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 +1,181 @@
|
|
|
1
|
+
"""Dependency-free JSON-lines worker transport.
|
|
2
|
+
|
|
3
|
+
The transport never invokes a shell. It exchanges one bounded JSON object per
|
|
4
|
+
process invocation so legacy workers cannot silently leak arbitrary logs into
|
|
5
|
+
the structured result channel.
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
from __future__ import annotations
|
|
9
|
+
|
|
10
|
+
import json
|
|
11
|
+
import os
|
|
12
|
+
from collections.abc import Mapping
|
|
13
|
+
from dataclasses import dataclass, field
|
|
14
|
+
from typing import Any, Protocol, runtime_checkable
|
|
15
|
+
|
|
16
|
+
from agentinputlab.contracts.errors import ExternalDependencyError, ValidationError
|
|
17
|
+
from agentinputlab.contracts.serialization import content_hash, freeze_mapping, to_jsonable
|
|
18
|
+
from agentinputlab.observability.models import ResourceLimits, ResourceReceipt
|
|
19
|
+
|
|
20
|
+
from .supervisor import (
|
|
21
|
+
AcceleratorSampler,
|
|
22
|
+
LocalSubprocessSupervisor,
|
|
23
|
+
SupervisedWorkerError,
|
|
24
|
+
WorkerSupervisor,
|
|
25
|
+
)
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
@runtime_checkable
|
|
29
|
+
class WorkerClient(Protocol):
|
|
30
|
+
worker_id: str
|
|
31
|
+
|
|
32
|
+
def call(self, payload: Mapping[str, Any]) -> Mapping[str, Any]:
|
|
33
|
+
"""Return one decoded worker response or raise a typed error."""
|
|
34
|
+
|
|
35
|
+
def fingerprint(self) -> Mapping[str, Any]:
|
|
36
|
+
"""Return image/source/protocol identity."""
|
|
37
|
+
|
|
38
|
+
|
|
39
|
+
@dataclass(slots=True)
|
|
40
|
+
class JsonLineSubprocessWorker:
|
|
41
|
+
"""Run a pinned worker command without shell interpolation."""
|
|
42
|
+
|
|
43
|
+
command: tuple[str, ...]
|
|
44
|
+
worker_id: str
|
|
45
|
+
identity: Mapping[str, Any]
|
|
46
|
+
timeout_seconds: float = 300.0
|
|
47
|
+
max_input_bytes: int = 8 * 1024 * 1024
|
|
48
|
+
max_output_bytes: int = 8 * 1024 * 1024
|
|
49
|
+
max_stderr_bytes: int = 1024 * 1024
|
|
50
|
+
environment: Mapping[str, str] | None = None
|
|
51
|
+
pass_environment: tuple[str, ...] = ()
|
|
52
|
+
working_directory: str | None = None
|
|
53
|
+
resource_limits: ResourceLimits | None = None
|
|
54
|
+
supervisor: WorkerSupervisor = field(default_factory=LocalSubprocessSupervisor)
|
|
55
|
+
accelerator_sampler: AcceleratorSampler | None = None
|
|
56
|
+
_calls: int = field(default=0, init=False, repr=False)
|
|
57
|
+
_last_resource_receipt: ResourceReceipt | None = field(default=None, init=False, repr=False)
|
|
58
|
+
|
|
59
|
+
def __post_init__(self) -> None:
|
|
60
|
+
self.command = tuple(self.command)
|
|
61
|
+
if not self.command or not all(isinstance(item, str) and item for item in self.command):
|
|
62
|
+
raise ValidationError("worker command must be a non-empty argv tuple")
|
|
63
|
+
if self.timeout_seconds <= 0:
|
|
64
|
+
raise ValidationError("worker timeout_seconds must be positive")
|
|
65
|
+
if min(self.max_input_bytes, self.max_output_bytes, self.max_stderr_bytes) < 1:
|
|
66
|
+
raise ValidationError("worker byte limits must be positive")
|
|
67
|
+
if not self.identity:
|
|
68
|
+
raise ValidationError("worker requires explicit pinned identity")
|
|
69
|
+
self.identity = freeze_mapping(self.identity)
|
|
70
|
+
if self.environment is not None:
|
|
71
|
+
self.environment = freeze_mapping(self.environment)
|
|
72
|
+
self.pass_environment = tuple(sorted(set(self.pass_environment)))
|
|
73
|
+
for name in self.pass_environment:
|
|
74
|
+
if (
|
|
75
|
+
not isinstance(name, str)
|
|
76
|
+
or not name
|
|
77
|
+
or not (name[0].isalpha() or name[0] == "_")
|
|
78
|
+
or not all(character.isalnum() or character == "_" for character in name)
|
|
79
|
+
):
|
|
80
|
+
raise ValidationError("passed worker environment names must be identifiers")
