diff --git a/docs/portfolio/inferedge_1page_architecture.ko.md b/docs/portfolio/inferedge_1page_architecture.ko.md index 220aeca..c337be3 100644 --- a/docs/portfolio/inferedge_1page_architecture.ko.md +++ b/docs/portfolio/inferedge_1page_architecture.ko.md @@ -17,7 +17,7 @@ ONNX model -> EdgeEnv -> Lab -> optional AIGuard --> optional Orchestrator operation context +-> optional Orchestrator operation context / operation_risk_rollup ``` ## 책임 경계 @@ -29,11 +29,11 @@ ONNX model | EdgeEnv | registry/comparability/regression evidence | Lab decision, public leaderboard | | Lab | report/API/job/deployment decision | production SaaS infrastructure | | AIGuard | deterministic diagnosis evidence | final decision owner | -| Orchestrator | queue/deadline/fallback operation context | Kubernetes/cloud orchestration | +| Orchestrator | queue/deadline/fallback operation context와 `operation_risk_rollup` | Kubernetes/cloud orchestration | ## Runtime Intelligence 흐름 -Runtime Intelligence는 새 repo나 monitoring SaaS가 아니다. Orchestrator operation feed, EdgeEnv telemetry/regression context, AIGuard deterministic warning evidence를 Lab report에 보존해 deployment risk를 더 쉽게 검토하게 만드는 local-first evidence extension이다. +Runtime Intelligence는 새 repo나 monitoring SaaS가 아니다. Orchestrator operation feed와 `operation_risk_rollup`, EdgeEnv telemetry/regression context, AIGuard deterministic warning evidence를 Lab report에 보존해 deployment risk를 더 쉽게 검토하게 만드는 local-first evidence extension이다. ## Reviewer focus diff --git a/docs/portfolio/inferedge_1page_architecture.md b/docs/portfolio/inferedge_1page_architecture.md index 8cd9bcf..6e4912a 100644 --- a/docs/portfolio/inferedge_1page_architecture.md +++ b/docs/portfolio/inferedge_1page_architecture.md @@ -8,7 +8,7 @@ InferEdge is an end-to-end Edge AI inference validation pipeline that builds dep Supporting sidecar: InferEdgeEnv is a local-first run evidence registry and comparability checker for Edge AI inference benchmark results. -Runtime Intelligence extension: Orchestrator can provide supplemental operation context, EdgeEnv preserves telemetry history and comparability-first regression evidence, AIGuard can explain deterministic runtime anomaly evidence, and Lab remains the deployment risk report and final decision owner. +Runtime Intelligence extension: Orchestrator can provide supplemental operation context including `operation_risk_rollup`, EdgeEnv preserves telemetry history and comparability-first regression evidence, AIGuard can explain deterministic runtime anomaly evidence, and Lab remains the deployment risk report and final decision owner. PDF-ready portfolio draft: [InferEdge Portfolio Submission](inferedge_portfolio_submission.md). Local PDF export uses pandoc + xelatex through `bash scripts/export_portfolio_pdf.sh`. @@ -34,7 +34,7 @@ Supporting sidecar: InferEdgeEnv -> local-first run evidence registry / comparability checker Runtime Intelligence smoke chain: -InferEdgeOrchestrator operation feed +InferEdgeOrchestrator operation feed / operation_risk_rollup -> InferEdgeEnv telemetry history / regression context -> optional InferEdgeAIGuard deterministic runtime anomaly evidence -> InferEdgeLab Runtime Intelligence Risk Summary / deployment risk report @@ -47,7 +47,7 @@ InferEdgeOrchestrator operation feed - **InferEdgeLab:** analysis/API/job/deployment decision owner. Compares Runtime results, generates reports, exposes API/job workflow contracts, preserves optional guard evidence, and owns the final `deployment_decision`. - **InferEdgeAIGuard:** optional rule + evidence diagnosis layer. Detects provenance/artifact/config mismatches and returns deterministic `guard_analysis` evidence for Lab to consume. - **InferEdgeEnv:** run evidence registry / comparability checker. Records benchmark artifacts, SQLite registry entries, evidence bundles, and comparability judgement without owning Lab deployment decisions. -- **InferEdgeOrchestrator:** runtime operation context provider. Supplies queue, deadline, fallback, thermal, and resource context as supplemental evidence without becoming the regression/comparability owner or final decision owner. +- **InferEdgeOrchestrator:** runtime operation context provider. Supplies queue, deadline, fallback, thermal, resource context, and compact `operation_risk_rollup` evidence without becoming the regression/comparability owner or final decision owner. Portfolio