themis-eval 0.2.3__py3-none-any.whl → 1.0.0__py3-none-any.whl

{themis_eval-0.2.3.dist-info → themis_eval-1.0.0.dist-info}/RECORD

@@ -1,13 +1,14 @@
-themis/__init__.py,sha256=rQL3njf3i5lnAcmu0HuRzGGMELbA9xX21hzw4HrbIxw,1394
-themis/_version.py,sha256=Tk4OCTQHYoZ61gm9JnkdgajR0vkBHbVm5OUjInzyJug,378
-themis/api.py,sha256=flZTbU-jRcbv7oXcfRKG4hkZjASmWlT52A4PghKj9G0,17700
+themis/__init__.py,sha256=YPexmyPqbANhr0Yzm46FDWgdBLjAX_UoZqywte9WJ84,1476
+themis/_version.py,sha256=b5-7SWk_hYOedKy8IUPxaubXc0KS0hRjhUsY8PkIbl0,818
+themis/api.py,sha256=zK6_RiiYTcXiV6uVQu1fNRdTqm-tXT_Df-CfrLXrIhk,15509
 themis/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+themis/session.py,sha256=H1Pkr71b90sa94YeRCIyffIxE2gQPmtfzo-zNnPLrdM,4311
 themis/backends/__init__.py,sha256=RWM5SnV5FrS_cVjpHHeZZM_b9CgqBu1rPS5DlT5YQTY,578
 themis/backends/execution.py,sha256=RAFuB9ri8TMil5PcnsisypKO2ViyLFXj08P_vjNYguU,6095
-themis/backends/storage.py,sha256=pQp20WagSCl8Vmd-Rgx0hDbpYFhCqARXtvGDw3DPgNQ,8021
+themis/backends/storage.py,sha256=3oMcL5Wliac3INxHVG6DFlbqe6Y-6fjA5eZJOLpxm0w,8168
 themis/cli/__init__.py,sha256=An2DrMHRfmiee5BYJ6TGqvbG7sXWECjjyvEgcoGJ7cE,99
 themis/cli/__main__.py,sha256=df2pOghoSuq18hZmVVikmGhaFSaRe-jeDOnrsu-1QDM,135
-themis/cli/main.py,sha256=AGBFxb1sPLQ-aUAq8RM3YI6gGNs6SdFmBzVSqwp_MSg,15482
+themis/cli/main.py,sha256=ZcB3rS0tMsNCEcHRqjCwfi_JwvQqWZUjZ1TWuSUj4N4,22236
 themis/cli/new_project.py,sha256=D8asV4QbjgQNYvmXt_WhK4nPM-wKHe_K0VJiBdgtO_E,1121
 themis/cli/utils.py,sha256=NAPyFiXspfpx5vBxA8aEcOMmWEDyt-R8ywoHo_8Nr4A,1307
 themis/cli/commands/__init__.py,sha256=CTx7su3qTtq96qxLNclDsE6UM_86NhaS01M9-x9wFiw,287
@@ -16,7 +17,7 @@ themis/cli/commands/comparison.py,sha256=Ki1_MMFFR4vBJkZTeIMWLh-_zdjbtJZurI3YyrE
 themis/cli/commands/config_commands.py,sha256=eL6GtdIllOIHo8GbNN2jOqLn5VUPBuqhnro9ooPxDog,7387
 themis/cli/commands/cost.py,sha256=fFdF6hKIzsbPsyrJ1nt6-2m43PpVGUj8jx5T90tBTNo,7233
 themis/cli/commands/demo.py,sha256=akQqjG-hbUDfeB3bI8K4F5-S0ibJqhflGBFQ5nvdUgE,2135
-themis/cli/commands/info.py,sha256=9maOaw-TFiBpuVhaqlMKukGuZ_zgESetqbMQ1Qdvjxs,2515
+themis/cli/commands/info.py,sha256=njms4dsaede2zyC2yoLbfoN5Bho_JTYciJl3Q7Twiv0,2250
 themis/cli/commands/leaderboard.py,sha256=AVvsYIwZAY18jn3sOq3QD45yNtfdHUEl7eixM4aMCKw,10615
 themis/cli/commands/math_benchmarks.py,sha256=nQ4TcPB7T9O3piAy4_TgrOQOQxh2Q8OyBreK_HoPCeQ,9946
 themis/cli/commands/mcq_benchmarks.py,sha256=Cls5W1jGd7TKizmw07CnZWY5N6ywR8VhJ6jKDnY_cRk,7026
@@ -24,7 +25,7 @@ themis/cli/commands/results.py,sha256=rdN3SaMoFnSfAoAXlfpeCTt3V6MwIp0Dk7FIjvPNF7
 themis/cli/commands/sample_run.py,sha256=r3Ymg5dVHg4IAVJvzoP0ZWUWWUE4Dia1t0062Yhdk9Q,9445
 themis/cli/commands/visualize.py,sha256=ZECkB0NjIltuOeBE-Q1JnndZEMXVzc8KgcrbaP-GSXo,9740
