agentinputlab 0.1.0a1__py3-none-any.whl
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.
- agentinputlab/__init__.py +28 -0
- agentinputlab/algorithms/__init__.py +6 -0
- agentinputlab/algorithms/acquire/__init__.py +145 -0
- agentinputlab/algorithms/acquire/adaptive_rag/__init__.py +43 -0
- agentinputlab/algorithms/acquire/adaptive_rag/replay.py +367 -0
- agentinputlab/algorithms/acquire/adaptive_rag/runtime.py +1202 -0
- agentinputlab/algorithms/acquire/adaptive_rag_common_support.py +542 -0
- agentinputlab/algorithms/acquire/hyde.py +296 -0
- agentinputlab/algorithms/acquire/hyde_controlled.py +200 -0
- agentinputlab/algorithms/acquire/hyde_external.py +148 -0
- agentinputlab/algorithms/acquire/ircot.py +28 -0
- agentinputlab/algorithms/acquire/ircot_common_support.py +478 -0
- agentinputlab/algorithms/acquire/ircot_native.py +355 -0
- agentinputlab/algorithms/acquire/ircot_replay.py +327 -0
- agentinputlab/algorithms/acquire/qrag/__init__.py +79 -0
- agentinputlab/algorithms/acquire/qrag/bridge.py +828 -0
- agentinputlab/algorithms/acquire/qrag/common_support.py +891 -0
- agentinputlab/algorithms/acquire/qrag/native_axis.py +71 -0
- agentinputlab/algorithms/acquire/qrag/reader_adapter.py +139 -0
- agentinputlab/algorithms/acquire/r2ag/__init__.py +57 -0
- agentinputlab/algorithms/acquire/r2ag/bridge.py +1312 -0
- agentinputlab/algorithms/acquire/r2ag/common_support.py +1029 -0
- agentinputlab/algorithms/acquire/r2ag/socket_runtime.py +255 -0
- agentinputlab/algorithms/clarify/__init__.py +61 -0
- agentinputlab/algorithms/clarify/_structured.py +64 -0
- agentinputlab/algorithms/clarify/active_task_disambiguation.py +371 -0
- agentinputlab/algorithms/clarify/ask_or_assume/__init__.py +17 -0
- agentinputlab/algorithms/clarify/ask_or_assume/monitor.py +410 -0
- agentinputlab/algorithms/clarify/future_turns.py +379 -0
- agentinputlab/algorithms/clarify/udcs.py +255 -0
- agentinputlab/algorithms/clarify/udcs_deterministic.py +196 -0
- agentinputlab/algorithms/curate/__init__.py +65 -0
- agentinputlab/algorithms/curate/acon/__init__.py +27 -0
- agentinputlab/algorithms/curate/acon/common.py +284 -0
- agentinputlab/algorithms/curate/acon/common_support.py +136 -0
- agentinputlab/algorithms/curate/acon/history_bridge.py +309 -0
- agentinputlab/algorithms/curate/acon/observation_bridge.py +290 -0
- agentinputlab/algorithms/curate/llmlingua/__init__.py +30 -0
- agentinputlab/algorithms/curate/llmlingua/adapters.py +537 -0
- agentinputlab/algorithms/curate/llmlingua/budget.py +48 -0
- agentinputlab/algorithms/curate/llmlingua/controlled.py +175 -0
- agentinputlab/algorithms/curate/llmlingua/multilingual.py +105 -0
- agentinputlab/algorithms/curate/llmlingua/span_map.py +167 -0
- agentinputlab/algorithms/curate/recomp/__init__.py +47 -0
- agentinputlab/algorithms/curate/recomp/external.py +1129 -0
- agentinputlab/algorithms/curate/recomp/selective.py +304 -0
- agentinputlab/algorithms/decompose/__init__.py +107 -0
- agentinputlab/algorithms/decompose/decomp/__init__.py +53 -0
- agentinputlab/algorithms/decompose/decomp/common_support.py +478 -0
- agentinputlab/algorithms/decompose/decomp/planner_component.py +199 -0
- agentinputlab/algorithms/decompose/decomp/worker_bridge.py +416 -0
- agentinputlab/algorithms/decompose/llm_compiler/__init__.py +35 -0
- agentinputlab/algorithms/decompose/llm_compiler/bridge.py +520 -0
- agentinputlab/algorithms/decompose/llm_compiler/common_support.py +504 -0
- agentinputlab/algorithms/decompose/plan_and_budget/__init__.py +43 -0
- agentinputlab/algorithms/decompose/plan_and_budget/cached_soft_budget.py +362 -0
- agentinputlab/algorithms/decompose/plan_and_budget/decomposer_component.py +107 -0
- agentinputlab/algorithms/decompose/plan_and_budget/full.py +205 -0
- agentinputlab/algorithms/decompose/plan_and_budget/hard_cap.py +246 -0
- agentinputlab/algorithms/decompose/plan_and_budget/regenerated_soft_budget.py +448 -0
- agentinputlab/algorithms/decompose/plan_and_solve/__init__.py +33 -0
- agentinputlab/algorithms/decompose/plan_and_solve/plan_component.py +192 -0
- agentinputlab/algorithms/decompose/plan_and_solve/ps_prompt.py +127 -0
- agentinputlab/algorithms/decompose/self_ask/__init__.py +27 -0
- agentinputlab/algorithms/decompose/self_ask/loop.py +362 -0
- agentinputlab/algorithms/decompose/self_ask/planner_component.py +117 -0
- agentinputlab/algorithms/reformulate/__init__.py +85 -0
- agentinputlab/algorithms/reformulate/infocqr/__init__.py +40 -0
- agentinputlab/algorithms/reformulate/infocqr/cache_replay.py +362 -0
- agentinputlab/algorithms/reformulate/infocqr/external.py +675 -0
- agentinputlab/algorithms/reformulate/infocqr/student.py +189 -0
- agentinputlab/algorithms/reformulate/query2doc/__init__.py +42 -0
- agentinputlab/algorithms/reformulate/query2doc/controlled_live.py +190 -0
- agentinputlab/algorithms/reformulate/query2doc/models.py +95 -0
- agentinputlab/algorithms/reformulate/query2doc/official_cached.py +335 -0
- agentinputlab/algorithms/reformulate/query2doc/querygym_bridge.py +223 -0
- agentinputlab/algorithms/reformulate/query2doc/querygym_runtime.py +326 -0
- agentinputlab/algorithms/reformulate/reformer/__init__.py +23 -0
- agentinputlab/algorithms/reformulate/reformer/external.py +417 -0
- agentinputlab/algorithms/reformulate/rrr.py +696 -0
- agentinputlab/annotation/__init__.py +5 -0
- agentinputlab/annotation/independent.py +428 -0
- agentinputlab/annotation/server.py +233 -0
- agentinputlab/annotation/static/app.js +121 -0
- agentinputlab/annotation/static/index.html +60 -0
- agentinputlab/annotation/static/review.css +2 -0
- agentinputlab/annotation/static/review.html +10 -0
- agentinputlab/annotation/static/review.js +11 -0
- agentinputlab/annotation/static/styles.css +55 -0
- agentinputlab/annotation/store.py +444 -0
- agentinputlab/api.py +368 -0
- agentinputlab/artifacts/__init__.py +15 -0
- agentinputlab/artifacts/json_file.py +104 -0
- agentinputlab/artifacts/manifest.py +98 -0
- agentinputlab/artifacts/store.py +157 -0
- agentinputlab/backends/__init__.py +98 -0
- agentinputlab/backends/embedding/__init__.py +25 -0
- agentinputlab/backends/embedding/base.py +145 -0
- agentinputlab/backends/embedding/callable.py +75 -0
- agentinputlab/backends/embedding/replay.py +129 -0
- agentinputlab/backends/llm/__init__.py +31 -0
- agentinputlab/backends/llm/base.py +149 -0
- agentinputlab/backends/llm/callable.py +108 -0
- agentinputlab/backends/llm/decomp_completion.py +88 -0
- agentinputlab/backends/llm/final_content_stop.py +72 -0
- agentinputlab/backends/llm/gpt_oss_final_channel.py +84 -0
- agentinputlab/backends/llm/openai_compatible.py +332 -0
- agentinputlab/backends/llm/transformers.py +171 -0
- agentinputlab/backends/retrieval/__init__.py +35 -0
- agentinputlab/backends/retrieval/base.py +50 -0
- agentinputlab/backends/retrieval/dense.py +777 -0
- agentinputlab/backends/retrieval/memory.py +119 -0
- agentinputlab/backends/retrieval/sqlite_fts.py +155 -0
- agentinputlab/backends/user/__init__.py +15 -0
- agentinputlab/backends/user/base.py +66 -0
- agentinputlab/backends/user/callable.py +54 -0
- agentinputlab/backends/user/replay.py +89 -0
- agentinputlab/backends/vector.py +38 -0
- agentinputlab/benchmarks/__init__.py +246 -0
- agentinputlab/benchmarks/acquisition.py +106 -0
- agentinputlab/benchmarks/ale.py +540 -0
- agentinputlab/benchmarks/appworld_ul.py +669 -0
- agentinputlab/benchmarks/assets.py +226 -0
- agentinputlab/benchmarks/base.py +68 -0
- agentinputlab/benchmarks/beir_archive.py +235 -0
- agentinputlab/benchmarks/beir_sample.py +269 -0
- agentinputlab/benchmarks/bfcl_native.py +245 -0
- agentinputlab/benchmarks/bfcl_v3.py +607 -0
- agentinputlab/benchmarks/cache.py +218 -0
- agentinputlab/benchmarks/clariq.py +203 -0
- agentinputlab/benchmarks/data/registry.json +321 -0
- agentinputlab/benchmarks/data/smoke_assets.json +511 -0
- agentinputlab/benchmarks/data/statebench_v0_8_1_task_ids.json +475 -0
- agentinputlab/benchmarks/data/toolsandbox_165848b_scenario_names.json +1 -0
- agentinputlab/benchmarks/execution.py +1111 -0
- agentinputlab/benchmarks/interactive.py +257 -0
- agentinputlab/benchmarks/ir_query.py +204 -0
- agentinputlab/benchmarks/longbench.py +203 -0
- agentinputlab/benchmarks/paired_replay.py +479 -0
- agentinputlab/benchmarks/preprocess.py +2313 -0
- agentinputlab/benchmarks/production.py +236 -0
- agentinputlab/benchmarks/qa.py +963 -0
- agentinputlab/benchmarks/registry.py +110 -0
- agentinputlab/benchmarks/retrieval.py +459 -0
- agentinputlab/benchmarks/runtime.py +509 -0
- agentinputlab/benchmarks/statebench.py +938 -0
- agentinputlab/benchmarks/tau3.py +725 -0
- agentinputlab/benchmarks/toolsandbox.py +626 -0
- agentinputlab/benchmarks/trec_dl_sample.py +263 -0
- agentinputlab/campaign.py +654 -0
- agentinputlab/campaign_runner.py +953 -0
- agentinputlab/cli/__init__.py +6 -0
- agentinputlab/cli/main.py +933 -0
- agentinputlab/connectors/__init__.py +46 -0
- agentinputlab/connectors/knowledge/__init__.py +4 -0
- agentinputlab/connectors/tools/__init__.py +30 -0
- agentinputlab/connectors/tools/base.py +57 -0
- agentinputlab/connectors/tools/mcp.py +222 -0
- agentinputlab/connectors/tools/memory.py +41 -0
- agentinputlab/connectors/tools/openapi.py +695 -0
- agentinputlab/connectors/tools/schema.py +102 -0
- agentinputlab/contracts/__init__.py +91 -0
- agentinputlab/contracts/availability.py +24 -0
- agentinputlab/contracts/dependency.py +58 -0
- agentinputlab/contracts/enums.py +82 -0
- agentinputlab/contracts/errors.py +63 -0
- agentinputlab/contracts/evidence.py +43 -0
- agentinputlab/contracts/intent.py +110 -0
- agentinputlab/contracts/lifecycle.py +164 -0
- agentinputlab/contracts/patch.py +167 -0
- agentinputlab/contracts/serialization.py +93 -0
- agentinputlab/contracts/state.py +127 -0
- agentinputlab/contracts/trace.py +59 -0
- agentinputlab/contracts/worker.py +42 -0
- agentinputlab/core/__init__.py +7 -0
- agentinputlab/data/__init__.py +32 -0
- agentinputlab/data/checksums.py +9 -0
- agentinputlab/data/downloader.py +235 -0
- agentinputlab/data/http.py +171 -0
- agentinputlab/data/index.py +21 -0
- agentinputlab/data/overlap.py +28 -0
- agentinputlab/data/preprocess.py +32 -0
- agentinputlab/data/storage.py +57 -0
- agentinputlab/data/trec_sample.py +427 -0
- agentinputlab/evaluation/__init__.py +5 -0
- agentinputlab/evaluation/outcomes.py +32 -0
- agentinputlab/experiments/__init__.py +278 -0
- agentinputlab/experiments/analysis.py +981 -0
- agentinputlab/experiments/artifacts.py +687 -0
- agentinputlab/experiments/automatic_measurement.py +678 -0
- agentinputlab/experiments/auxiliary_matrix.py +203 -0
- agentinputlab/experiments/bindings.py +430 -0
- agentinputlab/experiments/blueprints.py +147 -0
- agentinputlab/experiments/builders/__init__.py +45 -0
- agentinputlab/experiments/builders/executable_external_payload.py +423 -0
- agentinputlab/experiments/builders/external_process_arena.py +1018 -0
- agentinputlab/experiments/builders/registry_intervention.py +463 -0
- agentinputlab/experiments/cli.py +158 -0
- agentinputlab/experiments/controls.py +340 -0
- agentinputlab/experiments/driver_guard.py +205 -0
- agentinputlab/experiments/drivers/__init__.py +89 -0
- agentinputlab/experiments/drivers/external_arena.py +2908 -0
- agentinputlab/experiments/drivers/method_process.py +662 -0
- agentinputlab/experiments/drivers/parity_process.py +1484 -0
- agentinputlab/experiments/drivers/static_production.py +1574 -0
- agentinputlab/experiments/executor.py +448 -0
- agentinputlab/experiments/loader.py +605 -0
- agentinputlab/experiments/measurement.py +675 -0
- agentinputlab/experiments/paper_analysis.py +1818 -0
- agentinputlab/experiments/parity_identity.py +605 -0
- agentinputlab/experiments/plan.py +277 -0
- agentinputlab/experiments/production_builders.py +3739 -0
- agentinputlab/experiments/readiness.py +432 -0
- agentinputlab/experiments/schema.py +514 -0
- agentinputlab/experiments/secrets.py +177 -0
- agentinputlab/experiments/stateful_analysis.py +1738 -0
- agentinputlab/experiments/transfer_analysis.py +1316 -0
- agentinputlab/experiments/unseen_analysis.py +1025 -0
- agentinputlab/integrations/__init__.py +1 -0
- agentinputlab/integrations/action_space/__init__.py +1 -0
- agentinputlab/integrations/external_workers/__init__.py +76 -0
- agentinputlab/integrations/external_workers/ace.py +431 -0
- agentinputlab/integrations/external_workers/base.py +181 -0
- agentinputlab/integrations/external_workers/decomp_live.py +249 -0
- agentinputlab/integrations/external_workers/decomp_planner.py +30 -0
- agentinputlab/integrations/external_workers/infocqr_student.py +280 -0
- agentinputlab/integrations/external_workers/ircot.py +576 -0
- agentinputlab/integrations/external_workers/llmcompiler_live.py +360 -0
- agentinputlab/integrations/external_workers/longllmlingua.py +162 -0
- agentinputlab/integrations/external_workers/method.py +314 -0
- agentinputlab/integrations/external_workers/process.py +678 -0
- agentinputlab/integrations/external_workers/supervisor.py +1071 -0
- agentinputlab/integrations/full_architecture/__init__.py +1 -0
- agentinputlab/integrations/full_architecture/hyde.py +212 -0
- agentinputlab/integrations/full_architecture/plan_and_budget.py +188 -0
- agentinputlab/integrations/live_history.py +222 -0
- agentinputlab/integrations/prompt_optimization/__init__.py +1 -0
- agentinputlab/integrations/training/__init__.py +1 -0
- agentinputlab/metrics/__init__.py +50 -0
- agentinputlab/metrics/catalog.py +313 -0
- agentinputlab/metrics/harm.py +95 -0
- agentinputlab/metrics/paired.py +40 -0
- agentinputlab/metrics/qa.py +76 -0
- agentinputlab/metrics/text.py +33 -0
- agentinputlab/metrics/third_party.py +161 -0
- agentinputlab/models/__init__.py +4 -0
- agentinputlab/observability/__init__.py +56 -0
- agentinputlab/observability/budget.py +316 -0
- agentinputlab/observability/instrumentation.py +754 -0
- agentinputlab/observability/models.py +403 -0
- agentinputlab/observability/monitor.py +380 -0
- agentinputlab/optimizers/__init__.py +1 -0
- agentinputlab/optimizers/clarify/__init__.py +12 -0
- agentinputlab/optimizers/context/__init__.py +22 -0
- agentinputlab/optimizers/decompose/__init__.py +64 -0
- agentinputlab/optimizers/evidence/__init__.py +22 -0
- agentinputlab/optimizers/rewrite/__init__.py +30 -0
- agentinputlab/parity/__init__.py +5 -0
- agentinputlab/parity/base.py +206 -0
- agentinputlab/production_campaign.py +763 -0
- agentinputlab/py.typed +1 -0
- agentinputlab/recipes/__init__.py +333 -0
- agentinputlab/recipes/ace.py +418 -0
- agentinputlab/recipes/ace_bfcl.py +273 -0
- agentinputlab/recipes/ace_common_support.py +169 -0
- agentinputlab/recipes/ace_native_plan.py +322 -0
- agentinputlab/recipes/ace_online.py +711 -0
- agentinputlab/recipes/ace_retrieval.py +330 -0
- agentinputlab/recipes/ace_statebench.py +382 -0
- agentinputlab/recipes/ace_tau.py +463 -0
- agentinputlab/recipes/ace_toolsandbox.py +432 -0
- agentinputlab/recipes/acon_guideline.py +22 -0
- agentinputlab/recipes/adaptive_rag.py +822 -0
- agentinputlab/recipes/llmlingua2_inference.py +146 -0
- agentinputlab/recipes/llmlingua2_training.py +796 -0
- agentinputlab/recipes/mce.py +1486 -0
- agentinputlab/recipes/mce_final.py +335 -0
- agentinputlab/recipes/mce_text_fit.py +499 -0
- agentinputlab/recipes/mce_text_only.py +298 -0
- agentinputlab/recipes/qrag.py +729 -0
- agentinputlab/recipes/qrag_inference.py +364 -0
- agentinputlab/recipes/r2ag_features.py +236 -0
- agentinputlab/recipes/r2ag_inference.py +503 -0
- agentinputlab/recipes/r2ag_training.py +948 -0
- agentinputlab/recipes/recomp_abstractive.py +592 -0
- agentinputlab/recipes/recomp_extractive.py +525 -0
- agentinputlab/recipes/reformer.py +17 -0
- agentinputlab/recipes/rqrag.py +1516 -0
- agentinputlab/recipes/rqrag_common_support.py +193 -0
- agentinputlab/recipes/rqrag_completion.py +147 -0
- agentinputlab/recipes/skillopt.py +462 -0
- agentinputlab/recipes/skillopt_common_support.py +154 -0
- agentinputlab/recipes/skillopt_environment.py +165 -0
- agentinputlab/recipes/skillopt_environment_cli.py +5 -0
- agentinputlab/recipes/skillopt_retrieval.py +188 -0
- agentinputlab/recipes/smartsearch.py +897 -0
- agentinputlab/recipes/smartsearch_curriculum.py +283 -0
- agentinputlab/recipes/smartsearch_grpo_contracts.py +58 -0
- agentinputlab/recipes/smartsearch_grpo_data.py +314 -0
- agentinputlab/recipes/smartsearch_grpo_meter.py +324 -0
- agentinputlab/recipes/smartsearch_grpo_prefix.py +144 -0
- agentinputlab/recipes/smartsearch_grpo_rollout.py +119 -0
- agentinputlab/recipes/smartsearch_grpo_tools.py +364 -0
- agentinputlab/recipes/smartsearch_grpo_training.py +305 -0
- agentinputlab/recipes/smartsearch_grpo_transport.py +248 -0
- agentinputlab/recipes/smartsearch_grpo_worker.py +364 -0
- agentinputlab/recipes/smartsearch_inference.py +524 -0
- agentinputlab/recipes/smartsearch_native_data.py +172 -0
- agentinputlab/recipes/smartsearch_policy.py +616 -0
- agentinputlab/recipes/smartsearch_policy_worker.py +9 -0
- agentinputlab/recipes/smartsearch_sampling.py +232 -0
- agentinputlab/recipes/smartsearch_teacher.py +203 -0
- agentinputlab/recipes/smartsearch_training.py +265 -0
- agentinputlab/registry/__init__.py +46 -0
- agentinputlab/registry/beta.py +727 -0
- agentinputlab/registry/data/algorithms.json +1959 -0
- agentinputlab/registry/data/auxiliary_method_profiles.json +34 -0
- agentinputlab/registry/loader.py +100 -0
- agentinputlab/registry/result_manifest.py +184 -0
- agentinputlab/registry/runtime.py +195 -0
- agentinputlab/registry/schema.py +449 -0
- agentinputlab/runner/__init__.py +30 -0
- agentinputlab/runner/balance.py +248 -0
- agentinputlab/runner/cache.py +135 -0
- agentinputlab/runner/composite.py +130 -0
- agentinputlab/runner/engine.py +285 -0
- agentinputlab/runner/runtime_guard.py +268 -0
- agentinputlab/runner/worker.py +80 -0
- agentinputlab/runtime_bootstrap.py +288 -0
- agentinputlab/selectors/__init__.py +196 -0
- agentinputlab/selectors/advanced.py +1514 -0
- agentinputlab/selectors/artifacts.py +1087 -0
- agentinputlab/selectors/base.py +112 -0
- agentinputlab/selectors/campaigns.py +1893 -0
- agentinputlab/selectors/catalog.py +86 -0
- agentinputlab/selectors/discovery.py +1860 -0
- agentinputlab/selectors/evaluation.py +1348 -0
- agentinputlab/selectors/features.py +383 -0
- agentinputlab/selectors/learned.py +502 -0
- agentinputlab/selectors/no_op.py +28 -0
- agentinputlab/selectors/outcome_table.py +475 -0
- agentinputlab/selectors/producer.py +853 -0
- agentinputlab/selectors/rules.py +109 -0
- agentinputlab/selectors/runtime.py +420 -0
- agentinputlab/studies/__init__.py +16 -0
- agentinputlab/studies/design.py +165 -0
- agentinputlab/studies/paired.py +322 -0
- agentinputlab/studio/__init__.py +6 -0
- agentinputlab/studio/campaigns.py +819 -0
- agentinputlab/studio/engine.py +1408 -0
- agentinputlab/studio/model_backends.py +425 -0
- agentinputlab/studio/prepared.py +410 -0
- agentinputlab/studio/server.py +613 -0
- agentinputlab/studio/static/app.js +397 -0
- agentinputlab/studio/static/campaigns.html +175 -0
- agentinputlab/studio/static/campaigns.js +426 -0
- agentinputlab/studio/static/index.html +192 -0
- agentinputlab/studio/static/result.html +103 -0
- agentinputlab/studio/static/result.js +273 -0
- agentinputlab/studio/static/styles.css +467 -0
- agentinputlab-0.1.0a1.data/data/share/agentinputlab/CITATION.cff +22 -0
- agentinputlab-0.1.0a1.dist-info/METADATA +392 -0
- agentinputlab-0.1.0a1.dist-info/RECORD +469 -0
- agentinputlab-0.1.0a1.dist-info/WHEEL +5 -0
- agentinputlab-0.1.0a1.dist-info/entry_points.txt +6 -0
- agentinputlab-0.1.0a1.dist-info/licenses/LICENSE +21 -0
- agentinputlab-0.1.0a1.dist-info/licenses/NOTICE +31 -0
- agentinputlab-0.1.0a1.dist-info/licenses/THIRD_PARTY_NOTICES.md +94 -0
- agentinputlab-0.1.0a1.dist-info/top_level.txt +2 -0
- workers/README.md +94 -0
- workers/ace/README.md +261 -0
- workers/ace/__init__.py +1 -0
- workers/ace/compatibility_manifest.json +39 -0
- workers/ace/driver.py +1682 -0
- workers/acon/README.md +215 -0
- workers/acon/__init__.py +5 -0
- workers/acon/compatibility_manifest.json +93 -0
- workers/acon/driver.py +1859 -0
- workers/acon/fitted.py +395 -0
- workers/acon/guideline.py +470 -0
- workers/acon/guideline_source_manifest.json +20 -0
- workers/adaptive_rag/README.md +96 -0
- workers/adaptive_rag/__init__.py +15 -0
- workers/adaptive_rag/compatibility_manifest.json +61 -0
- workers/adaptive_rag/driver.py +1470 -0
- workers/adaptive_rag/fixture_runtime.py +163 -0
- workers/adaptive_rag/runtime.py +252 -0
- workers/ask_or_assume/README.md +124 -0
- workers/ask_or_assume/__init__.py +1 -0
- workers/ask_or_assume/compatibility_manifest.json +36 -0
- workers/ask_or_assume/driver.py +1227 -0
- workers/decomp/README.md +133 -0
- workers/decomp/UPSTREAM_LICENSE.Apache-2.0.txt +201 -0
- workers/decomp/__init__.py +5 -0
- workers/decomp/compatibility_manifest.json +205 -0
- workers/decomp/driver.py +1015 -0
- workers/decomp/live.py +137 -0
- workers/decomp/runtime.py +231 -0
- workers/infocqr/README.md +106 -0
- workers/infocqr/__init__.py +1 -0
- workers/infocqr/compatibility_manifest.json +84 -0
- workers/infocqr/driver.py +1766 -0
- workers/ircot/README.md +167 -0
- workers/ircot/__init__.py +7 -0
- workers/ircot/compatibility_manifest.json +121 -0
- workers/ircot/controlled.py +457 -0
- workers/ircot/driver.py +1373 -0
- workers/ircot/fixture_bindings.json +39 -0
- workers/ircot/fixture_config.json +48 -0
- workers/ircot/fixture_prompt.txt +2 -0
- workers/ircot/native_qa.py +398 -0
- workers/ircot/qa_compatibility_manifest.json +57 -0
- workers/ircot/runtime.py +97 -0
- workers/llmcompiler/README.md +160 -0
- workers/llmcompiler/__init__.py +1 -0
- workers/llmcompiler/compatibility_manifest.json +45 -0
- workers/llmcompiler/driver.py +1514 -0
- workers/llmcompiler/export_prompts.py +74 -0
- workers/llmcompiler/live.py +174 -0
- workers/llmcompiler/patch_manifest.json +147 -0
- workers/mce/README.md +198 -0
- workers/mce/__init__.py +29 -0
- workers/mce/compatibility_manifest.json +115 -0
- workers/mce/driver.py +2585 -0
- workers/mce/final_eval.py +173 -0
- workers/official_arenas/README.md +25 -0
- workers/official_arenas/__init__.py +1 -0
- workers/official_arenas/capture_records.py +110 -0
- workers/official_arenas/driver.py +619 -0
- workers/official_arenas/protocol.py +334 -0
- workers/official_arenas/providers/__init__.py +1 -0
- workers/official_arenas/providers/_live_history.py +228 -0
- workers/official_arenas/providers/_message_compat.py +77 -0
- workers/official_arenas/providers/appworld_ul.py +196 -0
- workers/official_arenas/providers/fixture.py +112 -0
- workers/official_arenas/providers/statebench.py +960 -0
- workers/official_arenas/providers/tau3.py +1672 -0
- workers/official_arenas/providers/tau_retail.py +586 -0
- workers/official_arenas/providers/toolsandbox.py +1321 -0
- workers/official_arenas/providers/tooltalk.py +699 -0
- workers/official_arenas/retail_capture.py +188 -0
- workers/official_arenas/retail_driver.py +275 -0
- workers/qrag/README.md +172 -0
- workers/qrag/__init__.py +1 -0
- workers/qrag/compatibility_manifest.json +64 -0
- workers/qrag/driver.py +673 -0
- workers/r2ag/README.md +154 -0
- workers/r2ag/__init__.py +5 -0
- workers/r2ag/compatibility_manifest.json +55 -0
- workers/r2ag/driver.py +1287 -0
- workers/r2ag/feature_schema.json +8 -0
- workers/r2ag/runtime.py +589 -0
- workers/recomp/README.md +120 -0
- workers/recomp/__init__.py +1 -0
- workers/recomp/compatibility_manifest.json +43 -0
- workers/recomp/driver.py +1653 -0
- workers/reformer/README.md +125 -0
- workers/reformer/__init__.py +1 -0
- workers/reformer/compatibility_manifest.json +68 -0
- workers/reformer/driver.py +1459 -0
- workers/reformer/induction.py +457 -0
- workers/rrr/README.md +126 -0
- workers/rrr/__init__.py +1 -0
- workers/rrr/compatibility_manifest.json +57 -0
- workers/rrr/driver.py +1133 -0
- workers/smartsearch/README.md +331 -0
- workers/smartsearch/__init__.py +1 -0
- workers/smartsearch/compatibility_manifest.json +41 -0
- workers/smartsearch/driver.py +1643 -0
|
@@ -0,0 +1,382 @@
|
|
|
1
|
+
"""Portable STATE-Bench tool-plan feedback for native ACE fitting/online APIs.
|
|
2
|
+
|
|
3
|
+
This adapter scores a noninteractive plan on original training tasks. It does
|
|
4
|
+
not replace the benchmark's closed-loop heldout agent or its native judge.
|
|
5
|
+
"""
|
|
6
|
+
|
|
7
|
+
from __future__ import annotations
|
|
8
|
+
|
|
9
|
+
from collections.abc import Mapping, Sequence
|
|
10
|
+
from pathlib import Path
|
|
11
|
+
from urllib.parse import urlsplit
|
|
12
|
+
|
|
13
|
+
from agentinputlab.api import _state
|
|
14
|
+
from agentinputlab.contracts import ValidationError
|
|
15
|
+
from agentinputlab.contracts.serialization import canonical_json, to_jsonable
|
|
16
|
+
|
|
17
|
+
from .ace import ACEExample
|
|
18
|
+
from .ace_native_plan import (
|
|
19
|
+
PLAN_INSTRUCTION,
|
|
20
|
+
NativePlanFeedback,
|
|
21
|
+
_save,
|
|
22
|
+
_sha,
|
|
23
|
+
native_process_scope,
|
|
24
|
+
source_file_pins,
|
|
25
|
+
verify_native_checkout,
|
|
26
|
+
)
|
|
27
|
+
|
|
28
|
+
SCHEMA = "agentinputlab.ace.STATE.native_feedback_input.v1"
|
|
29
|
+
|
|
30
|
+
|
|
31
|
+
def _provider():
|
|
32
|
+
from workers.official_arenas.providers.statebench import StateBenchProvider
|
|
33
|
+
|
|
34
|
+
return StateBenchProvider()
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
def _native_inputs(checkout, configuration, rows):
|
|
38
|
+
from state_bench.domain import get_domain_config
|
|
39
|
+
from state_bench.paths import domain_tasks_dir
|
|
40
|
+
from state_bench.protocol import load_default_protocol, load_split_task_ids
|
|
41
|
+
from state_bench.schemas import TaskDefinition
|
|
42
|
+
|
|
43
|
+
protocol = load_default_protocol()
|
|
44
|
+
if protocol.validate_prompt_hashes():
|
|
45
|
+
raise ValidationError("STATE protocol prompt checksums differ")
|
|
46
|
+
domains, tasks, files = (
|
|
47
|
+
{},
|
|
48
|
+
{},
|
|
49
|
+
[
|
|
50
|
+
checkout / "state_bench" / n
|
|
51
|
+
for n in (
|
|
52
|
+
"client.py",
|
|
53
|
+
"domain.py",
|
|
54
|
+
"env_loader.py",
|
|
55
|
+
"orchestrator.py",
|
|
56
|
+
"protocol.py",
|
|
57
|
+
"scoring.py",
|
|
58
|
+
)
|
|
59
|
+
],
|
|
60
|
+
)
|
|
61
|
+
for row in rows:
|
|
62
|
+
protected = row["protected_checkpoint"]["protected"]
|
|
63
|
+
name, task_id = protected["domain"], protected["task_id"]
|
|
64
|
+
if name not in configuration["domains"] or task_id not in load_split_task_ids(
|
|
65
|
+
name, "train", protocol.split_version
|
|
66
|
+
):
|
|
67
|
+
raise ValidationError("STATE feedback must use original official training tasks")
|
|
68
|
+
path = domain_tasks_dir(name) / (task_id + ".json")
|
|
69
|
+
task = TaskDefinition.load(path)
|
|
70
|
+
domain = get_domain_config(name)
|
|
71
|
+
state = _state(row["state"])
|
|
72
|
+
if (
|
|
73
|
+
row["case_id"] != name + ":" + task_id
|
|
74
|
+
or state.task_id != task_id
|
|
75
|
+
or state.query != task.opening_message
|
|
76
|
+
or protected["user_id"] != task.user_id
|
|
77
|
+
or protected["protocol_id"] != protocol.protocol_id
|
|
78
|
+
or protected["locked_dependency_hashes"] != configuration["locked_dependency_hashes"]
|
|
79
|
+
):
|
|
80
|
+
raise ValidationError("STATE fitting checkpoint, task or dependency binding differs")
|
|
81
|
+
instruction = domain.agent_system_prompt.format(now=task.now, user_id=task.user_id)
|
|
82
|
+
names = {t.get("name") or t.get("function", {}).get("name") for t in domain.tool_schemas}
|
|
83
|
+
if state.instruction != instruction or not names or names != set(state.tool_names):
|
|
84
|
+
raise ValidationError("STATE visible policy or native tool catalog differs")
|
|
85
|
+
domains[row["case_id"]], tasks[row["case_id"]] = domain, task
|
|
86
|
+
files.append(path)
|
|
87
|
+
files.extend(p for p in domain.prompts_dir.rglob("*") if p.is_file())
|
|
88
|
+
return domains, tasks, source_file_pins(checkout, files)
|
|
89
|
+
|
|
90
|
+
|
|
91
|
+
def build_ace_statebench_feedback_input(
|
|
92
|
+
*,
|
|
93
|
+
checkout,
|
|
94
|
+
captures: Sequence[Mapping],
|
|
95
|
+
native_configuration: Mapping,
|
|
96
|
+
train_ids: Sequence[str],
|
|
97
|
+
validation_ids: Sequence[str],
|
|
98
|
+
heldout_ids: Sequence[str],
|
|
99
|
+
target_model: Mapping,
|
|
100
|
+
environment_model: Mapping,
|
|
101
|
+
output_path,
|
|
102
|
+
):
|
|
103
|
+
"""Freeze existing official captures without a model call or generated labels.
|
|
104
|
+
|
|
105
|
+
IDs are native checkpoint IDs (``domain:task_id``). The split is checked
|
|
106
|
+
against the pinned protocol, not inferred from the caller's labels.
|
|
107
|
+
"""
|
|
108
|
+
provider = _provider()
|
|
109
|
+
root = verify_native_checkout(checkout, revision=provider.source_revision)
|
|
110
|
+
configuration = provider._validate_config(to_jsonable(native_configuration))
|
|
111
|
+
if configuration["split"] != "train":
|
|
112
|
+
raise ValidationError("STATE ACE fitting requires the official train split")
|
|
113
|
+
ids = [*train_ids, *validation_ids]
|
|
114
|
+
if (
|
|
115
|
+
not train_ids
|
|
116
|
+
or not validation_ids
|
|
117
|
+
or len(ids) != len(set(ids))
|
|
118
|
+
or set(ids) & set(heldout_ids)
|
|
119
|
+
or any(not isinstance(x, str) or not x for x in ids)
|
|
120
|
+
):
|
|
121
|
+
raise ValidationError("STATE ACE groups must be disjoint nonempty fitting IDs")
|
|
122
|
+
selected = {r["checkpoint_id"]: r for r in captures if r["checkpoint_id"] in ids}
|
|
123
|
+
if len(selected) != len(ids) or len([r for r in captures if r["checkpoint_id"] in ids]) != len(
|
|
124
|
+
ids
|
|
125
|
+
):
|
|
126
|
+
raise ValidationError("STATE capture IDs are missing or duplicated")
|
|
127
|
+
rows = []
|
|
128
|
+
for role, group in (("train", train_ids), ("validation", validation_ids)):
|
|
129
|
+
for key in group:
|
|
130
|
+
capture = to_jsonable(selected[key])
|
|
131
|
+
state = _state(capture["visible_state"])
|
|
132
|
+
rows.append(
|
|
133
|
+
dict(
|
|
134
|
+
case_id=key,
|
|
135
|
+
role=role,
|
|
136
|
+
official_split="train",
|
|
137
|
+
state=to_jsonable(state),
|
|
138
|
+
state_hash=state.state_hash,
|
|
139
|
+
protected_checkpoint=capture,
|
|
140
|
+
)
|
|
141
|
+
)
|
|
142
|
+
with native_process_scope(root, package="state_bench"):
|
|
143
|
+
_, _, pins = _native_inputs(root, configuration, rows)
|
|
144
|
+
if (
|
|
145
|
+
not isinstance(target_model, Mapping)
|
|
146
|
+
or not isinstance(environment_model, Mapping)
|
|
147
|
+
or target_model.get("model_id") != configuration.get("agent_model")
|
|
148
|
+
or environment_model.get("model_id") != configuration.get("evaluator_model")
|
|
149
|
+
):
|
|
150
|
+
raise ValidationError("STATE target/judge model fingerprints must match the deployment")
|
|
151
|
+
value = dict(
|
|
152
|
+
schema=SCHEMA,
|
|
153
|
+
benchmark=provider.benchmark_id,
|
|
154
|
+
native_source_revision=provider.source_revision,
|
|
155
|
+
source_files=pins,
|
|
156
|
+
native_configuration=configuration,
|
|
157
|
+
fitting_rows=rows,
|
|
158
|
+
heldout_ids=list(heldout_ids),
|
|
159
|
+
source_official_split="train",
|
|
160
|
+
target_model=to_jsonable(target_model),
|
|
161
|
+
environment_model=to_jsonable(environment_model),
|
|
162
|
+
builder_model_calls=0,
|
|
163
|
+
ToolTalk_not_used=True,
|
|
164
|
+
)
|
|
165
|
+
path = Path(output_path).expanduser().absolute()
|
|
166
|
+
if path.resolve().is_relative_to(root):
|
|
167
|
+
raise ValidationError("STATE feedback input must be outside the clean checkout")
|
|
168
|
+
if path.exists() or path.is_symlink():
|
|
169
|
+
raise ValidationError("STATE feedback input must be a new file")
|
|
170
|
+
_save(path, value)
|
|
171
|
+
return dict(path=str(path), sha256=_sha(path))
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
class ACEStateBenchFeedback(NativePlanFeedback):
|
|
175
|
+
"""Hash-pinned, split-isolated native STATE tools and judge feedback."""
