v0.1.2 - Docs and TensorRT evidence patch

Docs and TensorRT validation evidence patch release.

This release does not change runtime scheduler behavior.

Highlights:

• Records the TensorRT worker/inference path and Jetson TensorRT runtime telemetry evidence.
• Adds Jetson TensorRT contention evidence for scheduler/load-shedding behavior.
• Adds distinct generated detector/classifier TensorRT contention evidence with curated sample telemetry.
• Updates README and PORTFOLIO wording so the project is readable in 30 seconds without implying TensorRT/GPU throughput benchmark claims.
• Keeps raw reports, tegrastats logs, TensorRT plan files, and large model artifacts out of the repository.

Validation:

• Local full pytest: 60 passed.
• GitHub Actions pytest on Python 3.11: passed.

Boundary:
InferEdgeOrchestrator remains a lightweight edge runtime scheduler. v0.1.2 is not a Triton/DeepStream replacement release and not a TensorRT/GPU throughput benchmark.

InferEdgeOrchestrator v0.1.1

Docs and validation evidence patch release only.

No runtime scheduler behavior changes.

Added/packaged since v0.1.0:

• Portfolio brief in English and Korean
• Versioned sample telemetry artifacts
• Architecture documentation
• Validation evidence index
• Documentation link and language-pair pytest coverage
• Config guide documentation
• Tracked InferEdge handoff config sample
• Changelog promotion for v0.1.1

Validation:

• Local pytest: 29 passed
• GitHub Actions CI: pytest (Python 3.11) success

Key docs:

• CHANGELOG.md
• PORTFOLIO.md
• docs/validation_evidence.md
• docs/architecture.md
• configs/README.md
• examples/telemetry/README.md

InferEdgeOrchestrator v0.1.0

InferEdgeOrchestrator v0.1.0

Initial portfolio-ready release for InferEdgeOrchestrator, a lightweight edge inference runtime scheduler for priority/deadline-aware multi-task control, bounded queues, adaptive load shedding, ONNX Runtime workers, and Jetson telemetry.

Highlights

• Scheduler core with config-driven task registration, bounded per-task queues, priority/deadline-aware scheduling, dummy worker, load shedding, and telemetry JSON export.
• ONNX Runtime worker support with config-selectable worker execution and identity ONNX smoke validation.
• Synthetic overload comparison showing detector p95 end-to-end latency improvement from 782.0ms FIFO baseline to 8.0ms with scheduler + load shedding, while low-priority classifier work is intentionally dropped.
• Jetson Orin Nano dummy smoke validation on nano01 with telemetry generation, resource snapshots, and tegrastats parsing.
• Jetson ONNX Runtime smoke validation on nano01 using ONNX Runtime 1.23.2 and CPUExecutionProvider, with output metadata and resource snapshots recorded.
• InferEdge result.json file-based handoff for recommending Orchestrator task latency budgets without importing InferEdge internals.
• English/Korean documentation mirrors and GitHub Actions CI.

Validation

• Local pytest: 22 passed
• GitHub Actions CI: pytest (Python 3.11) success
• Latest main commit: 1b96bc2

Known Limitations

• This release is not a Triton or DeepStream replacement.
• Jetson ONNX Runtime smoke validates the worker path with CPUExecutionProvider; it is not TensorRT/GPU benchmark evidence.
• Jetson smoke artifacts are documented summaries, while raw reports remain ignored under reports/.