boundary: InferEdgeLab is the validation / decision layer. InferEdgeEnv is the run evidence registry / comparability layer. AIGuard and Orchestrator remain evidence providers, while Lab owns the final deployment decision. @@ -64,7 +64,7 @@ Portfolio boundary: InferEdgeLab is the validation / decision layer. InferEdgeEn - Runtime `worker_request` validation and `worker_response` dry-run export - Forge worker/runtime summary - AIGuard evidence diagnosis cases for provenance mismatch, bbox collapse, score saturation, temporal instability, and normal/pass paths -- Runtime Intelligence smoke evidence chain: Orchestrator `edgeenv_runtime_telemetry_feed` -> EdgeEnv telemetry history and producer-owned `history.telemetry_coverage` -> AIGuard deterministic runtime anomaly evidence -> Lab Runtime Intelligence Risk Summary +- Runtime Intelligence smoke evidence chain: Orchestrator `edgeenv_runtime_telemetry_feed` and `operation_risk_rollup` -> EdgeEnv telemetry history and producer-owned `history.telemetry_coverage` -> AIGuard deterministic runtime anomaly evidence -> Lab Runtime Intelligence Risk Summary - Bundle/report gates that preserve EdgeEnv coverage ownership, Orchestrator `edgeenv_mapping_hint`, AIGuard raw-context producer lineage handoff, and Lab-owned deployment risk wording - Lab decision/report guard evidence smoke - all repo README pipeline summaries synced diff --git a/docs/portfolio/inferedge_portfolio_submission.ko.md b/docs/portfolio/inferedge_portfolio_submission.ko.md index ff62507..fc3a6c0 100644 --- a/docs/portfolio/inferedge_portfolio_submission.ko.md +++ b/docs/portfolio/inferedge_portfolio_submission.ko.md @@ -16,7 +16,7 @@ Forge build provenance -> EdgeEnv registry / comparability / regression context -> Lab report / deployment decision -> optional AIGuard deterministic evidence --> optional Orchestrator operation context +-> optional Orchestrator operation context / operation_risk_rollup ``` ## 한눈에 보는 역할 분리 @@ -28,7 +28,7 @@ Forge build provenance | EdgeEnv | registry/comparability | 비교 가능한 조건인지 먼저 판정했는가 | | Lab | report/deployment decision | 최종 deploy/review/blocked 판단을 Lab이 소유하는가 | | AIGuard | deterministic diagnosis | runtime/output warning을 근거 기반으로 설명하는가 | -| Orchestrator | operation context | queue/deadline/fallback context가 supplemental evidence로 보존되는가 | +| Orchestrator | operation context | queue/deadline/fallback context와 `operation_risk_rollup`이 supplemental evidence로 보존되는가 | ## 강한 evidence @@ -36,6 +36,7 @@ Forge build provenance - Jetson TensorRT FP16 25W fixture: `10.066401 ms` mean / `15.548438 ms` p99 / `99.340373 FPS`. - 같은 demo pair 기준 TensorRT Jetson FP16은 ONNX Runtime CPU 대비 약 `4.51x` 빠르다. - Jetson EdgeEnv preservation smoke는 `device_local_starter`, live `tegrastats`, `runtime_operation_summary`, EdgeEnv run evidence, AIGuard warning, Lab deployment risk report까지 이어지는 local-first artifact chain을 보여준다. +- Runtime Intelligence chain은 Orchestrator `operation_risk_rollup`을 EdgeEnv handoff, AIGuard deterministic evidence, Lab Risk Summary로 연결하되 Lab final decision ownership은 유지한다. ## 경계 diff --git a/docs/portfolio/inferedge_portfolio_submission.md b/docs/portfolio/inferedge_portfolio_submission.md index 15af1ae..d29c5f8 100644 --- a/docs/portfolio/inferedge_portfolio_submission.md +++ b/docs/portfolio/inferedge_portfolio_submission.md @@ -10,7 +10,7 @@ InferEdge is not a benchmarking tool, but an end-to-end validation pipeline that InferEdgeEnv complements this pipeline as a local-first run evidence registry and comparability checker. Lab remains the validation / decision layer; Env records whether benchmark evidence can be trusted and compared. -Runtime Intelligence는 이 구조를 새 제품이나 monitoring SaaS로 키우지 않고, Orchestrator operation context -> EdgeEnv telemetry history/regression -> AIGuard deterministic evidence -> Lab deployment risk report 흐름으로 smoke evidence chain을 고정한다. +Runtime Intelligence는 이 구조를 새 제품이나 monitoring SaaS로 키우지 않고, Orchestrator operation context / `operation_risk_rollup` -> EdgeEnv telemetry history/regression -> AIGuard deterministic evidence -> Lab deployment risk report 흐름으로 smoke evidence chain을 고정한다. 이 프로젝트는 단순 latency benchmark가 아니라 artifact provenance, runtime result compatibility, deployment decision까지 연결한다. 