Skip to content

[milestone] Archive awarded SLIM-ARC release - #2

Merged
ouyangyipeng merged 10 commits into
mainfrom
feat/project-closeout
Aug 22, 2026
Merged

ouyangyipeng merged 10 commits into
mainfrom
feat/project-closeout

Conversation

@ouyangyipeng

Copy link
Copy Markdown
Member

Summary

  • archives the 2026 national first-prize final report, presentation, overview, and cross-platform evidence
  • adds a pinned, idempotent bootstrap/build/test workflow for new devices
  • establishes public Wiki sources, contribution/security/citation files, templates, and fast macOS/Ubuntu CI
  • removes internal Agent/plan/session state and generated artifacts from the final tree without rewriting history

Validation

  • 10 C++ core targets passed
  • 346 Python tests passed
  • make check passed, including public-tree and relative-link checks
  • pinned llama.cpp bootstrap passed twice; full macOS Release build passed
  • 41-page report rebuilt and visually inspected; final presentation/PDF are 39 pages; overview is 14 pages
  • gitleaks: 0 findings in current HEAD and complete reachable history

Release

After rebase merge and fresh-clone verification, publish v1.1.0 because the repository already contains an immutable historical v1.0.0 release.

Root cause: KV offload growth called Linux-only mremap directly, and an older integration fixture no longer modeled the required Qwen3-Next seam.

Solution: Add a cross-platform mapping resize helper, cover it with a dedicated unit test, expose list/all test entrypoints, and synchronize the integration fixture.

Risks: The non-Linux path maps the expanded file before releasing the old mapping, temporarily holding both virtual mappings.

Dependency: llama.cpp@360e1349f0009c5ad99d21e3c4546b707addc68a.

Links: docs/future-work.md
Root cause: New machines depended on an unpinned external llama.cpp tree, local absolute paths, and undocumented test entrypoints.

Solution: Add uv-managed tooling, pinned idempotent bootstrap, unified Make targets, portable demo environment variables, and public-tree validation.

Risks: Bootstrap requires network access and builds against one intentionally pinned upstream revision.

Dependency: llama.cpp@360e1349f0009c5ad99d21e3c4546b707addc68a.

Links: docs/guide/getting-started.md
Root cause: Final presentation sources, macOS native measurements, expert-review links, and preliminary evidence were split across local folders and stale repository locations.

Solution: Publish the verified 39-page deck, project overview, structured macOS evidence, expert index, historical roadmap, and preliminary experiment archive with repository-relative links.

Risks: Binary presentation assets increase clone size; raw logs are retained only when they support reproducibility.

Dependency: Qwen3-Next artifacts are identified by SHA-256 but are not distributed.

Links: https://www.bilibili.com/video/BV1iHbB6CEPb/
Root cause: The competition repository lacked a reproducible onboarding path, current license guidance, Wiki source, contribution policy, issue templates, and cross-platform CI.

Solution: Rebuild the README and guides, add community health files, make docs/wiki canonical, add Wiki synchronization, and define macOS/Ubuntu CI without model downloads.

Risks: GitHub Wiki synchronization is an explicit maintainer action and CI integration depends on upstream network availability.

Dependency: GitHub Actions and GitHub Wiki.

Links: https://slim.nexa-lang.com/
Root cause: The public tree retained internal Agent instructions, planning records, one-off GitLab publishers, generated LaTeX files, and accidental executable bits on evidence.

Solution: Remove those files from the current tree without rewriting history, strengthen ignore and binary attributes, and normalize documentation and log permissions.

Risks: Historical commits still contain non-secret internal records by design; they remain available for technical provenance.

Dependency: ef48fd8 archives the reusable technical material before cleanup.

Links: docs/archive/README.md
Root cause: Python bytecode generation inherited a host cache path, and Wiki-style extensionless links were not validated.

Solution: Keep bytecode caches inside the repository ignore boundary and validate public relative links, including GitHub Wiki page names.

Risks: The Markdown checker intentionally covers public maintenance documents rather than every archived competition note.

Dependency: NA

Links: NA
Root cause: NA

Solution: Rebuild the authoritative 41-page report from its committed LaTeX sources and verified structured evidence.

Risks: PDF metadata and binary encoding may differ across TeX Live versions while the rendered content remains equivalent.

Dependency: TeX Live 2026

Links: NA
Copilot AI lite review requested due to automatic review settings August 22, 2026 13:25

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR prepares the public, archived “award release” of SLIM-ARC by adding reproducible bootstrap/build/test automation, public-facing maintenance/docs/Wiki sources, and by cleaning internal planning artifacts from the tracked tree, while also landing a small portability refactor (KV mmap resize) with a regression test.