|
|
81
|
+
if set(self.pass_environment) & set((self.environment or {}).keys()):
|
|
82
|
+
raise ValidationError("passed worker environments cannot duplicate public values")
|
|
83
|
+
if self.resource_limits is None:
|
|
84
|
+
self.resource_limits = ResourceLimits(
|
|
85
|
+
wall_timeout_seconds=self.timeout_seconds,
|
|
86
|
+
)
|
|
87
|
+
elif not isinstance(self.resource_limits, ResourceLimits):
|
|
88
|
+
raise TypeError("resource_limits must be ResourceLimits")
|
|
89
|
+
else:
|
|
90
|
+
self.timeout_seconds = self.resource_limits.wall_timeout_seconds
|
|
91
|
+
if not isinstance(self.supervisor, WorkerSupervisor):
|
|
92
|
+
raise TypeError("supervisor must implement WorkerSupervisor")
|
|
93
|
+
if self.accelerator_sampler is not None and not isinstance(
|
|
94
|
+
self.accelerator_sampler,
|
|
95
|
+
AcceleratorSampler,
|
|
96
|
+
):
|
|
97
|
+
raise TypeError("accelerator_sampler must implement AcceleratorSampler")
|
|
98
|
+
|
|
99
|
+
def call(self, payload: Mapping[str, Any]) -> Mapping[str, Any]:
|
|
100
|
+
self._last_resource_receipt = None
|
|
101
|
+
missing = [name for name in self.pass_environment if not os.environ.get(name)]
|
|
102
|
+
if missing:
|
|
103
|
+
raise ExternalDependencyError(
|
|
104
|
+
"external worker is missing passed environments: " + ", ".join(missing)
|
|
105
|
+
)
|
|
106
|
+
environment = dict(self.environment or {})
|
|
107
|
+
environment.update({name: os.environ[name] for name in self.pass_environment})
|
|
108
|
+
request = json.dumps(
|
|
109
|
+
to_jsonable(dict(payload)),
|
|
110
|
+
ensure_ascii=False,
|
|
111
|
+
sort_keys=True,
|
|
112
|
+
separators=(",", ":"),
|
|
113
|
+
allow_nan=False,
|
|
114
|
+
)
|
|
115
|
+
try:
|
|
116
|
+
completed = self.supervisor.execute(
|
|
117
|
+
self.command,
|
|
118
|
+
input_text=f"{request}\n",
|
|
119
|
+
limits=self.resource_limits,
|
|
120
|
+
max_input_bytes=self.max_input_bytes,
|
|
121
|
+
max_stdout_bytes=self.max_output_bytes,
|
|
122
|
+
max_stderr_bytes=self.max_stderr_bytes,
|
|
123
|
+
cwd=self.working_directory,
|
|
124
|
+
environment=environment or None,
|
|
125
|
+
process_name=f"external worker {self.worker_id!r}",
|
|
126
|
+
image_identity=self.identity,
|
|
127
|
+
accelerator_sampler=self.accelerator_sampler,
|
|
128
|
+
)
|
|
129
|
+
except SupervisedWorkerError as exc:
|
|
130
|
+
self._last_resource_receipt = exc.resource_receipt
|
|
131
|
+
raise
|
|
132
|
+
self._last_resource_receipt = completed.resource_receipt
|
|
133
|
+
self._calls += 1
|
|
134
|
+
if completed.returncode != 0:
|
|
135
|
+
raise SupervisedWorkerError(
|
|
136
|
+
f"external worker {self.worker_id!r} exited with code {completed.returncode}",
|
|
137
|
+
resource_receipt=completed.resource_receipt,
|
|
138
|
+
)
|
|
139
|
+
lines = [line for line in completed.stdout.splitlines() if line.strip()]
|
|
140
|
+
if len(lines) != 1:
|
|
141
|
+
raise SupervisedWorkerError(
|
|
142
|
+
f"external worker {self.worker_id!r} must emit exactly one JSON line",
|
|
143
|
+
resource_receipt=completed.resource_receipt,
|
|
144
|
+
)
|
|
145
|
+
try:
|
|
146
|
+
response = json.loads(lines[0])
|
|
147
|
+
except json.JSONDecodeError as exc:
|
|
148
|
+
raise SupervisedWorkerError(
|
|
149
|
+
f"external worker {self.worker_id!r} emitted invalid JSON",
|
|
150
|
+
resource_receipt=completed.resource_receipt,
|
|
151
|
+
) from exc
|
|
152
|
+
if not isinstance(response, dict):
|
|
153
|
+
raise SupervisedWorkerError(
|
|
154
|
+
"external worker response must be a JSON object",
|
|
155
|
+
resource_receipt=completed.resource_receipt,
|
|
156
|
+
)
|
|
157
|
+
return response
|
|
158
|
+
|
|
159
|
+
@property
|
|
160
|
+
def last_resource_receipt(self) -> ResourceReceipt | None:
|
|
161
|
+
"""Most recent call receipt, including failed supervised calls."""