 themis/comparison/__init__.py,sha256=bRI8gDlcjMtnH77R7N5ARioq_V4daJcWWM4DXKsoE1k,679
-themis/comparison/engine.py,sha256=UkzXKmEFI2JiX0y8534oc6JFySxgA5v1emzRcGj33Kk,12133
+themis/comparison/engine.py,sha256=Mw4sQBO2NnKRUOWHNO1XnyjrVvmHX1KXRnl1LUw1hIY,11809
 themis/comparison/reports.py,sha256=126VJbd-lxj8C2YJqul53Fyr-nrZgmbrBsRA6Qkh0ro,10117
 themis/comparison/statistics.py,sha256=eLqKUtKFwSvXnbZax8S0lF8RiSepwYdhnmnDD7DcrZs,12929
 themis/config/__init__.py,sha256=YMdFG1iLvOQUnSPlc_ZJVn5zCCTbIozML64b4qUtGR8,476
@@ -34,7 +35,7 @@ themis/config/runtime.py,sha256=hU69_oND7fJfAOIBJONENmsuf7Y8roO7n-w9OwxzoT8,7475
 themis/config/schema.py,sha256=SMR9QHp8OBkSnb1dHyOgg-IJWSqpXfyAqywnBeMy46M,3196
 themis/core/__init__.py,sha256=S8G1x-39sZ3_NQ5DJ6R1yBTWXp_gO0WxOtVjeB9sTwY,113
 themis/core/conversation.py,sha256=wwO8RS4t4plDR0Sf1KjYv_ejonlvKe0ZwAD-4sfGak8,10155
-themis/core/entities.py,sha256=WV9kiYdZFGxn6oH0lPtqaViY4I8oq-hWi_SmCKjvRnc,4449
+themis/core/entities.py,sha256=a9R3J2wetyUiGcxH6K8pGXfVz_gLh4nsJ81y7cKBj5o,4548
 themis/core/serialization.py,sha256=cxfoSKwcZiNsnR8g_SAJAq1ZLrfLXM4S9_rVEDUT8qs,7071
 themis/core/tools.py,sha256=v0_ctsBCtinZGNC_I4C-h0GUPNM5ZeTi7z-U4iCtyp4,11035
 themis/core/types.py,sha256=I5rr9MMS0irX4lo-xlqGjosx-FjPgT64RzQAraM223A,3652
@@ -59,7 +60,8 @@ themis/datasets/super_gpqa.py,sha256=Mr1ag_FyAk1haxg6_ONX5F84wQYtbSVjV-MlMNmaHlI
 themis/evaluation/__init__.py,sha256=2Jl8tcVxYAsmHNAZev2mPS_mEwZcRzebqSM3QDc2cyY,36
 themis/evaluation/conditional.py,sha256=ayndI7FcwxdIMR8B4ddgcKZd5Jl5NQcBJUp7eXI6Djk,13881
 themis/evaluation/math_verify_utils.py,sha256=vXMvL11-IH16UHZ-mbi_r5hOFz7aUfR1J1laa6qmLMk,2213
-themis/evaluation/pipeline.py,sha256=OOowN59UdOMF2Hwy_G4ky5yzR4ajAnohil6xteWBHqE,1503
+themis/evaluation/metric_pipeline.py,sha256=_JXqf5UySbAyuSiN1waDCBfeek7ArluKvXXkm8qIEvs,329
+themis/evaluation/pipeline.py,sha256=7mQV_sJlPDd6zA9jd2TT5IhdMszzT3ftMLbM7Ww76aw,2217
 themis/evaluation/reports.py,sha256=9om7jzZUtmlMH7EeteXp_98gfHct4x09AyTFy3FSAdQ,8715
 themis/evaluation/extractors/__init__.py,sha256=BanoC_8e0iam-VU7l7uhvhac_6w_JJZYoYE4xXPUrGk,566
 themis/evaluation/extractors/error_taxonomy_extractor.py,sha256=RrRx-23l3LwTdG89kvSQJng438cfYI-IdtOGUD6gEDw,2462
@@ -86,9 +88,9 @@ themis/evaluation/metrics/nlp/bertscore.py,sha256=czlIqYkOTBWsfHiE6U1vkq1KHRQm8p
 themis/evaluation/metrics/nlp/bleu.py,sha256=o_aVkoFPSMmeOLYaHRMamIpSKlYSxrMA1OdntTIUe9g,4436
 themis/evaluation/metrics/nlp/meteor.py,sha256=QZT09s4aiUcVvDJDVPZYjzi5SxXdS2gn2IaOTNmKp78,5076
 themis/evaluation/metrics/nlp/rouge.py,sha256=YL05qluF-KsesHYFRfm5zELJlcvo6RvaKp7xKy6BuLI,4365
-themis/evaluation/pipelines/__init__.py,sha256=5YI1xaUULHisctFxrumN4XRpWYneoonX7nd9zBtsjvQ,384
-themis/evaluation/pipelines/composable_pipeline.py,sha256=nNP9MSvQQJvaSBw5_gO3FeyhGm9So2ZlGqh5qSvE8Ac,10905