Changes:

  • Added pinned bootstrap + unified Makefile targets + GitHub Actions CI to bootstrap, build, test, and verify idempotent integration.
  • Added public-facing governance/docs (SECURITY, CONTRIBUTING, citation/changelog/authors, Wiki source pages) and link/public-tree validation scripts.
  • Added a cross-platform KV mmap resize helper (resize_file_mapping) and a dedicated C++ regression test; removed generated report artifacts and internal plan docs from the final tree.

Reviewed changes

Copilot reviewed 61 out of 361 changed files in this pull request and generated 2 comments.

Show a summary per file
File Description
tests/test_apply_pressure_admission.py Extends upstream fixture coverage (adds src/models/ + Qwen3Next stub) for idempotent apply checks.
tests/run-cpp-unit.sh Adds list/all modes and allowlists a new KV-eviction unit test target.
tests/cpp/test-slim-arc-kv-eviction.cpp New regression test for KV mapping resize behavior.
patches/llama-upstream/slim-arc-kv-eviction.h Declares detail::resize_file_mapping helper for reuse/testing.
patches/llama-upstream/slim-arc-kv-eviction.cpp Factors mmap growth into resize_file_mapping and adds non-Linux fallback implementation.
scripts/bootstrap-dev.sh Adds pinned, idempotent bootstrap for fetching llama.cpp at a fixed revision and applying integration.
Makefile Adds bootstrap/build/test/check/docs/clean entrypoints for new-device onboarding and CI.
.github/workflows/ci.yml Adds fast macOS/Ubuntu checks + pinned integration bootstrap/idempotence build job.
pyproject.toml Establishes Python project metadata + dev dependencies + pytest/ruff config for tooling/scripts.
scripts/check-public-tree.py Adds guardrail to reject accidental internal state/oversized files/local-path markers.
scripts/check-markdown-links.py Adds repository-relative Markdown link validation for public docs.
scripts/sync-github-wiki.sh Adds a helper to sync docs/wiki/*.md to the GitHub Wiki repo.
SECURITY.md Adds public vulnerability reporting policy.
CONTRIBUTING.md Adds contribution workflow and evidence requirements.
CODE_OF_CONDUCT.md Adds community conduct policy.
CITATION.cff Adds citation metadata for the project and final technical report.
CHANGELOG.md Adds release notes for 1.1.0 and references earlier 1.0.0.
AUTHORS.md Adds author/advisor attribution list.
NOTICE Adds third-party notice for llama.cpp integration and referenced artifacts.
.editorconfig Adds repo-wide formatting defaults (LF, indentation, etc.).
.gitattributes Normalizes text eol behavior and marks binary artifacts as binary.
.gitignore Expands ignores for reproducible/report build artifacts and local demo/service state.
reports/Competition_Report/sections/07_appendix.tex Updates license description in report appendix.
reports/Competition_Report/main.blg Removes generated BibTeX log artifact from tracked tree.
plan/audit/01-v2-audit-fix-plan.md Removes internal audit plan from final public tree.
plan/23-v3-pressure-aware-prefetch.md Removes internal experiment plan from final public tree.
plan/22-v2-macos-80b-constrained-benchmark.md Removes internal experiment plan from final public tree.
plan/21-v2-integrate-team-branches.md Removes internal integration plan from final public tree.
plan/21-v1-integrate-team-branches.md Removes internal integration plan from final public tree.
plan/17-v2-gitlab-clean-final.md Removes internal GitLab clean-history plan from final public tree.
plan/17-v1-gitlab-clean-final.md Removes internal GitLab clean-history plan from final public tree.
plan/16-v1-report-round7.md Removes internal report iteration plan from final public tree.
plan/15-v1-report-round5.md Removes internal report iteration plan from final public tree.
plan/13-v1-readme-release-gitlab.md Removes internal release planning doc from final public tree.
docs/wiki/Home.md Adds canonical Wiki home page source (synced to GitHub Wiki).
docs/wiki/_Sidebar.md Adds canonical Wiki sidebar navigation source.
docs/wiki/Getting-Started.md Adds quickstart instructions for Wiki.
docs/wiki/Architecture.md Adds high-level architecture summary for Wiki.
docs/wiki/Results-and-Evidence.md Adds evidence/results index pointer for Wiki.
docs/wiki/Reproducing-Experiments.md Adds reproduction summary and pointer for Wiki.
docs/wiki/Model-Artifacts.md Documents model artifact policy and pointers.
docs/wiki/Hardware-Notes.md Documents platform scope and evidence entrypoint.
docs/wiki/FAQ.md Adds FAQ for common repository/repro questions.
docs/wiki/Contributing.md Adds Wiki-level contributing pointer.
docs/wiki/Citation.md Adds Wiki-level citation pointer.
docs/guide/getting-started.md Adds detailed getting-started guide (bootstrap + pinned upstream).
docs/guide/development.md Documents patch-module development model + test/allowlist expectations.
docs/guide/reproducing-experiments.md Adds reproduction contract guidance and evidence expectations.
docs/guide/model-artifacts.md Documents model upstream revisions + SHA-256 checks for reproducibility.
docs/guide/report-build.md Documents how to rebuild the finals report and which artifacts must not be committed.
docs/guide/release-maintenance.md Documents release policy, tag immutability, and security reporting.
docs/licensing.md Adds license matrix (Apache-2.0 code, CC BY 4.0 reports, etc.).
docs/media.md Adds media links (site/videos/pdfs) and notes release-asset handling.
docs/future-work.md Adds scoped future work roadmap for post-competition evolution.
docs/archive/README.md Introduces “archive” section and what is/ isn’t current guidance.
docs/archive/preliminary-results/INDEX.md Adds index for outdated/invalidated preliminary results.
logs/ablation/raw-80b/model-sha256-20260811-132145.txt Adds SHA-256 capture for model artifact used in evidence.
logs/ablation/raw-80b/env-yituodabian-20260811-132612.txt Adds environment capture for a raw evidence run.
logs/ablation/raw-80b/80b-yituodabian-baseline-pp32tg16-low-20260811-132643.txt Adds baseline raw evidence log.
logs/ablation/raw-80b/80b-yituodabian-baseline-pp32tg16-low-20260811-171246.txt Adds baseline raw evidence summary log.
logs/ablation/raw-80b/80b-yituodabian-slimarc-full-pp32tg16-low-20260811-181434.txt Adds SLIM-ARC raw evidence log (file added in PR metadata).
docs/macos_test_notes/2026-08-21/optimized-chat-response.json Adds macOS demo/evidence response capture.
docs/macos_test_notes/2026-08-21/native-metal/slim-arc-cpu-t14-tg32.log Adds native-metal macOS evidence log.
docs/macos_test_notes/2026-08-21/native-metal/metal-expert-diagnostic.log Adds native-metal macOS diagnostic log.
docs/macos_test_notes/2026-08-19/unconstrained-agent/harness-overlay.example.yml Adds example harness overlay for local SLIM-ARC server usage.
docs/macos_test_notes/2026-08-19/unconstrained-agent/harness-demo-preset/preset.yml Adds demo preset metadata for harness.
docs/macos_test_notes/2026-08-19/unconstrained-agent/harness-demo-preset/agent.cordis.yml Adds harness agent configuration used for a macOS demo.
docs/rk3588_test_notes/2026-08-05-昨日实验归档/debug日志/test-cold-hot.sh Adds archived RK3588 debug script for cold/hot test notes.
docs/pi5_4GB_test_notes/** Adds Raspberry Pi 5 evidence bundle (wall time, stdout/stderr, env/config snapshots).

