Skip to content

v0.1.0 - Evidence-first public foundation

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Xiaomiju-x Xiaomiju-x released this 03 Sep 10:00
v0.1.0
ab080d7

RDK Frontier Lab · first public release

An evidence-first, offline-first foundation for long-term RDK X5 / S100 / S600
algorithm portability work. This is an alpha tooling release and historical
research snapshot, not a finished library of validated board models.

Included

  • Dependency-free Python CLI: doctor, catalogue, arithmetic demo, manifest
    verification, scoped receipt evaluation and independent-target comparison.
  • Optional NumPy weighted-Kabsch geometry recipe and first-party MIT neural
    scorer source; no pretrained model or vendor runtime is bundled.
  • 59 dated registry entries (26 official / 17 frontier / 16 research), source
    references, asset-review notes and nine sanitized historical observations.
    Entry count is not a unique-model or successful-deployment count.
  • SwinIR and RTMPose failure studies that preserve the original accuracy gates.
  • English/Chinese entry points, contribution and maintenance policies, issue
    forms, versioned schemas and a deterministic source SPDX inventory.

Verified for this release

  • 74 regression tests; Linux and Windows matrix spans Python 3.10–3.14.
  • At least 90% branch-aware core coverage enforced by CI; local measurement 99%.
  • Lint, formatting, generated-data synchronization, schema/local-link checks,
    exact staged-blob publication audit, wheel/sdist inspection and clean wheel
    installation outside the checkout passed.
  • CI run
    and CodeQL run
    passed for commit ab080d723be0c4b6d94be3b864df5c65c845f449.
  • Distribution files attached here are from that successful hosted CI build.
    Verify them against SHA256SUMS. The SPDX attachment describes shipped source,
    not the user's installed dependencies or third-party model assets.

Quickstart

With Python 3.10 or newer, clone this repository and run:

python -m rdk_lab doctor
python -m rdk_lab demo
python -m rdk_lab catalog --target s600

Or download the attached wheel and install it in your own virtual environment:

python -m pip install --no-index --no-deps rdk_frontier_lab-0.1.0-py3-none-any.whl
rdk-lab demo

The demo returns DEMO_PASS, not BPU_PASS. No automatic network requests,
device connection, model loading, training, compiler launch or robot control.
The package is not published to PyPI in this release.

Important limits

Historical measurements are maintainer-reported. Original private receipts,
input media, checkpoints, ONNX and BIN/HBM files are not included. Original-file
digest commitments cannot independently prove their contents. Some compiled
models genuinely execute but fail numerical fidelity; those failures stay failed.

For the exact scope, see results,
reproducibility
and the roadmap.

中文摘要

首版已建立三板算法研究的长期维护基础：可运行工具、几何算法源码、双语入口、
算法台账、失败案例、许可与来源说明、跨平台 CI 和带哈希的发布包。
这是 v0.1.0 预发布版，不是“59 个算法全部可直接部署”的宣称。
后续重点是逐条补齐可公开复现的完整模型配方与真实数据评测；算法实验保持既有暂停状态。

Assets 6
Loading