@meetkai/mka1 0.54.23 → 0.54.24
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/README.md +294 -276
- package/bin/mcp-server.js +3844 -1779
- package/bin/mcp-server.js.map +42 -17
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js +136 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.js +138 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js +142 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js +142 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.js +130 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js +146 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/lib/config.d.ts +2 -2
- package/dist/commonjs/lib/config.js +2 -2
- package/dist/commonjs/mcp-server/mcp-server.js +1 -1
- package/dist/commonjs/mcp-server/server.d.ts.map +1 -1
- package/dist/commonjs/mcp-server/server.js +13 -1
- package/dist/commonjs/mcp-server/server.js.map +1 -1
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts +111 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.js +136 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.js.map +1 -0
- package/dist/commonjs/models/components/evalleaderboardrow.d.ts +85 -0
- package/dist/commonjs/models/components/evalleaderboardrow.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalleaderboardrow.js +129 -0
- package/dist/commonjs/models/components/evalleaderboardrow.js.map +1 -0
- package/dist/commonjs/models/components/evalrunoverview.d.ts +101 -0
- package/dist/commonjs/models/components/evalrunoverview.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalrunoverview.js +150 -0
- package/dist/commonjs/models/components/evalrunoverview.js.map +1 -0
- package/dist/commonjs/models/components/evalsampleobject.d.ts +32 -0
- package/dist/commonjs/models/components/evalsampleobject.d.ts.map +1 -1
- package/dist/commonjs/models/components/evalsampleobject.js +20 -0
- package/dist/commonjs/models/components/evalsampleobject.js.map +1 -1
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js +87 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts +35 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.js +78 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.js.map +1 -0
- package/dist/commonjs/models/components/importagentrunrequest.d.ts +164 -0
- package/dist/commonjs/models/components/importagentrunrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/importagentrunrequest.js +182 -0
- package/dist/commonjs/models/components/importagentrunrequest.js.map +1 -0
- package/dist/commonjs/models/components/index.d.ts +7 -0
- package/dist/commonjs/models/components/index.d.ts.map +1 -1
- package/dist/commonjs/models/components/index.js +7 -0
- package/dist/commonjs/models/components/index.js.map +1 -1
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js +71 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.d.ts +264 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.d.ts.map +1 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.js +371 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.js.map +1 -0
- package/dist/commonjs/models/operations/cancelevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/cancelevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/cancelevalrun.js +16 -0
- package/dist/commonjs/models/operations/cancelevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/createevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/createevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/createevalrun.js +16 -0
- package/dist/commonjs/models/operations/createevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/getevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/getevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/getevalrun.js +16 -0
- package/dist/commonjs/models/operations/getevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/getevalrunoverview.d.ts +24 -0
- package/dist/commonjs/models/operations/getevalrunoverview.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalrunoverview.js +73 -0
- package/dist/commonjs/models/operations/getevalrunoverview.js.map +1 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts +75 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.js +133 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.js.map +1 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts +32 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.js +83 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.js.map +1 -0
- package/dist/commonjs/models/operations/importagentevalrun.d.ts +262 -0
- package/dist/commonjs/models/operations/importagentevalrun.d.ts.map +1 -0
- package/dist/commonjs/models/operations/importagentevalrun.js +367 -0
- package/dist/commonjs/models/operations/importagentevalrun.js.map +1 -0
- package/dist/commonjs/models/operations/index.d.ts +6 -0
- package/dist/commonjs/models/operations/index.d.ts.map +1 -1
- package/dist/commonjs/models/operations/index.js +6 -0
- package/dist/commonjs/models/operations/index.js.map +1 -1
- package/dist/commonjs/models/operations/listevalruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalruns.js +16 -0
- package/dist/commonjs/models/operations/listevalruns.js.map +1 -1
- package/dist/commonjs/models/operations/listevalscheduleruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalscheduleruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalscheduleruns.js +16 -0
- package/dist/commonjs/models/operations/listevalscheduleruns.js.map +1 -1
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js +16 -0
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
- package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts +20 -0
- package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
- package/dist/commonjs/models/operations/rerunfailedevalsamples.js +16 -0
- package/dist/commonjs/models/operations/rerunfailedevalsamples.js.map +1 -1
- package/dist/commonjs/models/operations/retryfailedevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/retryfailedevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/retryfailedevalrun.js +16 -0
- package/dist/commonjs/models/operations/retryfailedevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts +31 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.js +86 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.js.map +1 -0
- package/dist/commonjs/react-query/index.d.ts +6 -0
- package/dist/commonjs/react-query/index.d.ts.map +1 -1
- package/dist/commonjs/react-query/index.js +6 -0
- package/dist/commonjs/react-query/index.js.map +1 -1
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js +47 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js +37 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts +40 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.js +60 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js +45 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js +65 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js +41 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js +65 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts +28 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.js +47 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js +47 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/sdk/evals.d.ts +42 -0
- package/dist/commonjs/sdk/evals.d.ts.map +1 -1
- package/dist/commonjs/sdk/evals.js +60 -0
- package/dist/commonjs/sdk/evals.js.map +1 -1
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.js +100 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.js +102 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js +106 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js +106 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.js +94 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js +110 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/lib/config.d.ts +2 -2
- package/dist/esm/lib/config.js +2 -2
- package/dist/esm/mcp-server/mcp-server.js +1 -1
- package/dist/esm/mcp-server/server.d.ts.map +1 -1
- package/dist/esm/mcp-server/server.js +13 -1
- package/dist/esm/mcp-server/server.js.map +1 -1
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/models/components/appendagenttrialsrequest.d.ts +111 -0
- package/dist/esm/models/components/appendagenttrialsrequest.d.ts.map +1 -0
- package/dist/esm/models/components/appendagenttrialsrequest.js +96 -0
- package/dist/esm/models/components/appendagenttrialsrequest.js.map +1 -0
- package/dist/esm/models/components/evalleaderboardrow.d.ts +85 -0
- package/dist/esm/models/components/evalleaderboardrow.d.ts.map +1 -0
- package/dist/esm/models/components/evalleaderboardrow.js +89 -0
- package/dist/esm/models/components/evalleaderboardrow.js.map +1 -0
- package/dist/esm/models/components/evalrunoverview.d.ts +101 -0
- package/dist/esm/models/components/evalrunoverview.d.ts.map +1 -0
- package/dist/esm/models/components/evalrunoverview.js +108 -0
- package/dist/esm/models/components/evalrunoverview.js.map +1 -0
- package/dist/esm/models/components/evalsampleobject.d.ts +32 -0
- package/dist/esm/models/components/evalsampleobject.d.ts.map +1 -1
- package/dist/esm/models/components/evalsampleobject.js +20 -0
- package/dist/esm/models/components/evalsampleobject.js.map +1 -1
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js +49 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
- package/dist/esm/models/components/evalsuiteleaderboard.d.ts +35 -0
- package/dist/esm/models/components/evalsuiteleaderboard.d.ts.map +1 -0
- package/dist/esm/models/components/evalsuiteleaderboard.js +40 -0
- package/dist/esm/models/components/evalsuiteleaderboard.js.map +1 -0
- package/dist/esm/models/components/importagentrunrequest.d.ts +164 -0
- package/dist/esm/models/components/importagentrunrequest.d.ts.map +1 -0
- package/dist/esm/models/components/importagentrunrequest.js +142 -0
- package/dist/esm/models/components/importagentrunrequest.js.map +1 -0
- package/dist/esm/models/components/index.d.ts +7 -0
- package/dist/esm/models/components/index.d.ts.map +1 -1
- package/dist/esm/models/components/index.js +7 -0
- package/dist/esm/models/components/index.js.map +1 -1
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js +33 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
- package/dist/esm/models/operations/appendagentevaltrials.d.ts +264 -0
- package/dist/esm/models/operations/appendagentevaltrials.d.ts.map +1 -0
- package/dist/esm/models/operations/appendagentevaltrials.js +323 -0
- package/dist/esm/models/operations/appendagentevaltrials.js.map +1 -0
- package/dist/esm/models/operations/cancelevalrun.d.ts +20 -0
- package/dist/esm/models/operations/cancelevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/cancelevalrun.js +16 -0
- package/dist/esm/models/operations/cancelevalrun.js.map +1 -1
- package/dist/esm/models/operations/createevalrun.d.ts +20 -0
- package/dist/esm/models/operations/createevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/createevalrun.js +16 -0
- package/dist/esm/models/operations/createevalrun.js.map +1 -1
- package/dist/esm/models/operations/getevalrun.d.ts +20 -0
- package/dist/esm/models/operations/getevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/getevalrun.js +16 -0
- package/dist/esm/models/operations/getevalrun.js.map +1 -1
- package/dist/esm/models/operations/getevalrunoverview.d.ts +24 -0
- package/dist/esm/models/operations/getevalrunoverview.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalrunoverview.js +35 -0
- package/dist/esm/models/operations/getevalrunoverview.js.map +1 -0
- package/dist/esm/models/operations/getevalsampletrajectory.d.ts +75 -0
- package/dist/esm/models/operations/getevalsampletrajectory.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalsampletrajectory.js +93 -0
- package/dist/esm/models/operations/getevalsampletrajectory.js.map +1 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts +32 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.js +45 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.js.map +1 -0
- package/dist/esm/models/operations/importagentevalrun.d.ts +262 -0
- package/dist/esm/models/operations/importagentevalrun.d.ts.map +1 -0
- package/dist/esm/models/operations/importagentevalrun.js +319 -0
- package/dist/esm/models/operations/importagentevalrun.js.map +1 -0
- package/dist/esm/models/operations/index.d.ts +6 -0
- package/dist/esm/models/operations/index.d.ts.map +1 -1
- package/dist/esm/models/operations/index.js +6 -0
- package/dist/esm/models/operations/index.js.map +1 -1
- package/dist/esm/models/operations/listevalruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalruns.js +16 -0
- package/dist/esm/models/operations/listevalruns.js.map +1 -1
- package/dist/esm/models/operations/listevalscheduleruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalscheduleruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalscheduleruns.js +16 -0
- package/dist/esm/models/operations/listevalscheduleruns.js.map +1 -1
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.js +16 -0
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
- package/dist/esm/models/operations/rerunfailedevalsamples.d.ts +20 -0
- package/dist/esm/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
- package/dist/esm/models/operations/rerunfailedevalsamples.js +16 -0
- package/dist/esm/models/operations/rerunfailedevalsamples.js.map +1 -1
- package/dist/esm/models/operations/retryfailedevalrun.d.ts +20 -0
- package/dist/esm/models/operations/retryfailedevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/retryfailedevalrun.js +16 -0
- package/dist/esm/models/operations/retryfailedevalrun.js.map +1 -1
- package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts +31 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.js +48 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.js.map +1 -0
- package/dist/esm/react-query/index.d.ts +6 -0
- package/dist/esm/react-query/index.d.ts.map +1 -1
- package/dist/esm/react-query/index.js +6 -0
- package/dist/esm/react-query/index.js.map +1 -1
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.js +42 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.js +32 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts +40 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.js +50 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js +40 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js +55 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js +36 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js +55 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts +28 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.js +42 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js +42 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/sdk/evals.d.ts +42 -0
- package/dist/esm/sdk/evals.d.ts.map +1 -1
- package/dist/esm/sdk/evals.js +60 -0
- package/dist/esm/sdk/evals.js.map +1 -1
- package/jsr.json +1 -1
- package/package.json +1 -1
- package/src/funcs/llmEvalsAppendAgentTrials.ts +188 -0
- package/src/funcs/llmEvalsGetRunOverview.ts +192 -0
- package/src/funcs/llmEvalsGetSampleTrajectory.ts +197 -0
- package/src/funcs/llmEvalsGetSuiteLeaderboard.ts +197 -0
- package/src/funcs/llmEvalsImportAgentRun.ts +181 -0
- package/src/funcs/llmEvalsUploadSampleTrajectory.ts +201 -0
- package/src/lib/config.ts +2 -2
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +13 -1
- package/src/mcp-server/tools/llmEvalsAppendAgentTrials.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetRunOverview.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetSampleTrajectory.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetSuiteLeaderboard.ts +37 -0
