.2",
|
|
141
141
|
"proxy-agent": "^6.4.0",
|
|
142
142
|
"python-shell": "^5.0.0",
|
|
143
|
-
"replicate": "^0.
|
|
143
|
+
"replicate": "^0.31.1",
|
|
144
144
|
"rfdc": "^1.4.1",
|
|
145
145
|
"rouge": "git+https://github.com/kenlimmj/rouge.git#f35111b599aca55f1d4dc1d4a3d15e28e7f7c55f",
|
|
146
146
|
"semver": "^7.6.2",
|
|
147
147
|
"socket.io": "^4.7.5",
|
|
148
148
|
"tiny-invariant": "^1.3.3",
|
|
149
149
|
"uuid": "^10.0.0",
|
|
150
|
-
"winston": "^3.13.0"
|
|
150
|
+
"winston": "^3.13.0",
|
|
151
|
+
"zod": "^3.23.8"
|
|
151
152
|
}
|
|
152
153
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"database.d.ts","sourceRoot":"","sources":["../../../src/__mocks__/database.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"database.d.ts","sourceRoot":"","sources":["../../../src/__mocks__/database.ts"],"names":[],"mappings":"AAAA,QAAA,MAAM,cAAc,IAInB,CAAC;AAEF,QAAA,MAAM,aAAa,0BAAY,CAAC;AAEhC,QAAA,MAAM,eAAe,0BAGnB,CAAC"}
|
|
@@ -23,12 +23,12 @@ var __importStar = (this && this.__importStar) || function (mod) {
|
|
|
23
23
|
return result;
|
|
24
24
|
};
|
|
25
25
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
26
|
-
exports.
|
|
26
|
+
exports.getDirectory = getDirectory;
|
|
27
|
+
exports.importModule = importModule;
|
|
27
28
|
const path = __importStar(require("path"));
|
|
28
29
|
function getDirectory() {
|
|
29
30
|
return '/test/dir';
|
|
30
31
|
}
|
|
31
|
-
exports.getDirectory = getDirectory;
|
|
32
32
|
function importModule(filePath, functionName) {
|
|
33
33
|
const mod = require(path.resolve(filePath));
|
|
34
34
|
if (functionName) {
|
|
@@ -36,5 +36,4 @@ function importModule(filePath, functionName) {
|
|
|
36
36
|
}
|
|
37
37
|
return mod;
|
|
38
38
|
}
|
|
39
|
-
exports.importModule = importModule;
|
|
40
39
|
//# sourceMappingURL=esm.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"esm.js","sourceRoot":"","sources":["../../../src/__mocks__/esm.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"esm.js","sourceRoot":"","sources":["../../../src/__mocks__/esm.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAEA,oCAEC;AAED,oCAMC;AAZD,2CAA6B;AAE7B,SAAgB,YAAY;IAC1B,OAAO,WAAW,CAAC;AACrB,CAAC;AAED,SAAgB,YAAY,CAAC,QAAgB,EAAE,YAAqB;IAClE,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC5C,IAAI,YAAY,EAAE,CAAC;QACjB,OAAO,GAAG,CAAC,YAAY,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}
|
package/dist/src/accounts.d.ts
CHANGED
|
@@ -1,3 +1,4 @@
|
|
|
1
|
-
export declare function getUserEmail(): string |
|
|
1
|
+
export declare function getUserEmail(): string | null;
|
|
2
2
|
export declare function setUserEmail(email: string): void;
|
|
3
|
+
export declare function getAuthor(): string | null;
|
|
3
4
|
//# sourceMappingURL=accounts.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"accounts.d.ts","sourceRoot":"","sources":["../../src/accounts.ts"],"names":[],"mappings":"AAEA,wBAAgB,YAAY,IAAI,MAAM,GAAG,
|
|
1
|
+
{"version":3,"file":"accounts.d.ts","sourceRoot":"","sources":["../../src/accounts.ts"],"names":[],"mappings":"AAEA,wBAAgB,YAAY,IAAI,MAAM,GAAG,IAAI,CAG5C;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,QAEzC;AAED,wBAAgB,SAAS,IAAI,MAAM,GAAG,IAAI,CAEzC"}
|
package/dist/src/accounts.js
CHANGED
|
@@ -1,14 +1,17 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
-
exports.