|
|
162
|
+
|
|
163
|
+
return self._last_resource_receipt
|
|
164
|
+
|
|
165
|
+
def fingerprint(self) -> Mapping[str, Any]:
|
|
166
|
+
return {
|
|
167
|
+
**dict(self.identity),
|
|
168
|
+
"worker_id": self.worker_id,
|
|
169
|
+
"protocol": "agentinputlab.jsonl.v1",
|
|
170
|
+
"command_hash": content_hash(self.command),
|
|
171
|
+
"working_directory_hash": content_hash(self.working_directory),
|
|
172
|
+
"environment_keys": sorted((self.environment or {}).keys()),
|
|
173
|
+
"environment_hash": content_hash(self.environment or {}),
|
|
174
|
+
"passed_environment_names": self.pass_environment,
|
|
175
|
+
"timeout_seconds": self.timeout_seconds,
|
|
176
|
+
"max_input_bytes": self.max_input_bytes,
|
|
177
|
+
"max_output_bytes": self.max_output_bytes,
|
|
178
|
+
"max_stderr_bytes": self.max_stderr_bytes,
|
|
179
|
+
"resource_limit_hash": self.resource_limits.limit_hash,
|
|
180
|
+
"supervisor": dict(self.supervisor.fingerprint()),
|
|
181
|
+
}
|
|
@@ -0,0 +1,249 @@
|
|
|
1
|
+
"""Supervised native DecomP with authenticated, budgeted host completions."""
|
|
2
|
+
from __future__ import annotations
|
|
3
|
+
|
|
4
|
+
import hashlib
|
|
5
|
+
import json
|
|
6
|
+
import math
|
|
7
|
+
import os
|
|
8
|
+
import secrets
|
|
9
|
+
import tempfile
|
|
10
|
+
import threading
|
|
11
|
+
from collections.abc import Mapping
|
|
12
|
+
from dataclasses import dataclass, field
|
|
13
|
+
from importlib import resources
|
|
14
|
+
from pathlib import Path
|
|
15
|
+
from typing import ClassVar
|
|
16
|
+
|
|
17
|
+
from agentinputlab.algorithms.decompose.decomp.worker_bridge import DecompWorkerRequest
|
|
18
|
+
from agentinputlab.backends.llm.decomp_completion import validate_native_parameters
|
|
19
|
+
from agentinputlab.contracts import ValidationError
|
|
20
|
+
from agentinputlab.contracts.serialization import content_hash, to_jsonable
|
|
21
|
+
|
|
22
|
+
from .base import JsonLineSubprocessWorker
|
|
23
|
+
from .llmcompiler_live import _server
|
|
24
|
+
from .supervisor import SupervisedWorkerError
|
|
25
|
+
|
|
26
|
+
_KEY_ENV = "AGENTINPUTLAB_DECOMP_CALLBACK_KEY"
|
|
27
|
+
_MODE = "controlled_native_completion_callbacks.v1"
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
class _Callbacks:
|
|
31
|
+
def __init__(self, owner, request, identity):
|
|
32
|
+
self.owner, self.request, self.identity = owner, request, identity
|
|
33
|
+
self.budget = request.execution_budget
|
|
34
|
+
self.input_bound = getattr(owner, "per_call_input_token_bound", None)
|
|
35
|
+
if (self.budget is None or self.budget.max_cost_usd is not None
|
|
36
|
+
or (self.budget.max_context_tokens is not None and self.input_bound is None)):
|
|
37
|
+
raise ValidationError("live DecomP requires output/call budgets; finite input/cost ceilings need an input bound or pricing estimator")
|
|
38
|
+