- package/src/mcp-server/tools/llmEvalsImportAgentRun.ts +37 -0
- package/src/mcp-server/tools/llmEvalsUploadSampleTrajectory.ts +38 -0
- package/src/models/components/appendagenttrialsrequest.ts +224 -0
- package/src/models/components/evalleaderboardrow.ts +200 -0
- package/src/models/components/evalrunoverview.ts +240 -0
- package/src/models/components/evalsampleobject.ts +48 -0
- package/src/models/components/evalsampletrajectoryuploadedobject.ts +98 -0
- package/src/models/components/evalsuiteleaderboard.ts +93 -0
- package/src/models/components/importagentrunrequest.ts +319 -0
- package/src/models/components/index.ts +7 -0
- package/src/models/components/uploadevalsampletrajectoryrequest.ts +79 -0
- package/src/models/operations/appendagentevaltrials.ts +678 -0
- package/src/models/operations/cancelevalrun.ts +36 -0
- package/src/models/operations/createevalrun.ts +36 -0
- package/src/models/operations/getevalrun.ts +36 -0
- package/src/models/operations/getevalrunoverview.ts +73 -0
- package/src/models/operations/getevalsampletrajectory.ts +199 -0
- package/src/models/operations/getevalsuiteleaderboard.ts +93 -0
- package/src/models/operations/importagentevalrun.ts +665 -0
- package/src/models/operations/index.ts +6 -0
- package/src/models/operations/listevalruns.ts +36 -0
- package/src/models/operations/listevalscheduleruns.ts +36 -0
- package/src/models/operations/listevalsuiteleaderboardruns.ts +36 -0
- package/src/models/operations/rerunfailedevalsamples.ts +36 -0
- package/src/models/operations/retryfailedevalrun.ts +36 -0
- package/src/models/operations/uploadevalsampletrajectory.ts +98 -0
- package/src/react-query/index.ts +6 -0
- package/src/react-query/llmEvalsAppendAgentTrials.ts +109 -0
- package/src/react-query/llmEvalsGetRunOverview.core.ts +77 -0
- package/src/react-query/llmEvalsGetRunOverview.ts +150 -0
- package/src/react-query/llmEvalsGetSampleTrajectory.core.ts +94 -0
- package/src/react-query/llmEvalsGetSampleTrajectory.ts +159 -0
- package/src/react-query/llmEvalsGetSuiteLeaderboard.core.ts +89 -0
- package/src/react-query/llmEvalsGetSuiteLeaderboard.ts +163 -0
- package/src/react-query/llmEvalsImportAgentRun.ts +109 -0
- package/src/react-query/llmEvalsUploadSampleTrajectory.ts +110 -0
- package/src/sdk/evals.ts +108 -0
|
@@ -33,7 +33,30 @@ export type EvalSampleObject = {
|
|
|
33
33
|
extractedOutput: string | null;
|
|
34
34
|
scores: { [k: string]: any };
|
|
35
35
|
judge: { [k: string]: any } | null;
|
|
36
|
+
/**
|
|
37
|
+
* The exception that ended this sample, if any. For a harness trial this also carries an agent timeout — which is still a *scored* trial, so do not read a non-null `error` here as an unscored one; check `status` and `scores`.
|
|
38
|
+
*/
|
|
36
39
|
error: { [k: string]: any } | null;
|
|
40
|
+
/**
|
|
41
|
+
* Spend for this one trial in USD. Null outside harness runs, and when the harness reported no cost.
|
|
42
|
+
*/
|
|
43
|
+
costUsd: number | null;
|
|
44
|
+
/**
|
|
45
|
+
* Prompt tokens for the whole trial session. Includes `cached_tokens`.
|
|
46
|
+
*/
|
|
47
|
+
inputTokens: number | null;
|
|
48
|
+
/**
|
|
49
|
+
* Completion tokens for the whole trial session.
|
|
50
|
+
*/
|
|
51
|
+
outputTokens: number | null;
|
|
52
|
+
/**
|
|
53
|
+
* The share of `input_tokens` served from the provider's prompt cache. Reported separately because it is what makes two agents at the same token count cost an order of magnitude apart.
|
|
54
|
+
*/
|
|
55
|
+
cachedTokens: number | null;
|
|
56
|
+
/**
|
|
57
|
+
* Per-phase wall clock for a harness trial: `environment_setup`, `agent_setup`, `agent_execution`, `verifier`, each `{ started_at, finished_at }` as ISO 8601 timestamps exactly as the harness recorded them. A phase the harness did not time is absent rather than null. Drives the trial timing bar. Null outside harness runs.
|
|
58
|
+
*/
|
|
59
|
+
phaseTimings: { [k: string]: any } | null;
|
|
37
60
|
createdAt: number;
|
|
38
61
|
startedAt: number | null;
|
|
39
62
|
completedAt: number | null;
|
|
@@ -62,6 +85,11 @@ export const EvalSampleObject$inboundSchema: z.ZodType<
|
|
|
62
85
|
scores: z.record(z.any()),
|
|
63
86
|
judge: z.nullable(z.record(z.any())),
|
|
64
87
|
error: z.nullable(z.record(z.any())),
|
|
88
|
+
cost_usd: z.nullable(z.number()),
|
|
89
|
+
input_tokens: z.nullable(z.number().int()),
|
|
90
|
+
output_tokens: z.nullable(z.number().int()),
|
|
91
|
+
cached_tokens: z.nullable(z.number().int()),
|
|
92
|
+
phase_timings: z.nullable(z.record(z.any())),
|
|
65
93
|
created_at: z.number().int(),
|
|
66
94
|
started_at: z.nullable(z.number().int()),
|
|
67
95
|
completed_at: z.nullable(z.number().int()),
|
|
@@ -74,6 +102,11 @@ export const EvalSampleObject$inboundSchema: z.ZodType<
|
|
|
74
102
|
"response_id": "responseId",
|
|
75
103
|
"output_text": "outputText",
|
|
76
104
|
"extracted_output": "extractedOutput",
|
|
105
|
+
"cost_usd": "costUsd",
|
|
106
|
+
"input_tokens": "inputTokens",
|
|
107
|
+
"output_tokens": "outputTokens",
|
|
108
|
+
"cached_tokens": "cachedTokens",
|
|
109
|
+
"phase_timings": "phaseTimings",
|
|
77
110
|
"created_at": "createdAt",
|
|
78
111
|
"started_at": "startedAt",
|
|
79
112
|
"completed_at": "completedAt",
|
|
@@ -98,6 +131,11 @@ export type EvalSampleObject$Outbound = {
|
|
|
98
131
|
scores: { [k: string]: any };
|
|
99
132
|
judge: { [k: string]: any } | null;
|
|
100
133
|
error: { [k: string]: any } | null;
|
|
134
|
+
cost_usd: number | null;
|
|
135
|
+
input_tokens: number | null;
|
|
136
|
+
output_tokens: number | null;
|
|
137
|
+
cached_tokens: number | null;
|
|
138
|
+
phase_timings: { [k: string]: any } | null;
|
|
101
139
|
created_at: number;
|
|
102
140
|
started_at: number | null;
|
|
103
141
|
completed_at: number | null;
|
|
@@ -126,6 +164,11 @@ export const EvalSampleObject$outboundSchema: z.ZodType<
|
|
|
126
164
|
scores: z.record(z.any()),
|
|
127
165
|
judge: z.nullable(z.record(z.any())),
|
|
128
166
|
error: z.nullable(z.record(z.any())),
|
|
167
|
+
costUsd: z.nullable(z.number()),
|
|
168
|
+
inputTokens: z.nullable(z.number().int()),
|
|
169
|
+
outputTokens: z.nullable(z.number().int()),
|
|
170
|
+
cachedTokens: z.nullable(z.number().int()),
|
|
171
|
+
phaseTimings: z.nullable(z.record(z.any())),
|
|
129
172
|
createdAt: z.number().int(),
|
|
130
173
|
startedAt: z.nullable(z.number().int()),
|
|
131
174
|
completedAt: z.nullable(z.number().int()),
|
|
@@ -138,6 +181,11 @@ export const EvalSampleObject$outboundSchema: z.ZodType<
|
|
|
138
181
|
responseId: "response_id",
|
|
139
182
|
outputText: "output_text",
|
|
140
183
|
extractedOutput: "extracted_output",
|
|
184
|
+
costUsd: "cost_usd",
|
|
185
|
+
inputTokens: "input_tokens",
|
|
186
|
+
outputTokens: "output_tokens",
|
|
187
|
+
cachedTokens: "cached_tokens",
|
|
188
|
+
phaseTimings: "phase_timings",
|
|
141
189
|
createdAt: "created_at",
|
|
142
190
|
startedAt: "started_at",
|
|
143
191
|
completedAt: "completed_at",
|
|
@@ -0,0 +1,98 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { remap as remap$ } from "../../lib/primitives.js";
|
|
7
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
8
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
9
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
10
|
+
|
|
11
|
+
export type EvalSampleTrajectoryUploadedObject = {
|
|
12
|
+
object: "eval.sample.trajectory.uploaded";
|
|
13
|
+
runId: string;
|
|
14
|
+
sampleIndex: number;
|
|
15
|
+
taskId: string;
|
|
16
|
+
model: string;
|
|
17
|
+
stepCount: number;
|
|
18
|
+
/**
|
|
19
|
+
* Measured from the stored payload, so it describes what was actually written.