목표는 "빠른 숫자"를 보여주는 것이 아니라, 어떤 모델과 산출물이 어떤 환경에서 실행되었고 그 결과를 배포해도 되는지 검토 가능한 evidence로 남기는 것이다. @@ -19,7 +19,7 @@ Runtime Intelligence는 이 구조를 새 제품이나 monitoring SaaS로 키우 - InferEdgeLab은 Runtime benchmark 결과를 분석해 comparison report, API response, async job result, deployment decision을 생성한다. - InferEdge 전체 흐름은 Forge build provenance -> Runtime real execution -> Lab compare/report/API/job/deployment_decision -> optional AIGuard diagnosis evidence로 구성된다. - Lab은 InferEdgeForge provenance metadata, InferEdge-Runtime C++ execution output, optional InferEdgeAIGuard diagnostic evidence를 하나의 검증 bundle로 연결한다. -- Runtime Intelligence smoke chain은 Orchestrator operation feed, EdgeEnv producer-owned telemetry coverage, AIGuard deterministic anomaly evidence, Lab-owned risk summary를 하나의 local-first artifact 흐름으로 연결한다. +- Runtime Intelligence smoke chain은 Orchestrator operation feed와 `operation_risk_rollup`, EdgeEnv producer-owned telemetry coverage, AIGuard deterministic anomaly evidence, Lab-owned risk summary를 하나의 local-first artifact 흐름으로 연결한다. - `yolov8n.onnx` manual smoke에서 Lab -> C++ Runtime CLI -> ONNX Runtime CPU execution -> Lab job result ingestion 경로가 dev-only minimal Runtime execution path로 검증되었다. - 현재 상태는 portfolio-grade pipeline foundation이며, production worker daemon, persistent queue/database, file upload, production frontend beyond Local Studio, auth/billing은 future work로 명확히 분리한다. @@ -38,7 +38,7 @@ Supporting sidecar: InferEdgeEnv -> local-first run evidence registry / comparability checker Runtime Intelligence smoke chain: -InferEdgeOrchestrator operation feed +InferEdgeOrchestrator operation feed / operation_risk_rollup -> InferEdgeEnv telemetry history / comparability-first regression context -> optional InferEdgeAIGuard deterministic runtime anomaly evidence -> InferEdgeLab Runtime Intelligence Risk Summary / deployment risk report @@ -102,7 +102,7 @@ Rule + evidence diagnosis layer. Forge summary, Runtime worker_response, Lab res Run evidence registry / comparability checker. Edge AI inference benchmark result를 local artifact와 SQLite registry로 고정하고, same-condition / conditional / no comparability judgement를 제공한다. Env는 deployment decision을 소유하지 않으며, Lab의 validation / decision layer와 분리된 evidence portability boundary다. **InferEdgeOrchestrator** -Runtime operation context provider. Orchestrator는 queue depth, deadline miss, fallback, thermal/resource pressure 같은 operation evidence를 EdgeEnv feed 후보로 제공한다. 이 context는 supplemental evidence이며, regression 계산 owner는 EdgeEnv, final deployment decision owner는 Lab으로 유지된다. +Runtime operation context provider. Orchestrator는 queue depth, deadline miss, fallback, thermal/resource pressure와 `operation_risk_rollup` 같은 operation evidence를 EdgeEnv feed 후보로 제공한다. 이 context는 supplemental evidence이며, regression 계산 owner는 EdgeEnv, final deployment decision owner는 Lab으로 유지된다. ## 5. Key Implemented Features @@ -121,7 +121,7 @@ Runtime operation context provider. Orchestrator는 queue depth, deadline miss, - AIGuard guard_analysis preservation in Lab deployment decision/report smoke - Local Studio browser workflow for Run, Import, Jetson command helper, demo evidence replay, Compare View, and Lab-owned Deployment Decision inspection - InferEdgeEnv run artifact bundle, SQLite registry, export/import, sampler metadata, resource lookup, and comparability-first report UX -- Runtime Intelligence smoke evidence chain from Orchestrator operation feed through EdgeEnv telemetry history/regression and AIGuard deterministic anomaly evidence into a Lab-owned Runtime Intelligence Risk Summary +- Runtime Intelligence smoke evidence chain from Orchestrator operation feed and `operation_risk_rollup` through EdgeEnv telemetry history/regression and AIGuard deterministic anomaly evidence into a Lab-owned Runtime Intelligence Risk Summary - Runtime Intelligence bundle/report gates for EdgeEnv producer-owned telemetry coverage, Orchestrator `edgeenv_mapping_hint`, AIGuard raw-context preservation, and Lab ownership wording - Core repository README pipeline summary sync plus InferEdgeEnv sidecar positioning @@ -145,7 +145,7 @@ Recent validation evidence: - Local Studio demo evidence: `/studio` can load bundled ONNX Runtime CPU and TensorRT Jetson Runtime result fixtures from `examples/studio_demo`, keep the demo pair selectable in Recent jobs while the local server process is alive, and show TensorRT Jetson vs ONNX Runtime CPU comparison in the browser. The fixture-backed evidence records ONNX Runtime CPU FP32 at mean `45.4299 ms` / p99 `49.2128 