if self.input_bound is not None and (type(self.input_bound) is not int or self.input_bound < 1):
|
|
39
|
+
raise ValidationError("DecomP input token bound must be a positive integer")
|
|
40
|
+
if content_hash(request.backend_binding) != content_hash(owner.backend.fingerprint()):
|
|
41
|
+
raise ValidationError("DecomP live backend differs from the context backend")
|
|
42
|
+
self.request_id = content_hash(to_jsonable(request))
|
|
43
|
+
self.events = []
|
|
44
|
+
self.seen = set()
|
|
45
|
+
self.lock = threading.RLock()
|
|
46
|
+
self.attested = self.closed = self.fatal = False
|
|
47
|
+
self.calls = self.input_tokens = self.output_tokens = self.reserved = self.inflight = 0
|
|
48
|
+
self.input_reserved = 0
|
|
49
|
+
self.unknown_usage_calls = 0
|
|
50
|
+
|
|
51
|
+
def usage(self):
|
|
52
|
+
return {"model_calls": self.calls, "retrieval_calls": 0, "input_tokens": self.input_tokens,
|
|
53
|
+
"output_tokens": self.output_tokens, "unpriced_model_calls": self.calls}
|
|
54
|
+
|
|
55
|
+
def dispatch(self, envelope):
|
|
56
|
+
if (not isinstance(envelope, dict) or set(envelope) != {"request_id", "call_id", "kind", "payload"}
|
|
57
|
+
or envelope["request_id"] != self.request_id or not isinstance(envelope["call_id"], str)
|
|
58
|
+
or not envelope["call_id"]):
|
|
59
|
+
raise ValueError("invalid DecomP callback envelope")
|
|
60
|
+
with self.lock:
|
|
61
|
+
if self.closed or self.fatal or envelope["call_id"] in self.seen:
|
|
62
|
+
raise ValueError("closed or duplicate DecomP callback")
|
|
63
|
+
self.seen.add(envelope["call_id"])
|
|
64
|
+
if envelope["kind"] == "attest":
|
|
65
|
+
if self.attested or envelope["payload"] != self.identity["worker_attestation"]:
|
|
66
|
+
raise ValueError("DecomP worker source or dependency attestation differs")
|
|
67
|
+
self.attested = True
|
|
68
|
+
self.events.append({"kind": "attestation", "matched": True})
|
|
69
|
+
return {"attested": True}
|
|
70
|
+
if envelope["kind"] != "model" or not self.attested:
|
|
71
|
+
raise ValueError("DecomP worker must attest before native model calls")
|
|
72
|
+
parameters = validate_native_parameters(envelope["payload"])
|
|
73
|
+
cap = parameters["max_tokens"]
|
|
74
|
+
input_reservation = self.input_bound or 0
|
|
75
|
+
with self.lock:
|
|
76
|
+
if (self.fatal or self.closed or self.calls >= self.budget.max_model_calls
|
|
77
|
+
or cap > self.owner.per_call_output_cap
|
|
78
|
+
or self.output_tokens + self.reserved + cap > self.budget.max_output_tokens
|
|
79
|
+
or (self.budget.max_context_tokens is not None and
|
|
80
|
+
self.input_tokens + self.input_reserved + input_reservation > self.budget.max_context_tokens)):
|
|
81
|
+
raise ValueError("DecomP native budget exhausted before dispatch")
|
|
82
|
+
self.calls += 1
|
|
83
|
+
self.inflight += 1
|
|
84
|
+
self.reserved += cap
|
|
85
|
+
self.input_reserved += input_reservation
|
|
86
|
+
event = {"kind": "model", "native_parameters": parameters, "seed": self.request.seed}
|
|
87
|
+
self.events.append(event)
|
|
88
|
+
try:
|
|
89
|
+
response = to_jsonable(self.owner.backend.complete_native(parameters, seed=self.request.seed))
|
|
90
|
+
event["response"] = response
|
|
91
|
+
usage = response["usage"]
|
|
92
|
+
counts = (usage["prompt_tokens"], usage["completion_tokens"])
|
|
93
|
+
if any(type(n) is not int or n < 0 for n in counts):
|
|
94
|
+
raise ValueError("invalid native usage")
|
|
95
|
+
with self.lock:
|
|
96
|
+
self.input_tokens += counts[0]
|
|
97
|
+
self.output_tokens += counts[1]
|
|
98
|
+
self.reserved -= cap
|
|
99
|
+
self.input_reserved -= input_reservation
|
|
100
|
+
event["usage_verified"] = True
|
|
101
|
+
if self.input_bound is not None and counts[0] > self.input_bound:
|
|
102
|
+
raise ValueError("provider exceeded DecomP native input token bound")
|
|
103
|
+
if counts[1] > cap:
|
|
104
|
+
raise ValueError("provider exceeded DecomP native output cap")
|
|
105
|
+
if response.get("model") != self.owner.backend.model_id:
|
|
106
|
+
raise ValueError("native completion returned a different model")
|
|
107
|
+
choices = response["choices"]
|
|
108
|
+
if not isinstance(choices, list) or len(choices) != 1 or not isinstance(choices[0]["text"], str):
|
|
109
|
+
raise ValueError("native completion requires exactly one text choice")
|
|
110
|
+
logs = choices[0].get("logprobs")
|
|
111
|
+
if parameters["logprobs"] is not None:
|
|
112
|
+
if not isinstance(logs, Mapping):
|
|
113
|
+
raise ValueError("native requested log probabilities are missing")
|
|
114
|
+
probs, tokens = logs["token_logprobs"], logs["tokens"]
|
|
115
|
+
if (not isinstance(probs, list) or not isinstance(tokens, list) or len(probs) != len(tokens)
|
|
116
|
+
or any(type(p) not in (float, int) or not math.isfinite(p) or p > 0 for p in probs)
|
|
117
|
+
or not all(isinstance(t, str) for t in tokens)):
|
|
118
|
+
raise ValueError("native token probabilities are invalid")
|
|
119
|
+
return response
|
|
120
|
+
except Exception as exc:
|
|
121
|
+
with self.lock:
|
|
122
|
+
self.fatal = True
|
|
123
|
+
self.unknown_usage_calls += int(not event.get("usage_verified", False))
|
|
124
|
+
event["error_type"] = type(exc).__name__
|
|
125
|
+
raise RuntimeError("bound DecomP completion failed; no retry") from None
|
|
126
|
+
finally:
|
|
127
|
+
with self.lock:
|
|
128
|
+
self.inflight -= 1
|
|
129
|
+
|
|
130
|
+
|
|
131
|
+
@dataclass
|
|
132
|
+
class DecompLiveRuntime:
|
|
133
|
+
"""Bind the original generator seam without granting child provider credentials.
|
|
134
|
+
|
|
135
|
+
Use a dedicated legacy worker interpreter and explicit NLTK data directory.
|
|
136
|
+
Backend I/O must have its own finite timeout. Every run gets an isolated HOME,
|
|
137
|
+
source-checked participants, uncached completions and request-local accounting.
|
|
138
|
+
"""