|
|
20
|
+
*/
|
|
21
|
+
sizeBytes: number;
|
|
22
|
+
};
|
|
23
|
+
|
|
24
|
+
/** @internal */
|
|
25
|
+
export const EvalSampleTrajectoryUploadedObject$inboundSchema: z.ZodType<
|
|
26
|
+
EvalSampleTrajectoryUploadedObject,
|
|
27
|
+
z.ZodTypeDef,
|
|
28
|
+
unknown
|
|
29
|
+
> = z.object({
|
|
30
|
+
object: z.literal("eval.sample.trajectory.uploaded"),
|
|
31
|
+
run_id: z.string(),
|
|
32
|
+
sample_index: z.number().int(),
|
|
33
|
+
task_id: z.string(),
|
|
34
|
+
model: z.string(),
|
|
35
|
+
step_count: z.number().int(),
|
|
36
|
+
size_bytes: z.number().int(),
|
|
37
|
+
}).transform((v) => {
|
|
38
|
+
return remap$(v, {
|
|
39
|
+
"run_id": "runId",
|
|
40
|
+
"sample_index": "sampleIndex",
|
|
41
|
+
"task_id": "taskId",
|
|
42
|
+
"step_count": "stepCount",
|
|
43
|
+
"size_bytes": "sizeBytes",
|
|
44
|
+
});
|
|
45
|
+
});
|
|
46
|
+
/** @internal */
|
|
47
|
+
export type EvalSampleTrajectoryUploadedObject$Outbound = {
|
|
48
|
+
object: "eval.sample.trajectory.uploaded";
|
|
49
|
+
run_id: string;
|
|
50
|
+
sample_index: number;
|
|
51
|
+
task_id: string;
|
|
52
|
+
model: string;
|
|
53
|
+
step_count: number;
|
|
54
|
+
size_bytes: number;
|
|
55
|
+
};
|
|
56
|
+
|
|
57
|
+
/** @internal */
|
|
58
|
+
export const EvalSampleTrajectoryUploadedObject$outboundSchema: z.ZodType<
|
|
59
|
+
EvalSampleTrajectoryUploadedObject$Outbound,
|
|
60
|
+
z.ZodTypeDef,
|
|
61
|
+
EvalSampleTrajectoryUploadedObject
|
|
62
|
+
> = z.object({
|
|
63
|
+
object: z.literal("eval.sample.trajectory.uploaded"),
|
|
64
|
+
runId: z.string(),
|
|
65
|
+
sampleIndex: z.number().int(),
|
|
66
|
+
taskId: z.string(),
|
|
67
|
+
model: z.string(),
|
|
68
|
+
stepCount: z.number().int(),
|
|
69
|
+
sizeBytes: z.number().int(),
|
|
70
|
+
}).transform((v) => {
|
|
71
|
+
return remap$(v, {
|
|
72
|
+
runId: "run_id",
|
|
73
|
+
sampleIndex: "sample_index",
|
|
74
|
+
taskId: "task_id",
|
|
75
|
+
stepCount: "step_count",
|
|
76
|
+
sizeBytes: "size_bytes",
|
|
77
|
+
});
|
|
78
|
+
});
|
|
79
|
+
|
|
80
|
+
export function evalSampleTrajectoryUploadedObjectToJSON(
|
|
81
|
+
evalSampleTrajectoryUploadedObject: EvalSampleTrajectoryUploadedObject,
|
|
82
|
+
): string {
|
|
83
|
+
return JSON.stringify(
|
|
84
|
+
EvalSampleTrajectoryUploadedObject$outboundSchema.parse(
|
|
85
|
+
evalSampleTrajectoryUploadedObject,
|
|
86
|
+
),
|
|
87
|
+
);
|
|
88
|
+
}
|
|
89
|
+
export function evalSampleTrajectoryUploadedObjectFromJSON(
|
|
90
|
+
jsonString: string,
|
|
91
|
+
): SafeParseResult<EvalSampleTrajectoryUploadedObject, SDKValidationError> {
|
|
92
|
+
return safeParse(
|
|
93
|
+
jsonString,
|
|
94
|
+
(x) =>
|
|
95
|
+
EvalSampleTrajectoryUploadedObject$inboundSchema.parse(JSON.parse(x)),
|
|
96
|
+
`Failed to parse 'EvalSampleTrajectoryUploadedObject' from JSON`,
|
|
97
|
+
);
|
|
98
|
+
}
|
|
@@ -0,0 +1,93 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { remap as remap$ } from "../../lib/primitives.js";
|
|
7
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
8
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
9
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
10
|
+
import {
|
|
11
|
+
EvalLeaderboardRow,
|
|
12
|
+
EvalLeaderboardRow$inboundSchema,
|
|
13
|
+
EvalLeaderboardRow$Outbound,
|
|
14
|
+
EvalLeaderboardRow$outboundSchema,
|
|
15
|
+
} from "./evalleaderboardrow.js";
|
|
16
|
+
|
|
17
|
+
export type EvalSuiteLeaderboard = {
|
|
18
|
+
object: "eval.leaderboard";
|
|
19
|
+
suiteId: string;
|
|
20
|
+
/**
|
|
21
|
+
* The version this board was restricted to, or null when it spans every version.