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment on lines +3 to +7
#include <cassert>
#include <cstdint>
#include <fcntl.h>
#include <sys/mman.h>
#include <unistd.h>
Comment thread scripts/check-public-tree.py Outdated
Comment on lines +19 to +20
TEXT_SUFFIXES = {".md", ".txt", ".toml", ".yml", ".yaml", ".json", ".tex", ".py", ".sh"}
FORBIDDEN_TEXT = ("/Users/bytedance/", "file:///Users/", "/private/tmp/")
Root cause: A C++ test relied on an indirect cstring include, while report build proofs assumed a complete TeX toolchain in every fast-test environment.

Solution: Include cstring explicitly and skip only TeX build proofs when xelatex or bibtex is unavailable; runtime and evidence-import tests remain mandatory.

Risks: Environments without TeX must use make docs on a documented report-build host to exercise the skipped build proofs.

Dependency: NA

Links: https://github.com/Nexa-Language/SLIM-ARC/actions/runs/32575705525
Root cause: Hot-cache tests expressed expert sizes as page counts, producing different cache capacity on 4 KiB Linux and 16 KiB macOS pages.

Solution: Use page-aligned fixed byte sizes so both platforms exercise identical one-entry and two-entry budget boundaries.

Risks: NA

Dependency: NA

Links: https://github.com/Nexa-Language/SLIM-ARC/actions/runs/32575938244
Root cause: Raw macOS logs and legacy demo scripts retained personal absolute paths, while the public-tree guard skipped several evidence suffixes.

Solution: Normalize macOS evidence paths, require portable Pi script inputs, scan additional text artifacts, and preserve remote /home paths only inside identified raw-evidence directories.

Risks: Pi experiment launchers now require callers to provide benchmark, model, and result paths explicitly.

Dependency: NA

Links: #2
@ouyangyipeng
ouyangyipeng merged commit cf44dc2 into main Aug 22, 2026
3 checks passed
@ouyangyipeng
ouyangyipeng deleted the feat/project-closeout branch August 22, 2026 13:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants