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 +1,85 @@
|
|
|
1
|
+
"""Query reformulation algorithms."""
|
|
2
|
+
|
|
3
|
+
from .infocqr import (
|
|
4
|
+
INFOCQR_POSTPROCESS_SHA256,
|
|
5
|
+
INFOCQR_PROMPT_SOURCE_SHA256,
|
|
6
|
+
INFOCQR_PROTOCOL_VERSION,
|
|
7
|
+
InfoCQRCache,
|
|
8
|
+
InfoCQREditReplay,
|
|
9
|
+
InfoCQRExternalControls,
|
|
10
|
+
InfoCQRFewShotReplay,
|
|
11
|
+
InfoCQRLiveTeacherExternal,
|
|
12
|
+
InfoCQRMode,
|
|
13
|
+
InfoCQRMountedCacheReplay,
|
|
14
|
+
InfoCQRRecord,
|
|
15
|
+
InfoCQRZeroShotReplay,
|
|
16
|
+
conversation_view_hash,
|
|
17
|
+
)
|
|
18
|
+
from .query2doc import (
|
|
19
|
+
OfficialQuery2DocRow,
|
|
20
|
+
Query2DocCache,
|
|
21
|
+
Query2DocOfficialCached,
|
|
22
|
+
QueryGymLiveBridge,
|
|
23
|
+
QueryGymMode,
|
|
24
|
+
QueryGymRawResult,
|
|
25
|
+
QueryGymRequest,
|
|
26
|
+
QueryGymRuntime,
|
|
27
|
+
compose_sparse_expansion,
|
|
28
|
+
select_demonstration_ids,
|
|
29
|
+
)
|
|
30
|
+
from .reformer import (
|
|
31
|
+
REFORMER_PROMPT_ARTIFACT_SHA256,
|
|
32
|
+
REFORMER_UPSTREAM_COMMIT,
|
|
33
|
+
ReFormeRControlledConcat,
|
|
34
|
+
ReFormeRControlledROnly,
|
|
35
|
+
ReFormeRControls,
|
|
36
|
+
ReFormeRExternal,
|
|
37
|
+
ReFormeRPaperConcat,
|
|
38
|
+
ReFormeRRepoNative,
|
|
39
|
+
)
|
|
40
|
+
from .rrr import (
|
|
41
|
+
RRR_UPSTREAM_COMMIT,
|
|
42
|
+
RRRCommonSupport,
|
|
43
|
+
RRRCommonSupportConfig,
|
|
44
|
+
RRRPromptControls,
|
|
45
|
+
RRRPromptExternal,
|
|
46
|
+
)
|
|
47
|
+
|
|
48
|
+
__all__ = [
|
|
49
|
+
"INFOCQR_POSTPROCESS_SHA256",
|
|
50
|
+
"INFOCQR_PROMPT_SOURCE_SHA256",
|
|
51
|
+
"INFOCQR_PROTOCOL_VERSION",
|
|
52
|
+
"REFORMER_PROMPT_ARTIFACT_SHA256",
|
|
53
|
+
"REFORMER_UPSTREAM_COMMIT",
|
|
54
|
+
"RRR_UPSTREAM_COMMIT",
|
|
55
|
+
"InfoCQRCache",
|
|
56
|
+
"InfoCQREditReplay",
|
|
57
|
+
"InfoCQRExternalControls",
|
|
58
|
+
"InfoCQRFewShotReplay",
|
|
59
|
+
"InfoCQRLiveTeacherExternal",
|
|
60
|
+
"InfoCQRMode",
|
|
61
|
+
"InfoCQRMountedCacheReplay",
|
|
62
|
+
"InfoCQRRecord",
|
|
63
|
+
"InfoCQRZeroShotReplay",
|
|
64
|
+
"OfficialQuery2DocRow",
|
|
65
|
+
"Query2DocCache",
|
|
66
|
+
"Query2DocOfficialCached",
|
|
67
|
+
"QueryGymLiveBridge",
|
|
68
|
+
"QueryGymMode",
|
|
69
|
+
"QueryGymRawResult",
|
|
70
|
+
"QueryGymRequest",
|
|
71
|
+
"QueryGymRuntime",
|
|
72
|
+
"RRRCommonSupport",
|
|
73
|
+
"RRRCommonSupportConfig",
|
|
74
|
+
"RRRPromptControls",
|
|
75
|
+
"RRRPromptExternal",
|
|
76
|
+
"ReFormeRControlledConcat",
|
|
77
|
+
"ReFormeRControlledROnly",
|
|
78
|
+
"ReFormeRControls",
|
|
79
|
+
"ReFormeRExternal",
|
|
80
|
+
"ReFormeRPaperConcat",
|
|
81
|
+
"ReFormeRRepoNative",
|
|
82
|
+
"compose_sparse_expansion",
|
|
83
|
+
"conversation_view_hash",
|
|
84
|
+
"select_demonstration_ids",
|
|
85
|
+
]
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
"""Checksum-locked InfoCQR artifact replay profiles."""
|
|
2
|
+
|
|
3
|
+
from .cache_replay import (
|
|
4
|
+
InfoCQRCache,
|
|
5
|
+
InfoCQREditReplay,
|
|
6
|
+
InfoCQRFewShotReplay,
|
|
7
|
+
InfoCQRMode,
|
|
8
|
+
InfoCQRRecord,
|
|
9
|
+
InfoCQRZeroShotReplay,
|
|
10
|
+
conversation_view_hash,
|
|
11
|
+
)
|
|
12
|
+
from .external import (
|
|
13
|
+
INFOCQR_POSTPROCESS_SHA256,
|
|
14
|
+
INFOCQR_PROMPT_SOURCE_SHA256,
|
|
15
|
+
INFOCQR_PROTOCOL_VERSION,
|
|
16
|
+
InfoCQRCommonSupportZeroShot,
|
|
17
|
+
InfoCQRExternalControls,
|
|
18
|
+
InfoCQRLiveTeacherExternal,
|
|
19
|
+
InfoCQRMountedCacheReplay,
|
|
20
|
+
)
|
|
21
|
+
from .student import InfoCQRStudentExternal, InfoCQRStudentRequest
|
|
22
|
+
|
|
23
|
+
__all__ = [
|
|
24
|
+
"INFOCQR_POSTPROCESS_SHA256",
|
|
25
|
+
"INFOCQR_PROMPT_SOURCE_SHA256",
|
|
26
|
+
"INFOCQR_PROTOCOL_VERSION",
|
|
27
|
+
"InfoCQRCache",
|
|
28
|
+
"InfoCQRCommonSupportZeroShot",
|
|
29
|
+
"InfoCQREditReplay",
|
|
30
|
+
"InfoCQRExternalControls",
|
|
31
|
+
"InfoCQRFewShotReplay",
|
|
32
|
+
"InfoCQRLiveTeacherExternal",
|
|
33
|
+
"InfoCQRMode",
|
|
34
|
+
"InfoCQRMountedCacheReplay",
|
|
35
|
+
"InfoCQRRecord",
|
|
36
|
+
"InfoCQRStudentExternal",
|
|
37
|
+
"InfoCQRStudentRequest",
|
|
38
|
+
"InfoCQRZeroShotReplay",
|
|
39
|
+
"conversation_view_hash",
|
|
40
|
+
]
|
|
@@ -0,0 +1,362 @@
|
|
|
1
|
+
"""Checksum-bound replay of externally supplied InfoCQR rewrites.
|
|
2
|
+
|
|
3
|
+
No upstream source or cache is redistributed. A caller supplies a checksum-
|
|
4
|
+
verified artifact and the adapter validates task, split, conversation view,
|
|
5
|
+
profile, post-processing identity, and leakage policy before applying it.
|
|
6
|
+
Byte verification does not establish authorship or the generating model;
|
|
7
|
+
retain those identities and historical costs in the source receipt.
|
|
8
|
+
"""
|
|
9
|
+
|
|
10
|
+
from __future__ import annotations
|
|
11
|
+
|
|
12
|
+
import hashlib
|
|
13
|
+
import json
|
|
14
|
+
from collections.abc import Mapping, Sequence
|
|
15
|
+
from dataclasses import dataclass, field
|
|
16
|
+
from enum import Enum
|
|
17
|
+
from pathlib import Path
|
|
18
|
+
from types import MappingProxyType
|
|
19
|
+
from typing import Any
|
|
20
|
+
|
|
21
|
+
from agentinputlab.contracts import (
|
|
22
|
+
AlgorithmOutputError,
|
|
23
|
+
ArtifactError,
|
|
24
|
+
ExecutionTrace,
|
|
25
|
+
FrozenAgentState,
|
|
26
|
+
InputPatch,
|
|
27
|
+
InterventionContext,
|
|
28
|
+
PatchOp,
|
|
29
|
+
PatchOperation,
|
|
30
|
+
SemanticFamily,
|
|
31
|
+
TraceEvent,
|
|
32
|
+
TracePhase,
|
|
33
|
+
ValidationError,
|
|
34
|
+
)
|
|
35
|
+
from agentinputlab.contracts.serialization import content_hash, freeze_mapping
|
|
36
|
+
|
|
37
|
+
INFOCQR_UPSTREAM_COMMIT = "2953c461aa0c6817b1b7656dbd5168050227b31d"
|
|
38
|
+
_GOLD_KEYS = frozenset({"answer", "gold", "qrels", "reference", "truth_rewrite"})
|
|
39
|
+
|
|
40
|
+
|
|
41
|
+
class InfoCQRMode(str, Enum):
|
|
42
|
+
RW_ZSL = "rw_zsl"
|
|
43
|
+
RW_FSL = "rw_fsl"
|
|
44
|
+
EDIT = "edit"
|
|
45
|
+
|
|
46
|
+
|
|
47
|
+
@dataclass(frozen=True, slots=True)
|
|
48
|
+
class InfoCQRRecord:
|
|
49
|
+
record_id: str
|
|
50
|
+
split: str
|
|
51
|
+
mode: InfoCQRMode
|
|
52
|
+
source_query: str
|
|
53
|
+
conversation_hash: str
|
|
54
|
+
rewrite: str
|
|
55
|
+
raw_candidate: str = ""
|
|
56
|
+
postprocess_id: str = "upstream_post_processing"
|
|
57
|
+
fallback_reason: str = ""
|
|
58
|
+
first_turn_truth_rewrite: bool = False
|
|
59
|
+
metadata: Mapping[str, Any] = field(default_factory=dict)
|
|
60
|
+
|
|
61
|
+
def __post_init__(self) -> None:
|
|
62
|
+
if not self.record_id.strip() or not self.split.strip():
|
|
63
|
+
raise ArtifactError("InfoCQR records require record_id and split")
|
|
64
|
+
if not isinstance(self.mode, InfoCQRMode):
|
|
65
|
+
try:
|
|
66
|
+
object.__setattr__(self, "mode", InfoCQRMode(str(self.mode)))
|
|
67
|
+
except ValueError as exc:
|
|
68
|
+
raise ArtifactError("InfoCQR record has an unsupported mode") from exc
|
|
69
|
+
if not self.source_query.strip() or not self.rewrite.strip():
|
|
70
|
+
raise ArtifactError("InfoCQR source_query and rewrite must not be empty")
|
|
71
|
+
if len(self.conversation_hash) != 64:
|
|
72
|
+
raise ArtifactError("InfoCQR conversation_hash must be a SHA-256 hex digest")
|
|
73
|
+
try:
|
|
74
|
+
int(self.conversation_hash, 16)
|
|
75
|
+
except ValueError as exc:
|
|
76
|
+
raise ArtifactError("InfoCQR conversation_hash must be hexadecimal") from exc
|
|
77
|
+
if not self.postprocess_id.strip():
|
|
78
|
+
raise ArtifactError("InfoCQR postprocess_id must not be empty")
|
|
79
|
+
if not isinstance(self.first_turn_truth_rewrite, bool):
|
|
80
|
+
raise ArtifactError("InfoCQR first_turn_truth_rewrite must be boolean")
|
|
81
|
+
lowered = {str(key).lower() for key in self.metadata}
|
|
82
|
+
if lowered & _GOLD_KEYS:
|
|
83
|
+
raise ArtifactError("InfoCQR solver-visible metadata contains a gold field")
|
|
84
|
+
object.__setattr__(self, "metadata", freeze_mapping(self.metadata))
|
|
85
|
+
|
|
86
|
+
@classmethod
|
|
87
|
+
def from_mapping(cls, value: Mapping[str, Any]) -> InfoCQRRecord:
|
|
88
|
+
metadata = value.get("metadata", {})
|
|
89
|
+
if not isinstance(metadata, Mapping):
|
|
90
|
+
raise ArtifactError("InfoCQR record metadata must be a mapping")
|
|
91
|
+
try:
|
|
92
|
+
return cls(
|
|
93
|
+
record_id=str(value["record_id"]),
|
|
94
|
+
split=str(value["split"]),
|
|
95
|
+
mode=InfoCQRMode(str(value["mode"])),
|
|
96
|
+
source_query=str(value["source_query"]),
|
|
97
|
+
conversation_hash=str(value["conversation_hash"]),
|
|
98
|
+
rewrite=str(value["rewrite"]),
|
|
99
|
+
raw_candidate=str(value.get("raw_candidate", "")),
|
|
100
|
+
postprocess_id=str(value.get("postprocess_id", "upstream_post_processing")),
|
|
101
|
+
fallback_reason=str(value.get("fallback_reason", "")),
|
|
102
|
+
first_turn_truth_rewrite=value.get("first_turn_truth_rewrite", False),
|
|
103
|
+
metadata=dict(metadata),
|
|
104
|
+
)
|
|
105
|
+
except (KeyError, ValueError) as exc:
|
|
106
|
+
raise ArtifactError("InfoCQR record is missing a required field") from exc
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
class InfoCQRCache:
|
|
110
|
+
"""Immutable keyed view of one externally supplied cache artifact."""
|
|
111
|
+
|
|
112
|
+
def __init__(
|
|
113
|
+
self,
|
|
114
|
+
records: Sequence[InfoCQRRecord],
|
|
115
|
+
*,
|
|
116
|
+
artifact_id: str,
|
|
117
|
+
artifact_sha256: str,
|
|
118
|
+
dataset_revision: str,
|
|
119
|
+
) -> None:
|
|
120
|
+
if not artifact_id.strip() or not dataset_revision.strip():
|
|
121
|
+
raise ArtifactError("InfoCQR cache requires artifact and dataset identities")
|
|
122
|
+
if len(artifact_sha256) != 64:
|
|
123
|
+
raise ArtifactError("InfoCQR artifact_sha256 must be a SHA-256 digest")
|
|
124
|
+
try:
|
|
125
|
+
int(artifact_sha256, 16)
|
|
126
|
+
except ValueError as exc:
|
|
127
|
+
raise ArtifactError("InfoCQR artifact_sha256 must be hexadecimal") from exc
|
|
128
|
+
indexed: dict[tuple[str, str, str], InfoCQRRecord] = {}
|
|
129
|
+
for record in records:
|
|
130
|
+
if not isinstance(record, InfoCQRRecord):
|
|
131
|
+
raise ArtifactError("InfoCQR cache entries must be InfoCQRRecord instances")
|
|
132
|
+
key = (record.mode.value, record.split, record.record_id)
|
|
133
|
+
if key in indexed:
|
|
134
|
+
raise ArtifactError(f"duplicate InfoCQR cache key: {key}")
|
|
135
|
+
indexed[key] = record
|
|
136
|
+
if not indexed:
|
|
137
|
+
raise ArtifactError("InfoCQR cache must not be empty")
|
|
138
|
+
self._records = MappingProxyType(dict(indexed))
|
|
139
|
+
self.artifact_id = artifact_id
|
|
140
|
+
self.artifact_sha256 = artifact_sha256.lower()
|
|
141
|
+
self.dataset_revision = dataset_revision
|
|
142
|
+
|
|
143
|
+
@classmethod
|
|
144
|
+
def from_jsonl(
|
|
145
|
+
cls,
|
|
146
|
+
path: str | Path,
|
|
147
|
+
*,
|
|
148
|
+
expected_sha256: str,
|
|
149
|
+
artifact_id: str,
|
|
150
|
+
dataset_revision: str,
|
|
151
|
+
) -> InfoCQRCache:
|
|
152
|
+
raw = Path(path).read_bytes()
|
|
153
|
+
actual = hashlib.sha256(raw).hexdigest()
|
|
154
|
+
if actual != expected_sha256.lower():
|
|
155
|
+
raise ArtifactError("InfoCQR cache checksum mismatch")
|
|
156
|
+
records: list[InfoCQRRecord] = []
|
|
157
|
+
for line_number, line in enumerate(raw.decode("utf-8").splitlines(), start=1):
|
|
158
|
+
if not line.strip():
|
|
159
|
+
continue
|
|
160
|
+
try:
|
|
161
|
+
value = json.loads(line)
|
|
162
|
+
except json.JSONDecodeError as exc:
|
|
163
|
+
raise ArtifactError(f"invalid InfoCQR JSONL line {line_number}") from exc
|
|
164
|
+
if not isinstance(value, Mapping):
|
|
165
|
+
raise ArtifactError(f"InfoCQR JSONL line {line_number} must be an object")
|
|
166
|
+
records.append(InfoCQRRecord.from_mapping(value))
|
|
167
|
+
return cls(
|
|
168
|
+
records,
|
|
169
|
+
artifact_id=artifact_id,
|
|
170
|
+
artifact_sha256=actual,
|
|
171
|
+
dataset_revision=dataset_revision,
|
|
172
|
+
)
|
|
173
|
+
|
|
174
|
+
def get(self, mode: InfoCQRMode, split: str, record_id: str) -> InfoCQRRecord:
|
|
175
|
+
try:
|
|
176
|
+
return self._records[(mode.value, split, record_id)]
|
|
177
|
+
except KeyError as exc:
|
|
178
|
+
raise ArtifactError(
|
|
179
|
+
f"InfoCQR cache has no {mode.value}/{split}/{record_id} record"
|
|
180
|
+
) from exc
|
|
181
|
+
|
|
182
|
+
|
|
183
|
+
def conversation_view_hash(state: FrozenAgentState) -> str:
|
|
184
|
+
"""Hash only the conversation fields consumed by the InfoCQR serializer."""
|
|
185
|
+
|
|
186
|
+
return content_hash(
|
|
187
|
+
{
|
|
188
|
+
"messages": state.messages,
|
|
189
|
+
"current_query": state.query,
|
|
190
|
+
"new_context": state.metadata.get("new_context", ""),
|
|
191
|
+
}
|
|
192
|
+
)
|
|
193
|
+
|
|
194
|
+
|
|
195
|
+
@dataclass(slots=True)
|
|
196
|
+
class _InfoCQRReplay:
|
|
197
|
+
cache: InfoCQRCache
|
|
198
|
+
split: str
|
|
199
|
+
mode: InfoCQRMode
|
|
200
|
+
release_id: str
|
|
201
|
+
allow_first_turn_truth_rewrite: bool = False
|
|
202
|
+
expected_postprocess_id: str = "upstream_post_processing"
|
|
203
|
+
method_id: str = "infocqr"
|
|
204
|
+
family: SemanticFamily = SemanticFamily.REFORMULATE
|
|
205
|
+
|
|
206
|
+
def __post_init__(self) -> None:
|
|
207
|
+
if not isinstance(self.cache, InfoCQRCache):
|
|
208
|
+
raise ValidationError("InfoCQR replay requires an InfoCQRCache")
|
|
209
|
+
if not self.split.strip() or not self.expected_postprocess_id.strip():
|
|
210
|
+
raise ValidationError("InfoCQR split and postprocess identity must not be empty")
|
|
211
|
+
|
|
212
|
+
def propose(self, state: FrozenAgentState, context: InterventionContext) -> InputPatch:
|
|
213
|
+
del context
|
|
214
|
+
record_id = state.task_id or str(state.metadata.get("record_id", ""))
|
|
215
|
+
if not record_id:
|
|
216
|
+
raise ArtifactError("InfoCQR replay requires task_id or metadata.record_id")
|
|
217
|
+
record = self.cache.get(self.mode, self.split, record_id)
|
|
218
|
+
if record.source_query != state.query:
|
|
219
|
+
raise ArtifactError("InfoCQR cached source query mismatch")
|
|
220
|
+
observed_conversation_hash = conversation_view_hash(state)
|
|
221
|
+
if record.conversation_hash != observed_conversation_hash:
|
|
222
|
+
raise ArtifactError("InfoCQR cached conversation view mismatch")
|
|
223
|
+
if record.postprocess_id != self.expected_postprocess_id:
|
|
224
|
+
raise ArtifactError("InfoCQR post-processing identity mismatch")
|
|
225
|
+
if record.first_turn_truth_rewrite and not self.allow_first_turn_truth_rewrite:
|
|
226
|
+
raise ArtifactError(
|
|
227
|
+
"InfoCQR first-turn Truth_rewrite is gold-derived and disabled by policy"
|
|
228
|
+
)
|
|
229
|
+
if not record.rewrite.strip():
|
|
230
|
+
raise AlgorithmOutputError("InfoCQR cache produced an empty final rewrite")
|
|
231
|
+
trace = ExecutionTrace(
|
|
232
|
+
method_id=self.method_id,
|
|
233
|
+
release_id=self.release_id,
|
|
234
|
+
events=(
|
|
235
|
+
TraceEvent(
|
|
236
|
+
TracePhase.PREPARE,
|
|
237
|
+
"cache_identity",
|
|
238
|
+
{
|
|
239
|
+
"artifact_id": self.cache.artifact_id,
|
|
240
|
+
"artifact_sha256": self.cache.artifact_sha256,
|
|
241
|
+
"dataset_revision": self.cache.dataset_revision,
|
|
242
|
+
"mode": self.mode.value,
|
|
243
|
+
"split": self.split,
|
|
244
|
+
"conversation_hash": observed_conversation_hash,
|
|
245
|
+
},
|
|
246
|
+
),
|
|
247
|
+
TraceEvent(
|
|
248
|
+
TracePhase.PROPOSE,
|
|
249
|
+
"cached_candidate",
|
|
250
|
+
{
|
|
251
|
+
"raw_candidate": record.raw_candidate,
|
|
252
|
+
"postprocess_id": record.postprocess_id,
|
|
253
|
+
"fallback_reason": record.fallback_reason,
|
|
254
|
+
},
|
|
255
|
+
),
|
|
256
|
+
TraceEvent(
|
|
257
|
+
TracePhase.FINALIZE,
|
|
258
|
+
"final_rewrite",
|
|
259
|
+
{
|
|
260
|
+
"rewrite": record.rewrite,
|
|
261
|
+
"first_turn_truth_rewrite": record.first_turn_truth_rewrite,
|
|
262
|
+
},
|
|
263
|
+
),
|
|
264
|
+
),
|
|
265
|
+
counters={"historical_model_calls": 0, "replay_records": 1},
|
|
266
|
+
warnings=(
|
|
267
|
+
("gold_derived_first_turn_rewrite_explicitly_allowed",)
|
|
268
|
+
if record.first_turn_truth_rewrite
|
|
269
|
+
else ()
|
|
270
|
+
),
|
|
271
|
+
)
|
|
272
|
+
return InputPatch(
|
|
273
|
+
method_id=self.method_id,
|
|
274
|
+
release_id=self.release_id,
|
|
275
|
+
source_state_hash=state.state_hash,
|
|
276