-themis/evaluation/pipelines/standard_pipeline.py,sha256=GI5_ImebBuM6D8GpGKLoNq4p3JhTq-ocOThlah8RxME,14754
+themis/evaluation/pipelines/__init__.py,sha256=EXBzgOpi_70yTvdXJr0cERmhBlaqnHZHNcDlPamyQLw,549
+themis/evaluation/pipelines/composable_pipeline.py,sha256=SWv4H1-pcFNAuTu0orkMDgKsdUcom75ded9GhIaHqo0,12838
+themis/evaluation/pipelines/standard_pipeline.py,sha256=Xe7hsOzHg0EBnmslBejb3xJ6Y2pvHF4mrwC_THurQQI,15423
 themis/evaluation/statistics/__init__.py,sha256=TTrScTLAW7EHNq0nbjuJs6iP3_HgDx1yy3EtYXx5JCk,1257
 themis/evaluation/statistics/bootstrap.py,sha256=JUQ8rtzFvW2e41I2pLJ7pqgSEjuJ1r6McyYLI42At9g,2409
 themis/evaluation/statistics/confidence_intervals.py,sha256=CN5EO2gWiSITQubuWuPryngnGXhGwczY9kO3mcG6JVc,3676
@@ -101,20 +103,19 @@ themis/evaluation/strategies/attempt_aware_evaluation_strategy.py,sha256=MFcBdtK
 themis/evaluation/strategies/default_evaluation_strategy.py,sha256=LShW-3Nxg_W4Ln-4qUvHJZqe5YMt64gHoK3uNJYLQNo,693
 themis/evaluation/strategies/evaluation_strategy.py,sha256=YFF-bXkz4Z52GuCw52FcklfEnf8dK8_z_I40DJRcmwE,669
 themis/evaluation/strategies/judge_evaluation_strategy.py,sha256=IRSgnnD2R6qrjiOTyA_PIOHUfQj4FqutkU3pKMth0CQ,2562
-themis/experiment/__init__.py,sha256=dGranqpESugmmfbQlTU9efwspazW6j3vcmAKEtAoWZk,182
-themis/experiment/builder.py,sha256=AEjCDeSOI2B0i0PBjkfY1GUDNrYGTGiqPvt0SxnDQFo,5618
-themis/experiment/cache_manager.py,sha256=Fd8Qxifrmyn8f2zjAyPrLv-ZU4Dcp-MKo8-09BoW7tY,4361
+themis/experiment/__init__.py,sha256=T8BEG9dmabQ-tnVZ5YUd5r-31m6_OY-FDfvUfWhl2hc,190
+themis/experiment/cache_manager.py,sha256=aB9QWcS5PV4y675cMVQehm1Rs2XuK84M0WMMFBYP3Hc,4901
 themis/experiment/comparison.py,sha256=Mr1L5Zj7i87xk9XUQ_UueLTsC-sDZH8YGwLwg_gG0VI,21562
 themis/experiment/cost.py,sha256=flhENfB5WKvyNWwPMDtygNZAv6y_yv4RoClsRz714Hc,10159
-themis/experiment/definitions.py,sha256=oOZBFfEQkSBiZd9CMutCQ5luH6oeUT9yAZFd7fpVjnw,2015
+themis/experiment/definitions.py,sha256=7x18qmQBX88Gn_kxY9n3UefnwvvsRnALAvuEmXSZqXY,2022
 themis/experiment/export.py,sha256=ujwiSvqQhLaO99WHyE8osdnmriHjyIM1C2zKf5o93Cw,29800
-themis/experiment/export_csv.py,sha256=80w3gEGjeLjuiNq539rRP73k3MBtwrzJy90hgE91AKw,6030
 themis/experiment/integration_manager.py,sha256=wTVTjDGcUkzz4tfnwSxa5nK1A4e2FKCPazDYGcdzYS8,3325
 themis/experiment/math.py,sha256=P2E9F_UKI7pb-aXepSztGdr_g309WEMe83zqg1nWO7A,6973
 themis/experiment/mcq.py,sha256=DDB99FHQsU_5vMIRDRhSZ7pReYvVf57wLmmo3OU_An4,6276
-themis/experiment/orchestrator.py,sha256=VeSasDmCXrYlrv1r47I698RUq14vEBR7c_uyZzM01hw,19304
+themis/experiment/orchestrator.py,sha256=fA4nPBrhMwcnsdPc9ETr2m8t0I690UHpujIf8dAOgzU,19527
 themis/experiment/pricing.py,sha256=fTM32yE3L8vahMP4sr1zr7dbp9zYCjiPN4D4VuZ8-q8,9346
-themis/experiment/storage.py,sha256=ujGiQTeRPOfS8hYHB1a7F9t-dQnXquhqomI1vDjqmno,55250