|
|
176
|
+
|
|
177
|
+
processor_id = "agentinputlab.ace.STATE.native_tool_plan_feedback.v1"
|
|
178
|
+
metric = "task_completion_pass"
|
|
179
|
+
|
|
180
|
+
def __init__(
|
|
181
|
+
self, *, checkout, source_input, output_root, environment_base_url=None, mode="offline"
|
|
182
|
+
):
|
|
183
|
+
self._initialize(
|
|
184
|
+
source_input=source_input, checkout=checkout, output_root=output_root, mode=mode
|
|
185
|
+
)
|
|
186
|
+
self.provider = _provider()
|
|
187
|
+
if (
|
|
188
|
+
self.source.get("schema") != SCHEMA
|
|
189
|
+
or self.source.get("benchmark") != self.provider.benchmark_id
|
|
190
|
+
):
|
|
191
|
+
raise ValidationError("STATE ACE feedback input schema or benchmark differs")
|
|
192
|
+
self.checkout = verify_native_checkout(checkout, revision=self.provider.source_revision)
|
|
193
|
+
if self.source["native_source_revision"] != self.provider.source_revision:
|
|
194
|
+
raise ValidationError("STATE ACE feedback source revision differs")
|
|
195
|
+
self._verify_source()
|
|
196
|
+
self.configuration = self.provider._validate_config(self.source["native_configuration"])
|
|
197
|
+
if (
|
|
198
|
+
self.configuration["split"] != "train"
|
|
199
|
+
or self.source["source_official_split"] != "train"
|
|
200
|
+
):
|
|
201
|
+
raise ValidationError("STATE ACE feedback excludes official test inputs")
|
|
202
|
+
with native_process_scope(self.checkout, package="state_bench"):
|
|
203
|
+
self.domains, self.tasks, actual = _native_inputs(
|
|
204
|
+
self.checkout, self.configuration, list(self.rows.values())
|
|
205
|
+
)
|
|
206
|
+
if actual != self.source["source_files"]:
|
|
207
|
+
raise ValidationError("STATE task, source or prompt file binding differs")
|
|
208
|
+
if (
|
|
209
|
+
self.source["target_model"].get("model_id") != self.configuration["agent_model"]
|
|
210
|
+
or self.source["environment_model"].get("model_id")
|
|
211
|
+
!= self.configuration["evaluator_model"]
|
|
212
|
+
):
|
|
213
|
+
raise ValidationError("STATE feedback target/judge deployment identity differs")
|
|
214
|
+
self.environment_base_url = environment_base_url or self.configuration["evaluator_base_url"]
|
|
215
|
+
url = (
|
|
216
|
+
urlsplit(self.environment_base_url)
|
|
217
|
+
if isinstance(self.environment_base_url, str)
|
|
218
|
+
else None
|
|
219
|
+
)
|
|
220
|
+
if (
|
|
221
|
+
url is None
|
|
222
|
+
or url.scheme not in {"http", "https"}
|
|
223
|
+
or not url.hostname
|
|
224
|
+
or url.username is not None
|
|
225
|
+
or url.password is not None
|
|
226
|
+
or url.query
|
|
227
|
+
or url.fragment
|
|
228
|
+
):
|
|
229
|
+
raise ValidationError("STATE feedback requires an explicit judge HTTP deployment")
|
|
230
|
+
self.tool_names = {k: set(s.tool_names) for k, s in self.states.items()}
|
|
231
|
+
self._create_output()
|
|
232
|
+
|
|
233
|
+
def fingerprint(self):
|
|
234
|
+
self._verify_source()
|
|
235
|
+
return dict(
|
|
236
|
+
processor_id=self.processor_id,
|
|
237
|
+
mode=self.mode,
|
|
238
|
+
source_input=self.source_input,
|
|
239
|
+
source_files=self.source["source_files"],
|
|
240
|
+
source_revision=self.provider.source_revision,
|
|
241
|
+
scorer_id=self.provider.scorer_id,
|
|
242
|
+
scorer_revision=self.provider.scorer_revision,
|
|
243
|
+
target_model=self.source["target_model"],
|
|
244
|
+
environment_model=self.source["environment_model"],
|
|
245
|
+
environment_base_url=self.environment_base_url,
|
|
246
|
+
provider_source_sha256=_sha(
|
|
247
|
+
__import__(self.provider.__module__, fromlist=["__file__"]).__file__
|
|
248
|
+
),
|
|
249
|
+
task_adapter_source_sha256=_sha(__file__),
|
|
250
|
+
feedback_metric=self.metric,
|
|
251
|
+
source_official_split="train",
|
|
252
|
+
groups={
|
|
253
|
+
role: [
|
|
254
|
+
dict(case_id=k, state_hash=r["state_hash"])
|
|
255
|
+
for k, r in self.rows.items()
|
|
256
|
+
if r["role"] == role
|
|
257
|
+
]
|
|
258
|
+
for role in ("train", "validation")
|
|
259
|
+
},
|
|
260
|
+
noninteractive_text_plan_NOT_native_closed_loop_agent=True,
|
|
261
|
+
heldout_feedback_NOT_used=True,
|
|
262
|
+
transductive_NOT_independent_heldout=self.mode == "online",
|
|
263
|
+
native_user_simulator_calls=0,
|
|
264
|
+
judge_costs_separate=True,
|
|
265
|
+
automatic_retry=False,
|
|
266
|
+
)
|
|
267
|
+
|
|
268
|
+
def _example(self, key, split):
|
|
269
|
+
state = self.states[key]
|
|
270
|
+
visible = dict(
|
|
271
|
+
query=state.query,
|
|
272
|
+
instruction=state.instruction,
|
|
273
|
+
messages=state.messages,
|
|
274
|
+
context=state.context,
|
|
275
|
+
tool_names=state.tool_names,
|
|
276
|
+
native_function_tools=to_jsonable(self.domains[key].tool_schemas),