|
|
3
|
+
exports.getUserEmail = getUserEmail;
|
|
4
|
+
exports.setUserEmail = setUserEmail;
|
|
5
|
+
exports.getAuthor = getAuthor;
|
|
4
6
|
const globalConfig_1 = require("./globalConfig");
|
|
5
7
|
function getUserEmail() {
|
|
6
8
|
const globalConfig = (0, globalConfig_1.readGlobalConfig)();
|
|
7
|
-
return globalConfig.account?.email;
|
|
9
|
+
return globalConfig.account?.email || null;
|
|
8
10
|
}
|
|
9
|
-
exports.getUserEmail = getUserEmail;
|
|
10
11
|
function setUserEmail(email) {
|
|
11
12
|
(0, globalConfig_1.writeGlobalConfig)({ account: { email } });
|
|
12
13
|
}
|
|
13
|
-
|
|
14
|
+
function getAuthor() {
|
|
15
|
+
return process.env.PROMPTFOO_AUTHOR || getUserEmail() || null;
|
|
16
|
+
}
|
|
14
17
|
//# sourceMappingURL=accounts.js.map
|
package/dist/src/accounts.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"accounts.js","sourceRoot":"","sources":["../../src/accounts.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"accounts.js","sourceRoot":"","sources":["../../src/accounts.ts"],"names":[],"mappings":";;AAEA,oCAGC;AAED,oCAEC;AAED,8BAEC;AAbD,iDAAqE;AAErE,SAAgB,YAAY;IAC1B,MAAM,YAAY,GAAG,IAAA,+BAAgB,GAAE,CAAC;IACxC,OAAO,YAAY,CAAC,OAAO,EAAE,KAAK,IAAI,IAAI,CAAC;AAC7C,CAAC;AAED,SAAgB,YAAY,CAAC,KAAa;IACxC,IAAA,gCAAiB,EAAC,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;AAC5C,CAAC;AAED,SAAgB,SAAS;IACvB,OAAO,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,YAAY,EAAE,IAAI,IAAI,CAAC;AAChE,CAAC"}
|
|
@@ -39,9 +39,9 @@ class AssertionsResult {
|
|
|
39
39
|
this.namedScores[metric] = (this.namedScores[metric] || 0) + result.score;
|
|
40
40
|
}
|
|
41
41
|
if (result.tokensUsed) {
|
|
42
|
-
this.tokensUsed.total += result.tokensUsed.total;
|
|
43
|
-
this.tokensUsed.prompt += result.tokensUsed.prompt;
|
|
44
|
-
this.tokensUsed.completion += result.tokensUsed.completion;
|
|
42
|
+
this.tokensUsed.total += result.tokensUsed.total || 0;
|
|
43
|
+
this.tokensUsed.prompt += result.tokensUsed.prompt || 0;
|
|
44
|
+
this.tokensUsed.completion += result.tokensUsed.completion || 0;
|
|
45
45
|
}
|
|
46
46
|
if (result.pass) {
|
|
47
47
|
return;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"AssertionsResult.js","sourceRoot":"","sources":["../../../src/assertions/AssertionsResult.ts"],"names":[],"mappings":";;;AAEA,MAAM,mBAAmB,GAAG;IAC1B,KAAK,EAAE,CAAC;IACR,MAAM,EAAE,CAAC;IACT,UAAU,EAAE,CAAC;CACd,CAAC;AAOF,MAAa,gBAAgB;IAC3B,MAAM,CAAC,eAAe;QACpB,OAAO;YACL,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,CAAC;YACR,MAAM,EAAE,eAAe;YACvB,UAAU,EAAE,EAAE,GAAG,mBAAmB,EAAE;YACtC,SAAS,EAAE,IAAI;SAChB,CAAC;IACJ,CAAC;IAcD,YAAY,EACV,SAAS,EACT,kBAAkB,MAIhB,EAAE;QAlBE,eAAU,GAAG;YACnB,GAAG,mBAAmB;SACvB,CAAC;QAGM,eAAU,GAAW,CAAC,CAAC;QACvB,gBAAW,GAAW,CAAC,CAAC;QAExB,qBAAgB,GAAoB,EAAE,CAAC;QACvC,gBAAW,GAA2B,EAAE,CAAC;QACzC,WAAM,GAAyB,IAAI,CAAC;QAS1C,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,mBAAmB,GAAG,kBAAkB,CAAC;IAChD,CAAC;IAED,IAAI,kBAAkB;QACpB,OAAO,IAAI,CAAC,mBAAmB,CAAC;IAClC,CAAC;IAED,SAAS,CAAC,EACR,KAAK,EACL,MAAM,EACN,MAAM,EACN,MAAM,GAAG,CAAC,GAMX;QACC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC;QACzC,IAAI,CAAC,WAAW,IAAI,MAAM,CAAC;QAC3B,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC;QAEtC,IAAI,MAAM,EAAE,CAAC;YACX,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC;QAC5E,CAAC;QAED,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YACtB,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC;
|
|
1
|
+
{"version":3,"file":"AssertionsResult.js","sourceRoot":"","sources":["../../../src/assertions/AssertionsResult.ts"],"names":[],"mappings":";;;AAEA,MAAM,mBAAmB,GAAG;IAC1B,KAAK,EAAE,CAAC;IACR,MAAM,EAAE,CAAC;IACT,UAAU,EAAE,CAAC;CACd,CAAC;AAOF,MAAa,gBAAgB;IAC3B,MAAM,CAAC,eAAe;QACpB,OAAO;YACL,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,CAAC;YACR,MAAM,EAAE,eAAe;YACvB,UAAU,EAAE,EAAE,GAAG,mBAAmB,EAAE;YACtC,SAAS,EAAE,IAAI;SAChB,CAAC;IACJ,CAAC;IAcD,YAAY,EACV,SAAS,EACT,kBAAkB,MAIhB,EAAE;QAlBE,eAAU,GAAG;YACnB,GAAG,mBAAmB;SACvB,CAAC;QAGM,eAAU,GAAW,CAAC,CAAC;QACvB,gBAAW,GAAW,CAAC,CAAC;QAExB,qBAAgB,GAAoB,EAAE,CAAC;QACvC,gBAAW,GAA2B,EAAE,CAAC;QACzC,WAAM,GAAyB,IAAI,CAAC;QAS1C,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,mBAAmB,GAAG,kBAAkB,CAAC;IAChD