ms` / `22.0119 FPS` and TensorRT Jetson FP16 25W at mean `10.066401 ms` / p99 `15.548438 ms` / `99.340373 FPS`, about a `4.51x` TensorRT speedup for this demo pair. - YOLOv8 COCO subset evaluation: a 10-image local person-detection subset with 89 ground-truth boxes is converted into a COCO-style annotation fixture and evaluated through the `yolov8_coco` preset. The generated report records metric backend `simplified`, mAP@50 0.1410, precision 0.2941, recall 0.1685, and structural validation passed. This is documented as subset workflow evidence, not a full COCO benchmark claim. `pycocotools` remains an optional explicit backend. - Validation problem cases: the demo bundle includes annotation-missing, invalid detection structure, contract shape mismatch, and latency regression reports. These show that InferEdge records review/block evidence explicitly instead of presenting every validation path as successful. -- Runtime Intelligence smoke chain: the committed bundle manifest and report gates verify Orchestrator `edgeenv_runtime_telemetry_feed`, EdgeEnv `runtime_telemetry_context.history.telemetry_coverage`, AIGuard deterministic runtime operation evidence, and Lab Runtime Intelligence Risk Summary ownership. This is a local-first artifact chain, not production observability or a runtime control plane. +- Runtime Intelligence smoke chain: the committed bundle manifest and report gates verify Orchestrator `edgeenv_runtime_telemetry_feed` / `operation_risk_rollup`, EdgeEnv `runtime_telemetry_context.history.telemetry_coverage`, AIGuard deterministic runtime operation evidence, and Lab Runtime Intelligence Risk Summary ownership. This is a local-first artifact chain, not production observability or a runtime control plane. - Jetson EdgeEnv preservation smoke: the InferEdge entrypoint replayed a 32-frame `device_local_starter` path on Jetson with a user-provided `yolov8n.onnx`, live `tegrastats`, process resource snapshot capture, and `--edgeenv-run-evidence`. The run reached max queue depth `6`, recorded dropped/fallback `29 / 29`, deadline misses `18`, parsed `4` `tegrastats` samples, observed max temperature / RAM `42.843 C / 999 MB`, stored EdgeEnv run `run-20260529-034704-fbf753f0` with `runtime_operation_summary`, and produced AIGuard `blocked/high` plus Lab `blocked` deployment risk evidence. This is device-local starter smoke, not decoded YOLO accuracy, live camera, production remote execution, or thermal endurance validation. The direct Runtime execution result includes `deployment_decision`. Its `unknown` value is expected before Lab compare/report because the worker response has not yet been compared by Lab. @@ -172,7 +172,7 @@ Forge summary - **Provenance-aware validation:** Artifact/source hash, manifest source model identity, and runtime provenance are treated as first-class deployment evidence. - **SaaS-ready API + async job workflow:** Lab has API response contracts, in-memory async job stubs, and worker request/response mapping without prematurely adding DB/queue infrastructure. - **Deterministic rule-based diagnosis:** AIGuard uses rule + evidence detectors instead of vague LLM judgement. -- **Runtime Intelligence evidence chain:** Orchestrator operation context, EdgeEnv telemetry coverage/regression, AIGuard deterministic anomaly evidence, and Lab deployment risk reporting are connected through lightweight fixtures and gates without adding a new repo or production monitoring platform. +- **Runtime Intelligence evidence chain:** Orchestrator operation context and `operation_risk_rollup`, EdgeEnv telemetry coverage/regression, AIGuard deterministic anomaly evidence, and Lab deployment risk reporting are connected through lightweight fixtures and gates without adding a new repo or production monitoring platform. - **Jetson Runtime Intelligence preservation evidence:** device-local ONNX Runtime probe and live `tegrastats` evidence can be preserved through EdgeEnv's local run registry and rendered back into a Lab-owned deployment risk report without changing existing Runtime result or compare contracts. - **Deployment decision ownership:** Lab keeps final deploy/review/blocked ownership while preserving optional guard evidence. - **Local-first Studio demo:** The browser UI can replay real validation evidence locally without adding DB, queue, upload, auth, billing, or production SaaS infrastructure.