|
+
operations=(
|
|
277
|
+
PatchOperation(
|
|
278
|
+
PatchOp.REPLACE_QUERY,
|
|
279
|
+
record.rewrite,
|
|
280
|
+
target="query",
|
|
281
|
+
rationale="checksum-locked InfoCQR cache replay",
|
|
282
|
+
),
|
|
283
|
+
),
|
|
284
|
+
trace=trace,
|
|
285
|
+
implementation_identity=(
|
|
286
|
+
f"artifact_replay:smartyfh/InfoCQR@{INFOCQR_UPSTREAM_COMMIT}:"
|
|
287
|
+
f"{self.cache.artifact_sha256}"
|
|
288
|
+
),
|
|
289
|
+
profile_kind="controlled_adapter",
|
|
290
|
+
metadata={
|
|
291
|
+
"implementation_origin": "compatibility_replay",
|
|
292
|
+
"artifact_id": self.cache.artifact_id,
|
|
293
|
+
"artifact_sha256": self.cache.artifact_sha256,
|
|
294
|
+
"dataset_revision": self.cache.dataset_revision,
|
|
295
|
+
"historical_generation_not_rebilled": True,
|
|
296
|
+
"gold_derived_first_turn": record.first_turn_truth_rewrite,
|
|
297
|
+
"record_metadata": dict(record.metadata),
|
|
298
|
+
},
|
|
299
|
+
)
|
|
300
|
+
|
|
301
|
+
|
|
302
|
+
class InfoCQRZeroShotReplay(_InfoCQRReplay):
|
|
303
|
+
def __init__(
|
|
304
|
+
self,
|
|
305
|
+
cache: InfoCQRCache,
|
|
306
|
+
*,
|
|
307
|
+
split: str,
|
|
308
|
+
allow_first_turn_truth_rewrite: bool = False,
|
|
309
|
+
) -> None:
|
|
310
|
+
super().__init__(
|
|
311
|
+
cache=cache,
|
|
312
|
+
split=split,
|
|
313
|
+
mode=InfoCQRMode.RW_ZSL,
|
|
314
|
+
release_id="rewrite.infocqr.cached_rw_zsl",
|
|
315
|
+
allow_first_turn_truth_rewrite=allow_first_turn_truth_rewrite,
|
|
316
|
+
)
|
|
317
|
+
|
|
318
|
+
|
|
319
|
+
class InfoCQRFewShotReplay(_InfoCQRReplay):
|
|
320
|
+
def __init__(
|
|
321
|
+
self,
|
|
322
|
+
cache: InfoCQRCache,
|
|
323
|
+
*,
|
|
324
|
+
split: str,
|
|
325
|
+
allow_first_turn_truth_rewrite: bool = False,
|
|
326
|
+
) -> None:
|
|
327
|
+
super().__init__(
|
|
328
|
+
cache=cache,
|
|
329
|
+
split=split,
|
|
330
|
+
mode=InfoCQRMode.RW_FSL,
|
|
331
|
+
release_id="rewrite.infocqr.cached_rw_fsl",
|
|
332
|
+
allow_first_turn_truth_rewrite=allow_first_turn_truth_rewrite,
|
|
333
|
+
)
|
|
334
|
+
|
|
335
|
+
|
|
336
|
+
class InfoCQREditReplay(_InfoCQRReplay):
|
|
337
|
+
def __init__(
|
|
338
|
+
self,
|
|
339
|
+
cache: InfoCQRCache,
|
|
340
|
+
*,
|
|
341
|
+
split: str,
|
|
342
|
+
allow_first_turn_truth_rewrite: bool = False,
|
|
343
|
+
) -> None:
|
|
344
|
+
super().__init__(
|
|
345
|
+
cache=cache,
|
|
346
|
+
split=split,
|
|
347
|
+
mode=InfoCQRMode.EDIT,
|
|
348
|
+
release_id="rewrite.infocqr.cached_edit",
|
|
349
|
+
allow_first_turn_truth_rewrite=allow_first_turn_truth_rewrite,
|
|
350
|
+
)
|
|
351
|
+
|
|
352
|
+
|
|
353
|
+
__all__ = [
|
|
354
|
+
"INFOCQR_UPSTREAM_COMMIT",
|
|
355
|
+
"InfoCQRCache",
|
|
356
|
+
"InfoCQREditReplay",
|
|
357
|
+
"InfoCQRFewShotReplay",
|
|
358
|
+
"InfoCQRMode",
|
|
359
|
+
"InfoCQRRecord",
|
|
360
|
+
"InfoCQRZeroShotReplay",
|
|
361
|
+
"conversation_view_hash",
|
|
362
|
+
]
|