|
|
139
|
+
|
|
140
|
+
python_executable: str
|
|
141
|
+
checkout: Path
|
|
142
|
+
nltk_data: Path
|
|
143
|
+
backend: object
|
|
144
|
+
timeout_seconds: float = 300.0
|
|
145
|
+
per_call_output_cap: int = 512
|
|
146
|
+
# A caller-attested upper bound, e.g. the bound provider's context window.
|
|
147
|
+
# Reserve it before each call, then replace it with actual reported input.
|
|
148
|
+
# It never truncates prompts or changes the native generation parameters.
|
|
149
|
+
per_call_input_token_bound: int | None = None
|
|
150
|
+
callback_events: list = field(default_factory=list, init=False, repr=False)
|
|
151
|
+
last_response: dict | None = field(default=None, init=False, repr=False)
|
|
152
|
+
last_resource_receipt: object = field(default=None, init=False, repr=False)
|
|
153
|
+
request_type: ClassVar[type] = DecompWorkerRequest
|
|
154
|
+
worker_id: ClassVar[str] = "agentinputlab.decomp.controlled-live.v1"
|
|
155
|
+
worker_flags: ClassVar[tuple[str, ...]] = ()
|
|
156
|
+
|
|
157
|
+
def __post_init__(self):
|
|
158
|
+
if (not isinstance(self.python_executable, str) or not self.python_executable
|
|
159
|
+
or type(self.per_call_output_cap) is not int or self.per_call_output_cap < 1):
|
|
160
|
+
raise ValidationError("DecomP worker interpreter and positive per-call cap required")
|
|
161
|
+
if (type(self.timeout_seconds) not in (int, float) or not math.isfinite(self.timeout_seconds)
|
|
162
|
+
or self.timeout_seconds <= 0):
|
|
163
|
+
raise ValidationError("DecomP finite positive worker timeout required")
|
|
164
|
+
if self.per_call_input_token_bound is not None and (
|
|
165
|
+
type(self.per_call_input_token_bound) is not int or self.per_call_input_token_bound < 1):
|
|
166
|
+
raise ValidationError("DecomP input token bound must be a positive integer or None")
|
|
167
|
+
if not callable(getattr(self.backend, "complete_native", None)) or not callable(getattr(self.backend, "fingerprint", None)):
|
|
168
|
+
raise ValidationError("DecomP requires a source-parameter-preserving completion backend")
|
|
169
|
+
self.checkout, self.nltk_data = Path(self.checkout), Path(self.nltk_data)
|
|
170
|
+
|
|
171
|
+
def fingerprint(self):
|
|
172
|
+
package = resources.files("workers").joinpath("decomp")
|
|
173
|
+
policy = package.joinpath("compatibility_manifest.json").read_bytes()
|
|
174
|
+
def sha(raw):
|
|
175
|
+
return hashlib.sha256(raw).hexdigest()
|
|
176
|
+
stopwords = self.nltk_data / "corpora" / "stopwords" / "english"
|
|
177
|
+
attestation = {"schema": "agentinputlab.decomp.live-attestation.v1",
|
|
178
|
+
"driver_sha256": sha(package.joinpath("driver.py").read_bytes()),
|
|
179
|
+
"live_sha256": sha(package.joinpath("live.py").read_bytes()),
|
|
180
|
+
"compatibility_manifest_sha256": sha(policy),
|
|
181
|
+
"required_versions": json.loads(policy)["live_required_versions"],
|
|
182
|
+
"english_stopwords_sha256": sha(stopwords.read_bytes()),
|
|
183
|
+
"source_files_verified": True, "import_origins_verified": True}
|
|
184
|
+
value = {"worker_id": self.worker_id, "profile_kind": "controlled_adapter",
|
|
185
|
+
"upstream_revision": "8b157f6fe9ff5676ae51bd054f1df7b015adf278",
|
|
186
|
+
"backend_binding_mode": _MODE, "worker_attestation": attestation,
|
|
187
|
+
"backend": dict(self.backend.fingerprint()), "per_call_output_cap": self.per_call_output_cap,
|
|
188
|
+
"timeout_seconds": self.timeout_seconds, "automatic_retry": False}
|
|
189
|
+
if self.per_call_input_token_bound is not None:
|
|
190
|
+
value["per_call_input_token_bound"] = self.per_call_input_token_bound
|
|
191
|
+
value["input_budget_accounting"] = "pre_call_upper_bound_then_actual_provider_usage"
|
|
192
|
+
return value
|
|
193
|
+
|
|
194
|
+
def run(self, request: DecompWorkerRequest):
|
|
195
|
+
if not isinstance(request, self.request_type):
|
|
196
|
+
raise TypeError("DecompLiveRuntime requires DecompWorkerRequest")
|
|
197
|
+
identity = self.fingerprint()
|
|
198
|
+
callbacks = _Callbacks(self, request, identity)
|
|
199
|
+
