You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

code-archive-40173

Source-code archive of an ML development workspace (/group/40173/zionyfeng), packaged 2026-10-05. 35 tar files, ~15.2 GB.

Contents

  • Source code and full .git history for 17 repositories (vllm, vllm-laya, verl, verl-omni, verl-omni-rebuild, DiffSynth-Studio, deepseek-harness, AReaL, ms-swift, sglang-omni, Flow-Factory, labs-molt, UniRL, D-OPSD, LightX2V, nano-diffusers, dmd2-official)
  • Standalone project directories: zyf_gly, joyai-upstream, joyai-runs, fast-inference, refs, tools, 4gpu-pi, magi2-* run dirs
  • Top-level scripts: modeltrain.py, installcc.sh

Each tar contains exactly one top-level directory, e.g. tar -xf vllm.tar yields ./vllm/.

Excluded from this archive

Model weights, datasets, and environment/caches were stripped: models/ (1.1T), outputs/ (2.4T), in-repo weights (nano-diffusers outputs/ 768G + data/ 279G, DiffSynth-Studio dataset/, D-OPSD experiment outputs, HF hub caches), virtualenvs and caches (uv_venv, uv_cache, hf_cache, wheels, worktrees, joyai-ref-env), and .claude session state.

Files

  • SHA256SUMS — checksums for all 35 tars
  • 资产盘点.md — full inventory report (git history per repo, size reconciliation, exclusion rules)

Access

This repository is gated: it is publicly listed, but downloading files requires requesting access and being approved by the owner.

Downloads last month
23