,CAAC;IAED,IAAI,kBAAkB;QACpB,OAAO,IAAI,CAAC,mBAAmB,CAAC;IAClC,CAAC;IAED,SAAS,CAAC,EACR,KAAK,EACL,MAAM,EACN,MAAM,EACN,MAAM,GAAG,CAAC,GAMX;QACC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC;QACzC,IAAI,CAAC,WAAW,IAAI,MAAM,CAAC;QAC3B,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC;QAEtC,IAAI,MAAM,EAAE,CAAC;YACX,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC;QAC5E,CAAC;QAED,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YACtB,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,MAAM,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC;YACtD,IAAI,CAAC,UAAU,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,IAAI,CAAC,CAAC;YACxD,IAAI,CAAC,UAAU,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,CAAC,UAAU,IAAI,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;YAChB,OAAO;QACT,CAAC;QAED,IAAI,CAAC,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC;QAElC,IAAI,OAAO,CAAC,GAAG,CAAC,qCAAqC,EAAE,CAAC;YACtD,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACjC,CAAC;IACH,CAAC;IAED,UAAU;QACR,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChB,OAAO,IAAI,CAAC,MAAM,CAAC;QACrB,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,WAAW,CAAC;QACjD,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC;QAE9B,IAAI,MAAM,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC;QAE9E,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACnB,wFAAwF;YACxF,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC;YAE/B,IAAI,IAAI,EAAE,CAAC;gBACT,MAAM,GAAG,mBAAmB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,SAAS,YAAY,CAAC;YAC/E,CAAC;iBAAM,CAAC;gBACN,MAAM,GAAG,mBAAmB,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,SAAS,YAAY,CAAC;YAC/E,CAAC;QACH,CAAC;QAED,IAAI,CAAC,MAAM,GAAG;YACZ,IAAI;YACJ,KAAK;YACL,MAAM;YACN,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,SAAS,EAAE,IAAI;SAChB,CAAC;QAEF,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;CACF;AA3GD,4CA2GC"}
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
-
exports.
|
|
3
|
+
exports.AssertValiationError = void 0;
|
|
4
|
+
exports.validateAssertions = validateAssertions;
|
|
4
5
|
class AssertValiationError extends Error {
|
|
5
6
|
constructor(message, testCase) {
|
|
6
7
|
const testCaseDescription = testCase.description || JSON.stringify(testCase);
|
|
@@ -31,5 +32,4 @@ function validateAssertions(tests) {
|
|
|
31
32
|
}
|
|
32
33
|
}
|
|
33
34
|
}
|
|
34
|
-
exports.validateAssertions = validateAssertions;
|
|
35
35
|
//# sourceMappingURL=validateAssertions.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"validateAssertions.js","sourceRoot":"","sources":["../../../src/assertions/validateAssertions.ts"],"names":[],"mappings":";;;
|
|
1
|
+
{"version":3,"file":"validateAssertions.js","sourceRoot":"","sources":["../../../src/assertions/validateAssertions.ts"],"names":[],"mappings":";;;AAyBA,gDAUC;AAjCD,MAAa,oBAAqB,SAAQ,KAAK;IAC7C,YAAY,OAAe,EAAE,QAAkB;QAC7C,MAAM,mBAAmB,GAAG,QAAQ,CAAC,WAAW,IAAI,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QAE7E,KAAK,CAAC,GAAG,OAAO,SAAS,mBAAmB,EAAE,CAAC,CAAC;QAChD,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAPD,oDAOC;AAED,SAAS,iBAAiB,CAAC,SAAiB,EAAE,IAAc;IAC1D,IAAI,CAAC,CAAC,QAAQ,IAAI,SAAS,CAAC,EAAE,CAAC;QAC7B,MAAM,IAAI,oBAAoB,CAAC,2CAA2C,EAAE,IAAI,CAAC,CAAC;IACpF,CAAC;IAED,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC;QACrC,MAAM,IAAI,oBAAoB,CAAC,oDAAoD,EAAE,IAAI,CAAC,CAAC;IAC7F,CAAC;IAED,IAAI,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,KAAK,YAAY,CAAC,EAAE,CAAC;QAC1E,MAAM,IAAI,oBAAoB,CAAC,4CAA4C,EAAE,IAAI,CAAC,CAAC;IACrF,CAAC;AACH,CAAC;AAED,SAAgB,kBAAkB,CAAC,KAA6D;IAC9F,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChB,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;gBACpC,IAAI,SAAS,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;oBACpC,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;gBACrC,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;AACH,CAAC"}
|
package/dist/src/assertions.js
CHANGED
|
@@ -26,7 +26,12 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
|
|
|
26
26
|
return (mod && mod.__esModule) ? mod : { "default": mod };
|
|
27
27
|
};
|
|
28
28
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
29
|
-
exports.