|
|
22
|
+
*/
|
|
23
|
+
suiteVersion: number | null;
|
|
24
|
+
/**
|
|
25
|
+
* The `scores` key `accuracy` averages. Echoed back so a board rendered from a non-default metric is self-describing.
|
|
26
|
+
*/
|
|
27
|
+
metric: string;
|
|
28
|
+
/**
|
|
29
|
+
* Rows ordered by `accuracy` descending, nulls last, then by agent, model, and effort.
|
|
30
|
+
*/
|
|
31
|
+
data: Array<EvalLeaderboardRow>;
|
|
32
|
+
};
|
|
33
|
+
|
|
34
|
+
/** @internal */
|
|
35
|
+
export const EvalSuiteLeaderboard$inboundSchema: z.ZodType<
|
|
36
|
+
EvalSuiteLeaderboard,
|
|
37
|
+
z.ZodTypeDef,
|
|
38
|
+
unknown
|
|
39
|
+
> = z.object({
|
|
40
|
+
object: z.literal("eval.leaderboard"),
|
|
41
|
+
suite_id: z.string(),
|
|
42
|
+
suite_version: z.nullable(z.number().int()),
|
|
43
|
+
metric: z.string(),
|
|
44
|
+
data: z.array(EvalLeaderboardRow$inboundSchema),
|
|
45
|
+
}).transform((v) => {
|
|
46
|
+
return remap$(v, {
|
|
47
|
+
"suite_id": "suiteId",
|
|
48
|
+
"suite_version": "suiteVersion",
|
|
49
|
+
});
|
|
50
|
+
});
|
|
51
|
+
/** @internal */
|
|
52
|
+
export type EvalSuiteLeaderboard$Outbound = {
|
|
53
|
+
object: "eval.leaderboard";
|
|
54
|
+
suite_id: string;
|
|
55
|
+
suite_version: number | null;
|
|
56
|
+
metric: string;
|
|
57
|
+
data: Array<EvalLeaderboardRow$Outbound>;
|
|
58
|
+
};
|
|
59
|
+
|
|
60
|
+
/** @internal */
|
|
61
|
+
export const EvalSuiteLeaderboard$outboundSchema: z.ZodType<
|
|
62
|
+
EvalSuiteLeaderboard$Outbound,
|
|
63
|
+
z.ZodTypeDef,
|
|
64
|
+
EvalSuiteLeaderboard
|
|
65
|
+
> = z.object({
|
|
66
|
+
object: z.literal("eval.leaderboard"),
|
|
67
|
+
suiteId: z.string(),
|
|
68
|
+
suiteVersion: z.nullable(z.number().int()),
|
|
69
|
+
metric: z.string(),
|
|
70
|
+
data: z.array(EvalLeaderboardRow$outboundSchema),
|
|
71
|
+
}).transform((v) => {
|
|
72
|
+
return remap$(v, {
|
|
73
|
+
suiteId: "suite_id",
|
|
74
|
+
suiteVersion: "suite_version",
|
|
75
|
+
});
|
|
76
|
+
});
|
|
77
|
+
|
|
78
|
+
export function evalSuiteLeaderboardToJSON(
|
|
79
|
+
evalSuiteLeaderboard: EvalSuiteLeaderboard,
|
|
80
|
+
): string {
|
|
81
|
+
return JSON.stringify(
|
|
82
|
+
EvalSuiteLeaderboard$outboundSchema.parse(evalSuiteLeaderboard),
|
|
83
|
+
);
|
|
84
|
+
}
|
|
85
|
+
export function evalSuiteLeaderboardFromJSON(
|
|
86
|
+
jsonString: string,
|
|
87
|
+
): SafeParseResult<EvalSuiteLeaderboard, SDKValidationError> {
|
|
88
|
+
return safeParse(
|
|
89
|
+
jsonString,
|
|
90
|
+
(x) => EvalSuiteLeaderboard$inboundSchema.parse(JSON.parse(x)),
|
|
91
|
+
`Failed to parse 'EvalSuiteLeaderboard' from JSON`,
|
|
92
|
+
);
|
|
93
|
+
}
|
|
@@ -0,0 +1,319 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { remap as remap$ } from "../../lib/primitives.js";
|
|
7
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
8
|
+
import { ClosedEnum } from "../../types/enums.js";
|
|
9
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
10
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
11
|
+
|
|
12
|
+
/**
|
|
13
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
14
|
+
*/
|
|
15
|
+
export const ImportAgentRunRequestStatus = {
|
|
16
|
+
Completed: "completed",
|
|
17
|
+
Failed: "failed",
|
|
18
|
+
} as const;
|
|
19
|
+
/**
|
|
20
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
21
|
+
*/
|
|
22
|
+
export type ImportAgentRunRequestStatus = ClosedEnum<
|
|
23
|
+
typeof ImportAgentRunRequestStatus
|
|
24
|
+
>;
|
|
25
|
+
|
|
26
|
+
export type ImportAgentRunRequestTrials = {
|
|
27
|
+
/**
|
|
28
|
+
* Must already exist in the target suite version's manifest; unknown ids are rejected rather than invented.
|
|
29
|
+
*/
|
|
30
|
+
taskId: string;
|
|
31
|
+
/**
|
|
32
|
+
* Defaults to the trial's position in `trials`. Unique per (task, model), not per run.
|
|
33
|
+
*/
|
|
34
|
+
sampleIndex?: number | undefined;
|
|
35
|
+
/**
|
|
36
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
37
|
+
*/
|
|
38
|
+
status?: ImportAgentRunRequestStatus | undefined;
|
|
39
|
+
/**
|
|
40
|
+
* The task instruction.
|
|
41
|
+
*/
|
|
42
|
+
prompt?: string | undefined;
|
|
43
|
+
target?: string | null | undefined;
|
|
44
|
+
/**
|
|
45
|
+
* What decided the score — the verifier's output. The agent's session goes to the trajectory route.
|
|
46
|
+
*/
|
|
47
|
+
outputText?: string | null | undefined;
|
|
48
|
+
/**
|
|
49
|
+
* Metric values for this trial, e.g. `{ "reward": 1, "resolved": 1 }`. The board averages `reward` by default.