+themis/experiment/share.py,sha256=vbwZ93jFQo0r7uxfzkixwkfwbZZFc0FeFgCj4XS1L4o,8209
+themis/experiment/storage.py,sha256=rqyRyWopwR1Td-9d1O6kslPT1QEepzX29Nl3dtYcDIk,56991
 themis/experiment/visualization.py,sha256=dJYHrp3mntl8CPc5HPI3iKqPztVsddQB3ogRkd_FCNc,18473
 themis/generation/__init__.py,sha256=6KVwCQYMpPIsXNuWDZOGuqHkUkA45lbSacIFn8ZbD4s,36
 themis/generation/agentic_runner.py,sha256=armBQBk7qZDBEwT8HqjIWomYDQm57NfrP5CZJzay2uA,13669
@@ -122,8 +123,8 @@ themis/generation/batching.py,sha256=ddpgpn1pq_EwipvTg-K4WcoSs3c2rbW37jEA5Pa_spo
 themis/generation/clients.py,sha256=6apXCp_VNQosnpnmohTHOhHGXw-VZgsUyLds8MwtYUE,4910
 themis/generation/conversation_runner.py,sha256=kSZHwEvfqzxZ-eQYxmg5OkNZcgEHggZExjad6nBOeTM,7980
 themis/generation/plan.py,sha256=k6_gdKFM12nrKz7ac1c5vTZsFanIKJJgyQ8IhvakDNQ,17158
-themis/generation/router.py,sha256=jZc0KFL483f8TrYtt9yxzFKs-T9CG2CoE2kfOQdHMEc,1082
-themis/generation/runner.py,sha256=pH4Dw77qskMQk3yxEkaHYAl1PItTofI7OXdvevnFiCA,8984
+themis/generation/router.py,sha256=ce3Hbth3JrJxO9dBgo0izCMzvqJ3ARaBNRhIJT-JXRQ,1692
+themis/generation/runner.py,sha256=E7hN4gAxzJLCYwvGjwKCBMhXlFv-RhSRjQs1kYvarXo,9579
 themis/generation/strategies.py,sha256=hjqaVkNycFxJWh_edJ7ilBl7HS6bL-8pYm24zTfoAvg,2975
 themis/generation/templates.py,sha256=ut_6akp8Y6Ey_9O3s64jDbwCB74pw62Zf8URlYcKHkA,2325
 themis/generation/turn_strategies.py,sha256=w33qhzpQbGTsfeOgOgMDovV0wEeXeNZUUBm5yZy1naw,10973
@@ -135,7 +136,7 @@ themis/integrations/huggingface.py,sha256=vrLwYwn65pU4W3FUe0ImCOZxKKlpRshDqMoLFs
 themis/integrations/wandb.py,sha256=LJOPojjlqG05EIPxcjy3QmA15Gxgs1db3encDWVzYYw,2545
 themis/interfaces/__init__.py,sha256=78dNE_eHfFmb9hXNy5sLZ1jOTGWS8TzdVE_eiYQPFVc,5967
 themis/presets/__init__.py,sha256=w58fJcy4eNiE034qHO2xE5pp-H-4LNLXo5hLMuC7wIQ,533
-themis/presets/benchmarks.py,sha256=s9JxRogHwZs8oiuiI7Z7uiUBZXEp3gg7AQZnBvdGieA,12026
+themis/presets/benchmarks.py,sha256=wO5xAVafUuL3HEjPO4eAsfzoeexINxIIGUXiwz_31zU,31752
 themis/presets/models.py,sha256=c6-I_drHa4vMLIajSkCcrFbsJOsauFjY8fU1leBxZLg,5173
 themis/project/__init__.py,sha256=vgLv2nS62yz1XsFSFzFf7eIo6FyQJXpOY9OPRUcTQLQ,465
 themis/project/definitions.py,sha256=vHARw0IjFOWE4RL4mGRwvke36A6GWQGep6cQFIRcpJg,3329
@@ -143,15 +144,21 @@ themis/project/patterns.py,sha256=2J51Q9Jq7X-2N57uexvR191gaZKwusef5vIuIVUQY-E,77
 themis/providers/__init__.py,sha256=K5nG0DsK_YPY0cT9MBLk5BLcLbBo0wBP0vQvLjpAw_Y,189
 themis/providers/registry.py,sha256=Za5Kg3-A-35wS_jiGpPXV2q1k6he_dRIWVqt36dKN-4,1056
 themis/server/__init__.py,sha256=Hp0qGI5nvO4bhLAez3jQxim7H433l72EYE2IA8Xp2hA,731
-themis/server/app.py,sha256=OZ39gCC47AXVqZxroC_4KtIYBYx_rfpde7C25AF3EI0,11166
+themis/server/app.py,sha256=p8UQp2cU5i4DgZC0EhRdRejOfFvlg-93WOC3Z45apHs,11510