|
|
29
|
+
exports.MODEL_GRADED_ASSERTION_TYPES = void 0;
|
|
30
|
+
exports.isSql = isSql;
|
|
31
|
+
exports.runAssertion = runAssertion;
|
|
32
|
+
exports.runAssertions = runAssertions;
|
|
33
|
+
exports.runCompareAssertion = runCompareAssertion;
|
|
34
|
+
exports.readAssertions = readAssertions;
|
|
30
35
|
const ajv_1 = __importDefault(require("ajv"));
|
|
31
36
|
const ajv_formats_1 = __importDefault(require("ajv-formats"));
|
|
32
37
|
const async_1 = __importDefault(require("async"));
|
|
@@ -157,7 +162,6 @@ async function isSql(outputString, renderedValue, inverse, assertion) {
|
|
|
157
162
|
assertion,
|
|
158
163
|
};
|
|
159
164
|
}
|
|
160
|
-
exports.isSql = isSql;
|
|
161
165
|
async function runAssertion({ prompt, provider, assertion, test, output, latencyMs, logProbs, cost, }) {
|
|
162
166
|
let pass = false;
|
|
163
167
|
let score = 0.0;
|
|
@@ -990,7 +994,6 @@ ${isMultiline
|
|
|
990
994
|
}
|
|
991
995
|
throw new Error('Unknown assertion type: ' + assertion.type);
|
|
992
996
|
}
|
|
993
|
-
exports.runAssertion = runAssertion;
|
|
994
997
|
async function runAssertions({ prompt, provider, test, output, latencyMs, logProbs, cost, }) {
|
|
995
998
|
if (!test.assert || test.assert.length < 1) {
|
|
996
999
|
return AssertionsResult_1.AssertionsResult.noAssertsResult();
|
|
@@ -1055,7 +1058,6 @@ async function runAssertions({ prompt, provider, test, output, latencyMs, logPro
|
|
|
1055
1058
|
});
|
|
1056
1059
|
return mainAssertResult.testResult();
|
|
1057
1060
|
}
|
|
1058
|
-
exports.runAssertions = runAssertions;
|
|
1059
1061
|
async function runCompareAssertion(test, assertion, outputs) {
|
|
1060
1062
|
(0, tiny_invariant_1.default)(typeof assertion.value === 'string', 'select-best must have a string value');
|
|
1061
1063
|
test = getFinalTest(test, assertion);
|
|
@@ -1065,7 +1067,6 @@ async function runCompareAssertion(test, assertion, outputs) {
|
|
|
1065
1067
|
assertion,
|
|
1066
1068
|
}));
|
|
1067
1069
|
}
|
|
1068
|
-
exports.runCompareAssertion = runCompareAssertion;
|
|
1069
1070
|
async function readAssertions(filePath) {
|
|
1070
1071
|
try {
|
|
1071
1072
|
const assertions = js_yaml_1.default.load(fs_1.default.readFileSync(filePath, 'utf-8'));
|
|
@@ -1078,7 +1079,6 @@ async function readAssertions(filePath) {
|
|
|
1078
1079
|
throw new Error(`Failed to read assertions from ${filePath}:\n${err}`);
|
|
1079
1080
|
}
|
|
1080
1081
|
}
|
|
1081
|
-
exports.readAssertions = readAssertions;
|
|
1082
1082
|
// These exports are used by the node.js package (index.ts)
|
|
1083
1083
|
exports.default = {
|
|
1084
1084
|
runAssertion,
|