|
|
50
|
+
*/
|
|
51
|
+
scores?: { [k: string]: number } | undefined;
|
|
52
|
+
/**
|
|
53
|
+
* The exception this trial recorded, if any. Does not by itself make the trial a failure.
|
|
54
|
+
*/
|
|
55
|
+
error?: { [k: string]: any } | null | undefined;
|
|
56
|
+
costUsd?: number | null | undefined;
|
|
57
|
+
inputTokens?: number | null | undefined;
|
|
58
|
+
outputTokens?: number | null | undefined;
|
|
59
|
+
/**
|
|
60
|
+
* The share of `input_tokens` served from the provider's prompt cache.
|
|
61
|
+
*/
|
|
62
|
+
cachedTokens?: number | null | undefined;
|
|
63
|
+
phaseTimings?: { [k: string]: any } | null | undefined;
|
|
64
|
+
};
|
|
65
|
+
|
|
66
|
+
export type ImportAgentRunRequest = {
|
|
67
|
+
/**
|
|
68
|
+
* Existing suite to import into. Its manifest defines the task set, so create the suite first.
|
|
69
|
+
*/
|
|
70
|
+
suiteId: string;
|
|
71
|
+
/**
|
|
72
|
+
* Defaults to the suite's active version.
|
|
73
|
+
*/
|
|
74
|
+
suiteVersion?: number | undefined;
|
|
75
|
+
/**
|
|
76
|
+
* The harness agent, e.g. `omp`.
|
|
77
|
+
*/
|
|
78
|
+
agentName: string;
|
|
79
|
+
agentVersion?: string | null | undefined;
|
|
80
|
+
/**
|
|
81
|
+
* Reasoning-effort setting, e.g. `medium`. Part of the leaderboard's row identity, so the same agent and model at two efforts stay two rows.
|
|
82
|
+
*/
|
|
83
|
+
agentEffort?: string | null | undefined;
|
|
84
|
+
model: string;
|
|
85
|
+
/**
|
|
86
|
+
* Total spend for the job. Falls back to the sum of the trials' own costs when omitted.
|
|
87
|
+
*/
|
|
88
|
+
costUsd?: number | null | undefined;
|
|
89
|
+
/**
|
|
90
|
+
* When the job ran, unix seconds. Defaults to now — set it when backfilling, since the board orders by this.
|
|
91
|
+
*/
|
|
92
|
+
createdAt?: number | undefined;
|
|
93
|
+
metadata?: { [k: string]: string } | undefined;
|
|
94
|
+
/**
|
|
95
|
+
* Reusing a key returns the run created the first time instead of importing a second copy. Without one, a retried upload silently doubles the trials behind a board row.
|
|
96
|
+
*/
|
|
97
|
+
idempotencyKey?: string | undefined;
|
|
98
|
+
/**
|
|
99
|
+
* May be empty when opening a run to stream trials into. A one-shot import sends them all here.
|
|
100
|
+
*/
|
|
101
|
+
trials?: Array<ImportAgentRunRequestTrials> | undefined;
|
|
102
|
+
/**
|
|
103
|
+
* How many trials the job will produce. Sets the run's total up front so the console shows real progress (12/89) while it streams, instead of 12/12 that grows.
|
|
104
|
+
*/
|
|
105
|
+
expectedTrials?: number | undefined;
|
|
106
|
+
/**
|
|
107
|
+
* False opens the run `in_progress` for `POST /evals/runs/{run_id}/trials` to fill. Leaderboards only count completed runs, so a streaming run stays off the board until it is finalized.
|
|
108
|
+
*/
|
|
109
|
+
finalize?: boolean | undefined;
|
|
110
|
+
};
|
|
111
|
+
|
|
112
|
+
/** @internal */
|
|
113
|
+
export const ImportAgentRunRequestStatus$inboundSchema: z.ZodNativeEnum<
|
|
114
|
+
typeof ImportAgentRunRequestStatus
|
|
115
|
+
> = z.nativeEnum(ImportAgentRunRequestStatus);
|
|
116
|
+
/** @internal */
|
|
117
|
+
export const ImportAgentRunRequestStatus$outboundSchema: z.ZodNativeEnum<
|
|
118
|
+
typeof ImportAgentRunRequestStatus
|
|
119
|
+
> = ImportAgentRunRequestStatus$inboundSchema;
|
|
120
|
+
|
|
121
|
+
/** @internal */
|
|
122
|
+
export const ImportAgentRunRequestTrials$inboundSchema: z.ZodType<
|
|
123
|
+
ImportAgentRunRequestTrials,
|
|
124
|
+
z.ZodTypeDef,
|
|
125
|
+
unknown
|
|
126
|
+
> = z.object({
|
|
127
|
+
task_id: z.string(),
|
|
128
|
+
sample_index: z.number().int().optional(),
|
|
129
|
+
status: ImportAgentRunRequestStatus$inboundSchema.default("completed"),
|
|
130
|
+
prompt: z.string().default(""),
|
|
131
|
+
target: z.nullable(z.string()).optional(),
|
|
132
|
+
output_text: z.nullable(z.string()).optional(),
|
|
133
|
+
scores: z.record(z.number()).optional(),
|
|
134
|
+
error: z.nullable(z.record(z.any())).optional(),
|
|
135
|
+
cost_usd: z.nullable(z.number()).optional(),
|
|
136
|
+
input_tokens: z.nullable(z.number().int()).optional(),
|
|
137
|
+
output_tokens: z.nullable(z.number().int()).optional(),
|
|
138
|
+
cached_tokens: z.nullable(z.number().int()).optional(),
|
|
139
|
+
phase_timings: z.nullable(z.record(z.any())).optional(),
|
|
140
|
+
}).transform((v) => {
|
|
141
|
+
return remap$(v, {
|
|
142
|
+
"task_id": "taskId",
|
|
143
|
+
"sample_index": "sampleIndex",
|
|
144
|
+
"output_text": "outputText",
|
|
145
|
+
"cost_usd": "costUsd",
|
|
146
|
+
"input_tokens": "inputTokens",
|
|
147
|
+
"output_tokens": "outputTokens",
|
|
148
|
+
"cached_tokens": "cachedTokens",
|
|
149
|
+
"phase_timings": "phaseTimings",
|
|
150
|
+
});
|
|
151
|
+
});
|
|
152
|
+
/** @internal */
|
|
153
|
+
export type ImportAgentRunRequestTrials$Outbound = {
|
|
154
|
+
task_id: string;
|
|
155
|
+
sample_index?: number | undefined;
|
|
156
|
+
status: string;
|
|
157
|