|
|
277
|
+
)
|
|
278
|
+
return ACEExample(
|
|
279
|
+
example_id=key,
|
|
280
|
+
input=dict(question=PLAN_INSTRUCTION + canonical_json(visible)),
|
|
281
|
+
expected_output=key,
|
|
282
|
+
split=split,
|
|
283
|
+
metadata=dict(
|
|
284
|
+
source_state_hash=state.state_hash,
|
|
285
|
+
task_processor=self.processor_id,
|
|
286
|
+
evaluator_target_is_opaque_case_id=True,
|
|
287
|
+
),
|
|
288
|
+
)
|
|
289
|
+
|
|
290
|
+
def _native_score(self, key, plan, folder):
|
|
291
|
+
with native_process_scope(self.checkout, package="state_bench"):
|
|
292
|
+
from state_bench.env_loader import load_task_environment
|
|
293
|
+
from state_bench.protocol import load_default_protocol
|
|
294
|
+
from state_bench.schemas import StateDiff
|
|
295
|
+
from state_bench.scoring import (
|
|
296
|
+
TaskRequirementsJudge,
|
|
297
|
+
combine_task_completion,
|
|
298
|
+
evaluate_state_requirements,
|
|
299
|
+
)
|
|
300
|
+
|
|
301
|
+
import workers.official_arenas.providers.statebench as module
|
|
302
|
+
|
|
303
|
+
task, domain = self.tasks[key], self.domains[key]
|
|
304
|
+
data, _ = load_task_environment(domain, task)
|
|
305
|
+
env = domain.environment_class(data.deep_copy(), now=task.now)
|
|
306
|
+
before = env.get_full_snapshot()
|
|
307
|
+
calls = []
|
|
308
|
+
conversation = [dict(role="user", content=self.states[key].query)]
|
|
309
|
+
for action in plan["actions"]:
|
|
310
|
+
result = env.tool_handlers[action["name"]](action["arguments"])
|
|
311
|
+
call = dict(
|
|
312
|
+
name=action["name"], arguments=action["arguments"], result=to_jsonable(result)
|
|
313
|
+
)
|
|
314
|
+
calls.append(call)
|
|
315
|
+
conversation.append(dict(role="assistant", content="", tool_calls=[call]))
|
|
316
|
+
conversation.append(
|
|
317
|
+
dict(role="tool", name=action["name"], content=canonical_json(result))
|
|
318
|
+
)
|
|
319
|
+
conversation.append(dict(role="assistant", content=plan["response"]))
|
|
320
|
+
diff = StateDiff.compute(before, env.get_full_snapshot())
|
|
321
|
+
state_result = evaluate_state_requirements(task, diff)
|
|
322
|
+
_save(
|
|
323
|
+
folder / "before-native-judge.json",
|
|
324
|
+
dict(
|
|
325
|
+
case_id=key,
|
|
326
|
+
source_state_hash=self.states[key].state_hash,
|
|
327
|
+
plan=plan,
|
|
328
|
+
native_tool_calls=calls,
|
|
329
|
+
conversation=conversation,
|
|
330
|
+
state_diff=diff.to_dict(),
|
|
331
|
+
state_requirements_score=state_result.score,
|
|
332
|
+
evaluated_agent_model_calls=0,
|
|
333
|
+
noninteractive_plan_execution_NOT_native_agent_rollout=True,
|
|
334
|
+
),
|
|
335
|
+
)
|
|
336
|
+
cfg = self.configuration
|
|
337
|
+
model = cfg["evaluator_model"]
|
|
338
|
+
client = module._openai_compatible_client(
|
|
339
|
+
model=model,
|
|
340
|
+
base_url=self.environment_base_url,
|
|
341
|
+
api_key_var=cfg["evaluator_api_key_var"],
|
|
342
|
+
)
|
|
343
|
+
module._install_statebench_judge_json_compat(client)
|
|
344
|
+
protocol = load_default_protocol()
|
|
345
|
+
effort = module.normalize_qwen38_reasoning_effort(
|
|
346
|
+
model, cfg["score_reasoning_effort"] or protocol.judge_reasoning_effort
|
|
347
|
+
)
|
|
348
|
+
judge = TaskRequirementsJudge(
|
|
349
|
+
client=client,
|
|
350
|
+
prompts_dir=domain.prompts_dir,
|
|
351
|
+
system_prompt=domain.judge_system_prompt,
|
|
352
|
+
reasoning_effort=effort,
|
|
353
|
+
)
|
|
354
|
+
try:
|
|
355
|
+
task_result = judge.evaluate(task, conversation, calls, diff)
|
|
356
|
+
finally:
|
|
357
|
+
client._client.close()
|
|
358
|
+
if task_result is None:
|
|
359
|
+
raise ValidationError("original native STATE task judge did not complete")
|
|
360
|
+
score = combine_task_completion(state_result, task_result)
|
|
361
|
+
if score is None:
|
|
362
|
+
raise ValidationError("original native STATE completion is absent")
|
|
363
|
+
value = dict(
|
|
364
|
+
status="native_STATE_tool_plan_scored",
|
|
365
|
+
case_id=key,
|
|
366
|
+
native_metrics=dict(
|
|
367
|
+
task_completion_pass=float(score),
|
|
368
|
+
state_requirements_met=float(state_result.score),
|
|
369
|
+
task_requirements_met=float(task_result.score),
|
|
370
|
+
),
|
|
371
|
+
evaluator_agent_model_calls=0,
|
|
372
|
+
native_user_simulator_calls=0,
|
|
373
|
+
task_judge_model=model,
|
|
374
|
+
original_judge_controls=True,
|
|
375
|
+
)
|
|
376
|
+
_save(folder / "native-score.json", value)
|
|
377
|
+
return value
|
|
378
|
+
|
|
379
|
+
|
|
380
|
+
def create_statebench_feedback(**configuration):
|
|
381
|
+
"""Trusted startup factory for ``python -m workers.ace.driver``."""
|
|
382
|
+
return ACEStateBenchFeedback(**configuration)
|