self.callback_events, self.last_response, self.last_resource_receipt = callbacks.events, None, None
|
|
200
|
+
key = secrets.token_urlsafe(32)
|
|
201
|
+
server = _server(callbacks, key, 4 * 1024 * 1024)
|
|
202
|
+
thread = threading.Thread(target=server.serve_forever, daemon=True)
|
|
203
|
+
thread.start()
|
|
204
|
+
raw = to_jsonable(request)
|
|
205
|
+
payload = {"request_id": callbacks.request_id, "request": raw}
|
|
206
|
+
try:
|
|
207
|
+
with tempfile.TemporaryDirectory(prefix="agentinputlab-decomp-worker-") as home:
|
|
208
|
+
worker = JsonLineSubprocessWorker(
|
|
209
|
+
command=(self.python_executable, "-m", "workers.decomp.live", "--checkout", str(self.checkout),
|
|
210
|
+
"--callback-url", f"http://127.0.0.1:{server.server_port}/callback", "--callback-key-env", _KEY_ENV,
|
|
211
|
+
*self.worker_flags),
|
|
212
|
+
worker_id=identity["worker_id"], identity=identity, timeout_seconds=self.timeout_seconds,
|
|
213
|
+
environment={"PATH": os.environ.get("PATH", ""), "HOME": home, "NLTK_DATA": str(self.nltk_data),
|
|
214
|
+
"CUDA_VISIBLE_DEVICES": "", "PYTHONDONTWRITEBYTECODE": "1", _KEY_ENV: key})
|
|
215
|
+
try:
|
|
216
|
+
response = dict(worker.call(payload))
|
|
217
|
+
except SupervisedWorkerError:
|
|
218
|
+
response = {"status": "error", "failure_code": "LIVE_WORKER_EXECUTION_FAILED"}
|
|
219
|
+
finally:
|
|
220
|
+
self.last_resource_receipt = worker.last_resource_receipt
|
|
221
|
+
with callbacks.lock:
|
|
222
|
+
callbacks.closed = True
|
|
223
|
+
incomplete = callbacks.inflight > 0
|
|
224
|
+
usage = callbacks.usage()
|
|
225
|
+
if incomplete:
|
|
226
|
+
response = {"status": "error", "failure_code": "CALLBACK_USAGE_INCOMPLETE"}
|
|
227
|
+
if response.get("status") == "ok" and response["payload"]["usage"]["model_calls"] != usage["model_calls"]:
|
|
228
|
+
response = {"status": "error", "failure_code": "CALLBACK_USAGE_MISMATCH"}
|
|
229
|
+
response.update(usage=usage, consumed=bool(usage["model_calls"]), callback_diagnostics={"unknown_usage_calls": callbacks.unknown_usage_calls,
|
|
230
|
+
"usage_incomplete": incomplete, "automatic_retry": False})
|
|
231
|
+
self.last_response = response
|
|
232
|
+
if response.get("status") != "ok":
|
|
233
|
+
# The request's external-status hook raises only after the
|
|
234
|
+
# monitoring boundary has charged this complete actual usage.
|
|
235
|
+
return response
|
|
236
|
+
value = response["payload"]
|
|
237
|
+
value["usage"] = usage
|
|
238
|
+
value["native_artifacts"].update(backend_binding_mode=identity["backend_binding_mode"],
|
|
239
|
+
token_accounting="actual_native_provider_usage", provider_cache="disabled_controlled_profile")
|
|
240
|
+
return self._parse_payload(value)
|
|
241
|
+
finally:
|
|
242
|
+
callbacks.closed = True
|
|
243
|
+
server.shutdown()
|
|
244
|
+
server.server_close()
|
|
245
|
+
thread.join(timeout=2)
|
|
246
|
+
|
|
247
|
+
def _parse_payload(self, value):
|
|
248
|
+
from workers.decomp.runtime import _parse_response
|
|
249
|
+
return _parse_response(value)
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
"""Dedicated planner-only native subprocess; cannot satisfy the full bridge."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
from agentinputlab.algorithms.decompose.decomp.planner_component import (
|
|
6
|
+
DECOMP_PLANNER_MODE,
|
|
7
|
+
DecompPlannerRequest,
|
|
8
|
+
)
|
|
9
|
+
|
|
10
|
+
from .decomp_live import DecompLiveRuntime
|
|
11
|
+
|
|
12
|
+
|
|
13
|
+
class DecompPlannerLiveRuntime(DecompLiveRuntime):
|
|
14
|
+
request_type = DecompPlannerRequest
|
|
15
|
+
worker_id = "agentinputlab.decomp.native-planner.v1"
|
|
16
|
+
worker_flags = ("--planner-component",)
|
|
17
|
+
|
|
18
|
+
def fingerprint(self):
|
|
19
|
+
return {
|
|
20
|
+
**super().fingerprint(),
|
|
21
|
+
"backend_binding_mode": DECOMP_PLANNER_MODE,
|
|
22
|
+
"component_only": True,
|
|
23
|
+
"max_native_transitions": 1,
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
def _parse_payload(self, value):
|
|
27
|
+
return value
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
__all__ = ["DecompPlannerLiveRuntime"]
|