+
prompt: string;
|
|
158
|
+
target?: string | null | undefined;
|
|
159
|
+
output_text?: string | null | undefined;
|
|
160
|
+
scores?: { [k: string]: number } | undefined;
|
|
161
|
+
error?: { [k: string]: any } | null | undefined;
|
|
162
|
+
cost_usd?: number | null | undefined;
|
|
163
|
+
input_tokens?: number | null | undefined;
|
|
164
|
+
output_tokens?: number | null | undefined;
|
|
165
|
+
cached_tokens?: number | null | undefined;
|
|
166
|
+
phase_timings?: { [k: string]: any } | null | undefined;
|
|
167
|
+
};
|
|
168
|
+
|
|
169
|
+
/** @internal */
|
|
170
|
+
export const ImportAgentRunRequestTrials$outboundSchema: z.ZodType<
|
|
171
|
+
ImportAgentRunRequestTrials$Outbound,
|
|
172
|
+
z.ZodTypeDef,
|
|
173
|
+
ImportAgentRunRequestTrials
|
|
174
|
+
> = z.object({
|
|
175
|
+
taskId: z.string(),
|
|
176
|
+
sampleIndex: z.number().int().optional(),
|
|
177
|
+
status: ImportAgentRunRequestStatus$outboundSchema.default("completed"),
|
|
178
|
+
prompt: z.string().default(""),
|
|
179
|
+
target: z.nullable(z.string()).optional(),
|
|
180
|
+
outputText: z.nullable(z.string()).optional(),
|
|
181
|
+
scores: z.record(z.number()).optional(),
|
|
182
|
+
error: z.nullable(z.record(z.any())).optional(),
|
|
183
|
+
costUsd: z.nullable(z.number()).optional(),
|
|
184
|
+
inputTokens: z.nullable(z.number().int()).optional(),
|
|
185
|
+
outputTokens: z.nullable(z.number().int()).optional(),
|
|
186
|
+
cachedTokens: z.nullable(z.number().int()).optional(),
|
|
187
|
+
phaseTimings: z.nullable(z.record(z.any())).optional(),
|
|
188
|
+
}).transform((v) => {
|
|
189
|
+
return remap$(v, {
|
|
190
|
+
taskId: "task_id",
|
|
191
|
+
sampleIndex: "sample_index",
|
|
192
|
+
outputText: "output_text",
|
|
193
|
+
costUsd: "cost_usd",
|
|
194
|
+
inputTokens: "input_tokens",
|
|
195
|
+
outputTokens: "output_tokens",
|
|
196
|
+
cachedTokens: "cached_tokens",
|
|
197
|
+
phaseTimings: "phase_timings",
|
|
198
|
+
});
|
|
199
|
+
});
|
|
200
|
+
|
|
201
|
+
export function importAgentRunRequestTrialsToJSON(
|
|
202
|
+
importAgentRunRequestTrials: ImportAgentRunRequestTrials,
|
|
203
|
+
): string {
|
|
204
|
+
return JSON.stringify(
|
|
205
|
+
ImportAgentRunRequestTrials$outboundSchema.parse(
|
|
206
|
+
importAgentRunRequestTrials,
|
|
207
|
+
),
|
|
208
|
+
);
|
|
209
|
+
}
|
|
210
|
+
export function importAgentRunRequestTrialsFromJSON(
|
|
211
|
+
jsonString: string,
|
|
212
|
+
): SafeParseResult<ImportAgentRunRequestTrials, SDKValidationError> {
|
|
213
|
+
return safeParse(
|
|
214
|
+
jsonString,
|
|
215
|
+
(x) => ImportAgentRunRequestTrials$inboundSchema.parse(JSON.parse(x)),
|
|
216
|
+
`Failed to parse 'ImportAgentRunRequestTrials' from JSON`,
|
|
217
|
+
);
|
|
218
|
+
}
|
|
219
|
+
|
|
220
|
+
/** @internal */
|
|
221
|
+
export const ImportAgentRunRequest$inboundSchema: z.ZodType<
|
|
222
|
+
ImportAgentRunRequest,
|
|
223
|
+
z.ZodTypeDef,
|
|
224
|
+
unknown
|
|
225
|
+
> = z.object({
|
|
226
|
+
suite_id: z.string(),
|
|
227
|
+
suite_version: z.number().int().optional(),
|
|
228
|
+
agent_name: z.string(),
|
|
229
|
+
agent_version: z.nullable(z.string()).optional(),
|
|
230
|
+
agent_effort: z.nullable(z.string()).optional(),
|
|
231
|
+
model: z.string(),
|
|
232
|
+
cost_usd: z.nullable(z.number()).optional(),
|
|
233
|
+
created_at: z.number().int().optional(),
|
|
234
|
+
metadata: z.record(z.string()).optional(),
|
|
235
|
+
idempotency_key: z.string().optional(),
|
|
236
|
+
trials: z.array(z.lazy(() => ImportAgentRunRequestTrials$inboundSchema))
|
|
237
|
+
.optional(),
|
|
238
|
+
expected_trials: z.number().int().optional(),
|
|
239
|
+
finalize: z.boolean().default(true),
|
|
240
|
+
}).transform((v) => {
|
|
241
|
+
return remap$(v, {
|
|
242
|
+
"suite_id": "suiteId",
|
|
243
|
+
"suite_version": "suiteVersion",
|
|
244
|
+
"agent_name": "agentName",
|
|
245
|
+
"agent_version": "agentVersion",
|
|
246
|
+
"agent_effort": "agentEffort",
|
|
247
|
+
"cost_usd": "costUsd",
|
|
248
|
+
"created_at": "createdAt",
|
|
249
|
+
"idempotency_key": "idempotencyKey",
|
|
250
|
+
"expected_trials": "expectedTrials",
|
|
251
|
+
});
|
|
252
|
+
});
|
|
253
|
+
/** @internal */
|
|
254
|
+
export type ImportAgentRunRequest$Outbound = {
|
|
255
|
+
suite_id: string;
|
|
256
|
+
suite_version?: number | undefined;
|
|
257
|
+
agent_name: string;
|
|
258
|
+
agent_version?: string | null | undefined;
|
|
259
|
+
agent_effort?: string | null | undefined;
|
|
260
|
+
model: string;
|
|
261
|
+
cost_usd?: number | null | undefined;
|
|
262
|
+
created_at?: number | undefined;
|
|
263
|
+
metadata?: { [k: string]: string } | undefined;
|
|
264
|
+
idempotency_key?: string | undefined;
|
|
265
|
+
trials?: Array<ImportAgentRunRequestTrials$Outbound> | undefined;
|
|
266
|
+
expected_trials?: number | undefined;
|
|
267
|
+
finalize: boolean;
|
|
268
|
+
};
|
|
269
|
+
|
|
270
|
+
/** @internal */
|
|
271
|
+
export const ImportAgentRunRequest$outboundSchema: z.ZodType<
|
|
272
|
+
ImportAgentRunRequest$Outbound,
|
|
273
|
+
z.ZodTypeDef,
|
|
274
|
+
ImportAgentRunRequest
|
|
275
|
+
> = z.object({
|
|
276
|
+
suiteId: z.string(),
|
|
277
|
+
suiteVersion: z.number().int().optional(),
|
|
278
|
+
agentName: z.string(),
|
|
279
|
+
agentVersion: z.nullable(z.string()).optional(),
|
|
280
|
+
agentEffort: z.nullable(z.string()).optional(),
|
|
281
|
+
model: z.string(),
|
|
282
|
+
costUsd: z.nullable(z.number()).optional(),
|
|
283
|
+
createdAt: z.number().int().optional(),
|
|
284
|
+
metadata: z.record(z.string()).optional(),
|
|
285
|
+
idempotencyKey: z.string().optional(),
|
|
286
|
+
trials: z.array(z.lazy(() => ImportAgentRunRequestTrials$outboundSchema))
|
|
287
|
+
.optional(),
|
|
288
|
+
expectedTrials: z.number().int().optional(),
|
|
289
|
+
finalize: z.boolean().default(true),
|
|
290
|
+
}).transform((v) => {
|
|
291
|
+
return remap$(v, {
|
|
292
|
+
suiteId: "suite_id",
|
|
293
|
+
suiteVersion: "suite_version",
|
|
294
|
+
agentName: "agent_name",
|
|
295
|
+
agentVersion: "agent_version",
|
|
296
|
+
agentEffort: "agent_effort",
|
|
297
|
+
costUsd: "cost_usd",
|
|
298
|
+
createdAt: "created_at",
|
|
299
|
+
idempotencyKey: "idempotency_key",
|
|
300
|
+
expectedTrials: "expected_trials",
|
|
301
|
+
});
|
|
302
|
+
});
|
|
303
|
+
|
|
304
|
+
export function importAgentRunRequestToJSON(
|
|
305
|
+
importAgentRunRequest: ImportAgentRunRequest,
|
|
306
|
+
): string {
|
|
307
|
+
return JSON.stringify(
|
|
308
|
+
ImportAgentRunRequest$outboundSchema.parse(importAgentRunRequest),
|
|
309
|
+
);
|
|
310
|
+
}
|
|
311
|
+
export function importAgentRunRequestFromJSON(
|
|
312
|
+
jsonString: string,
|
|
313
|
+
): SafeParseResult<ImportAgentRunRequest, SDKValidationError> {
|
|
314
|
+
return safeParse(
|
|
315
|
+
jsonString,
|
|
316
|
+
(x) => ImportAgentRunRequest$inboundSchema.parse(JSON.parse(x)),
|
|
317
|
+
`Failed to parse 'ImportAgentRunRequest' from JSON`,
|
|
318
|
+
);
|
|
319
|
+
}
|
|
@@ -27,6 +27,7 @@ export * from "./agenttextconfiguration.js";
|
|
|
27
27
|
export * from "./agentversion.js";
|
|
28
28
|
export * from "./agentversionlist.js";
|
|
29
29
|
export * from "./allowedtools.js";
|
|
30
|
+
export * from "./appendagenttrialsrequest.js";
|
|
30
31
|
export * from "./authorizeintent.js";
|
|
31
32
|
export * from "./autochunkingstrategy.js";
|
|
32
33
|
export * from "./autoendpoint.js";
|
|
@@ -233,6 +234,7 @@ export * from "./evaldatasetformat.js";
|
|
|
233
234
|
export * from "./evalfewshotconfig.js";
|
|
234
235
|
export * from "./evalgenerationconfig.js";
|
|
235
236
|
export * from "./evalhistoricalimportobject.js";
|
|
237
|
+
export * from "./evalleaderboardrow.js";
|
|
236
238
|
export * from "./evalleaderboardrungroup.js";
|
|
237
239
|
export * from "./evalmetric.js";
|
|
238
240
|
export * from "./evaloutputextraction.js";
|
|
@@ -242,15 +244,18 @@ export * from "./evalpythongraderexecution.js";
|
|
|
242
244
|
export * from "./evalpythongradermodelaccess.js";
|
|
243
245
|
export * from "./evalpythonpreprocessor.js";
|
|
244
246
|
export * from "./evalrundeletedobject.js";
|
|
247
|
+
export * from "./evalrunoverview.js";
|
|
245
248
|
export * from "./evalrunstatus.js";
|
|
246
249
|
export * from "./evalsampleaudioobject.js";
|
|
247
250
|
export * from "./evalsamplelist.js";
|
|
248
251
|
export * from "./evalsampleobject.js";
|
|
249
252
|
export * from "./evalsamplestatus.js";
|
|
253
|
+
export * from "./evalsampletrajectoryuploadedobject.js";
|
|
250
254
|
export * from "./evalscheduledeletedobject.js";
|
|
251
255
|
export * from "./evalscheduleoverlappolicy.js";
|
|
252
256
|
export * from "./evalschedulerunconfig.js";
|
|
253
257
|
export * from "./evalsuitedeletedobject.js";
|
|
258
|
+
export * from "./evalsuiteleaderboard.js";
|
|
254
259
|
export * from "./evalsuitelist.js";
|
|
255
260
|
export * from "./evalsuitemanifest.js";
|
|
256
261
|
export * from "./evalsuiteobject.js";
|
|
@@ -309,6 +314,7 @@ export * from "./imagegenerationcallwithid.js";
|
|
|
309
314
|
export * from "./imagegenerationrequest.js";
|
|
310
315
|
export * from "./imagegenerationresponse.js";
|
|
311
316
|
export * from "./imagegenerationtooldefinition.js";
|
|
317
|
+
export * from "./importagentrunrequest.js";
|
|
312
318
|
export * from "./importhistoricalevalresultsrequest.js";
|
|
313
319
|
export * from "./incompletedetails.js";
|
|
314
320
|
export * from "./inputaudio.js";
|
|
@@ -607,6 +613,7 @@ export * from "./updatememorystorerequest.js";
|
|
|
607
613
|
export * from "./updatepromptresponse.js";
|
|
608
614
|
export * from "./updateresponserequest.js";
|
|
609
615
|
export * from "./updatevectorstorefileattributesrequest.js";
|
|
616
|
+
export * from "./uploadevalsampletrajectoryrequest.js";
|
|
610
617
|
export * from "./urlcitation.js";
|
|
611
618
|
export * from "./usage.js";
|
|
612
619
|
export * from "./usagecostsresponse.js";
|