+themis/specs/__init__.py,sha256=7eJcZFKgMHflUF4C_Pg1M5B_gpJXCsACEZruv4SHx70,256
+themis/specs/execution.py,sha256=11EeecjIdcKKRdF_kQ_WQkAIxIR8kUdifqRPnLo8XrU,698
+themis/specs/experiment.py,sha256=dnOP1KmsUIcA-VC4AR0bOFsNDMFUBEGLPJVVzPoovR0,971
+themis/specs/storage.py,sha256=5jEqBPv1P2DTXlriUZvfNLnMGn2P-CLx6h82M5RpTmo,383
+themis/storage/__init__.py,sha256=1axYCdtNSE3sYehO6T1hWRjGP_iYnlGl-dBlKiMNX8g,276
+themis/storage/experiment_storage.py,sha256=S7tQ9DtYhWwX2rmxsWrjMW69WGAZORFu49s0eVmwS40,197
 themis/utils/api_generator.py,sha256=3oQ7mGZlFx2Dpm45pMg3rNIqNK2Smj05PjOMXp5RIkQ,10776
 themis/utils/cost_tracking.py,sha256=9_Z2iTfNaQse9G_bnqn4hme4T0fG2W-fxOLEDeF_3VI,11545
 themis/utils/dashboard.py,sha256=2yiIu9_oENglTde_J3G1d5cpQ5VtSnfbUvdliw5Og1E,13008
 themis/utils/logging_utils.py,sha256=buC64X-xOu-2SZ0wVkz3nCXzYVGiqKbxK-8DGSGsAdM,1173
 themis/utils/progress.py,sha256=HS0-yVbRT7Ai9zRlsJcex_OKP6dUiKx1vOp_IsobiHM,2097
 themis/utils/tracing.py,sha256=VTeiRjcW_B5fOOoSeAp37nrmlwP1DiqPcoe6OtIQ7dk,8468
-themis_eval-0.2.3.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
-themis_eval-0.2.3.dist-info/METADATA,sha256=4N7tBOyUi8PAlFT2qJseKIABjHOzkFmLtfqVVUSFz84,15235
-themis_eval-0.2.3.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
-themis_eval-0.2.3.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
-themis_eval-0.2.3.dist-info/RECORD,,
+themis_eval-1.0.0.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
+themis_eval-1.0.0.dist-info/METADATA,sha256=mXVE6G4L1O18OhtsOgACvvjvx4zJ_onVmgF4SO9mJFw,15513
+themis_eval-1.0.0.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+themis_eval-1.0.0.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
+themis_eval-1.0.0.dist-info/RECORD,,

{themis_eval-0.2.3.dist-info → themis_eval-1.0.0.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.10.1)
+Generator: setuptools (80.10.2)
 Root-Is-Purelib: true
 Tag: py3-none-any
 

themis/experiment/builder.py

@@ -1,151 +0,0 @@
-"""Utilities for assembling experiments from reusable components."""
-
-from __future__ import annotations
-
-from pathlib import Path
-from typing import Any, Callable, Mapping, Sequence, Type
-
-from themis.config import schema as config
-from themis.core import entities as core_entities
-from themis.evaluation import pipeline as evaluation_pipeline
-from themis.evaluation import strategies as evaluation_strategies
-from themis.experiment import orchestrator
-from themis.experiment import storage as experiment_storage
-from themis.experiment.cache_manager import CacheManager
-from themis.experiment.definitions import (
-    BuiltExperiment,
-    ExperimentDefinition,
-    ModelBinding,
-)
-from themis.experiment.integration_manager import IntegrationManager
-from themis.generation import plan as generation_plan
-from themis.generation import router as generation_router
-from themis.generation import runner as generation_runner
-from themis.generation import strategies as generation_strategies
-from themis.interfaces import ModelProvider
-from themis.providers import create_provider
-
-
-class ExperimentBuilder:
-    """Composable builder for constructing experiment components."""
-
-    def __init__(
-        self,
-        *,
-        extractor,
-        metrics,
-        runner_cls: Type[
-            generation_runner.GenerationRunner
-        ] = generation_runner.GenerationRunner,
-        runner_kwargs: Mapping[str, Any] | None = None,
-        pipeline_cls: Type[
-            evaluation_pipeline.EvaluationPipeline
-        ] = evaluation_pipeline.EvaluationPipeline,
-        pipeline_kwargs: Mapping[str, Any] | None = None,
-        router_cls: Type[ModelProvider] = generation_router.ProviderRouter,
-        router_kwargs: Mapping[str, Any] | None = None,
-        strategy_resolver: Callable[
-            [core_entities.GenerationTask], generation_strategies.GenerationStrategy
-        ]
-        | None = None,
-        evaluation_strategy_resolver: Callable[
-            [core_entities.GenerationRecord], evaluation_strategies.EvaluationStrategy
-        ]
-        | None = None,
-    ) -> None:
-        self._extractor = extractor
-        self._metrics = list(metrics)
-        self._runner_cls = runner_cls
-        self._runner_kwargs = dict(runner_kwargs or {})
-        self._pipeline_cls = pipeline_cls
-        self._pipeline_kwargs = dict(pipeline_kwargs or {})
-        self._router_cls = router_cls
-        self._router_kwargs = dict(router_kwargs or {})
-        self._strategy_resolver = strategy_resolver
-        self._evaluation_strategy_resolver = evaluation_strategy_resolver
-
-    def build(
-        self,
-        definition: ExperimentDefinition,
-        *,
-        storage_dir: str | Path | None = None,
-    ) -> BuiltExperiment:
-        plan_obj = self._build_plan(definition)
-        router = self._build_router(definition.model_bindings)
-        runner_kwargs = dict(self._runner_kwargs)
-        if self._strategy_resolver is not None:
-            runner_kwargs.setdefault("strategy_resolver", self._strategy_resolver)
-        runner = self._runner_cls(provider=router, **runner_kwargs)
-        pipeline_kwargs = dict(self._pipeline_kwargs)
-        if self._evaluation_strategy_resolver is not None:
-            pipeline_kwargs.setdefault(
-                "strategy_resolver", self._evaluation_strategy_resolver
-            )
-        pipeline = self._pipeline_cls(
-            extractor=self._extractor,
-            metrics=self._metrics,
-            **pipeline_kwargs,
-        )
-
-        # Create storage backend
-        storage = (
-            experiment_storage.ExperimentStorage(storage_dir)
-            if storage_dir is not None
-            else None
-        )
-
-        # Create managers for better separation of concerns
-        cache_manager = CacheManager(
-            storage=storage,
-            enable_resume=True,
-            enable_cache=True,
-        )
-        integration_manager = IntegrationManager(config=config.IntegrationsConfig())
-
-        # Create orchestrator with managers
-        orchestrator_obj = orchestrator.ExperimentOrchestrator(
-            generation_plan=plan_obj,
-            generation_runner=runner,
-            evaluation_pipeline=pipeline,
-            cache_manager=cache_manager,
-            integration_manager=integration_manager,
-        )
-
-        return BuiltExperiment(
-            orchestrator=orchestrator_obj,
-            plan=plan_obj,
-            runner=runner,
-            pipeline=pipeline,
-            storage=storage,
-            router=router,
-        )
-
-    def _build_plan(
-        self, definition: ExperimentDefinition
-    ) -> generation_plan.GenerationPlan:
-        return generation_plan.GenerationPlan(
-            templates=list(definition.templates),
-            models=[binding.spec for binding in definition.model_bindings],
-            sampling_parameters=list(definition.sampling_parameters),
-            dataset_id_field=definition.dataset_id_field,
-            reference_field=definition.reference_field,
-            metadata_fields=tuple(definition.metadata_fields),
-            context_builder=definition.context_builder,
-        )
-
-    def _build_router(self, bindings: Sequence[ModelBinding]) -> ModelProvider:
-        providers: dict[str, ModelProvider] = {}
-        for binding in bindings:
-            providers[binding.spec.identifier] = create_provider(
-                binding.provider_name,
-                **binding.provider_options,
-            )
-        return self._router_cls(providers, **self._router_kwargs)
-
-
-__all__ = [
-    "ExperimentBuilder",
-    "ExperimentDefinition",
-    "ModelBinding",
-    "BuiltExperiment",
-]

themis/experiment/export_csv.py

@@ -1,159 +0,0 @@
-"""CSV export functionality for experiment reports."""
-
-from __future__ import annotations
-
-import csv
-from pathlib import Path
-from typing import MutableMapping, Sequence
-
-from themis.core import entities as core_entities
-from themis.experiment import orchestrator
-
-
-def export_report_csv(
-    report: orchestrator.ExperimentReport,
-    path: str | Path,
-    *,
-    include_failures: bool = True,
-) -> Path:
-    """Write per-sample metrics to a CSV file for offline analysis.
-
-    Args:
-        report: Experiment report to export
-        path: Output path for CSV file
-        include_failures: Whether to include failures column
-
-    Returns:
-        Path to created CSV file
-    """
-    path = Path(path)
-    path.parent.mkdir(parents=True, exist_ok=True)
-    metadata_by_condition, metadata_fields = _collect_sample_metadata(
-        report.generation_results
-    )
-
-    # Create a proper index mapping generation records to their metadata
-    gen_record_index = {}
-    for gen_record in report.generation_results:
-        sample_id = gen_record.task.metadata.get(
-            "dataset_id"
-        ) or gen_record.task.metadata.get("sample_id")
-        prompt_template = gen_record.task.prompt.spec.name
-        model_identifier = gen_record.task.model.identifier
-        sampling_temp = gen_record.task.sampling.temperature
-        sampling_max_tokens = gen_record.task.sampling.max_tokens
-        condition_id = f"{sample_id}_{prompt_template}_{model_identifier}_{sampling_temp}_{sampling_max_tokens}"
-        gen_record_index[condition_id] = gen_record
-
-    metric_names = sorted(report.evaluation_report.metrics.keys())
-    fieldnames = (
-        ["sample_id"] + metadata_fields + [f"metric:{name}" for name in metric_names]
-    )
-    if include_failures:
-        fieldnames.append("failures")
-
-    with path.open("w", encoding="utf-8", newline="") as handle:
-        writer = csv.DictWriter(handle, fieldnames=fieldnames)
-        writer.writeheader()
-
-        # Process evaluation records in the same order as generation records
-        for i, eval_record in enumerate(report.evaluation_report.records):
-            # Find the corresponding generation record by index
-            if i < len(report.generation_results):
-                gen_record = report.generation_results[i]
-                sample_id = gen_record.task.metadata.get(
-                    "dataset_id"
-                ) or gen_record.task.metadata.get("sample_id")
-                prompt_template = gen_record.task.prompt.spec.name
-                model_identifier = gen_record.task.model.identifier
-                sampling_temp = gen_record.task.sampling.temperature
-                sampling_max_tokens = gen_record.task.sampling.max_tokens
-                condition_id = f"{sample_id}_{prompt_template}_{model_identifier}_{sampling_temp}_{sampling_max_tokens}"
-                metadata = metadata_by_condition.get(condition_id, {})
-            else:
-                # Fallback for extra evaluation records
-                sample_id = eval_record.sample_id or ""
-                metadata = {}
-
-            row: dict[str, object] = {"sample_id": sample_id}
-            for field in metadata_fields:
-                row[field] = metadata.get(field, "")
-            score_by_name = {
-                score.metric_name: score.value for score in eval_record.scores
-            }
-            for name in metric_names:
-                row[f"metric:{name}"] = score_by_name.get(name, "")
-            if include_failures:
-                row["failures"] = "; ".join(eval_record.failures)
-            writer.writerow(row)
-    return path
-
-
-def _collect_sample_metadata(
-    records: Sequence[core_entities.GenerationRecord],
-) -> tuple[dict[str, MutableMapping[str, object]], list[str]]:
-    """Collect metadata from generation records.
-
-    Args:
-        records: Generation records
-
-    Returns:
-        Tuple of (metadata by condition ID, list of metadata fields)
-    """
-    metadata: dict[str, MutableMapping[str, object]] = {}
-    for index, record in enumerate(records):
-        sample_id = _extract_sample_id(record.task.metadata)
-        if sample_id is None:
-            sample_id = f"sample-{index}"
-
-        # Create unique identifier for each experimental condition
-        prompt_template = record.task.prompt.spec.name
-        model_identifier = record.task.model.identifier
-        sampling_temp = record.task.sampling.temperature
-        sampling_max_tokens = record.task.sampling.max_tokens
-
-        # Create unique condition key
-        condition_id = f"{sample_id}_{prompt_template}_{model_identifier}_{sampling_temp}_{sampling_max_tokens}"
-
-        # Store metadata with unique condition ID
-        condition_metadata = _metadata_from_task(record)
-        metadata[condition_id] = condition_metadata
-
-    # Collect all field names from all conditions
-    fields = sorted({field for meta in metadata.values() for field in meta.keys()})
-
-    return metadata, fields
-
-
-def _extract_sample_id(metadata: dict[str, object]) -> str | None:
-    """Extract sample ID from metadata.
-
-    Args:
-        metadata: Task metadata
-
-    Returns:
-        Sample ID or None
-    """
-    value = metadata.get("dataset_id") or metadata.get("sample_id")
-    if value is None:
-        return None
-    return str(value)
-
-
-def _metadata_from_task(record: core_entities.GenerationRecord) -> dict[str, object]:
-    """Build metadata dict from generation record.
-
-    Args:
-        record: Generation record
-
-    Returns:
-        Metadata dictionary
-    """
-    metadata = dict(record.task.metadata)
-    metadata.setdefault("model_identifier", record.task.model.identifier)
-    metadata.setdefault("model_provider", record.task.model.provider)
-    metadata.setdefault("prompt_template", record.task.prompt.spec.name)
-    metadata.setdefault("sampling_temperature", record.task.sampling.temperature)
-    metadata.setdefault("sampling_top_p", record.task.sampling.top_p)
-    metadata.setdefault("sampling_max_tokens", record.task.sampling.max_tokens)
-    return metadata