lmc7150/latent-profiler-archive

latent_profiler完整归档存储库，是lucid-computing-labs/latent_profiler仓库在2026-04-26的快照，包含所有模型检查点和扫描结果等重文件（即被.gitignore忽略的工件）。数据集包含32,748个文件，约47 GB，采用单一存储桶、转储式结构，包括代码、论文、结果和所有权重。具体内容包括：Qwen2.5-7B世界模型检查点（三个版本，每个14 GB）、xlarge cVAE检查点（约470 MB每个）、较小cVAE变体（约119 MB每个）、潜在GPU容量扫描（约1.7 GB）、自改进循环迭代（约908 MB）、B200 fp16映射扫描、形状立方体计划扫描（约209 MB）、论文LaTeX源文件（约6 MB）、外部引用PDF（约116 MB）、Python包源代码（约1.3 MB）以及脚本和配置。缺失内容包括：.git目录（需从GitHub克隆获取历史）、.venv虚拟环境（需通过uv sync重建）、外部数据（需从上游URL重新克隆）和LaTeX构建工件（由latexmk重新生成）。恢复方法包括从GitHub克隆代码、从该数据集下载重文件，并克隆外部语料库。

latent_profiler -- full archive bucket, snapshot of the lucid-computing-labs/latent_profiler repo on 2026-04-26, including all model checkpoints and sweep results that are heavy artifacts (i.e., .gitignore-d artifacts). The dataset contains 32,748 files, approximately 47 GB, in a single bucket, dump-style structure, encompassing code, papers, results, and all weights. Key contents include: Qwen2.5-7B world-model checkpoints (three versions, 14 GB each), xlarge cVAE checkpoints (~470 MB each), smaller cVAE variants (~119 MB each), latent-GPU capacity sweeps (~1.7 GB), self-improvement loop iterations (~908 MB), B200 fp16 mapping sweeps, shape-cube plan sweeps (~209 MB), paper LaTeX sources (~6 MB), external citation PDFs (~116 MB), Python package source (~1.3 MB), and scripts and configurations. Missing items include: .git directory (clone from GitHub for history), .venv environment (recreate via uv sync), external corpora (re-clone from upstream URLs), and LaTeX build artifacts (regenerated by latexmk). Restoration involves cloning code from GitHub, downloading heavy artifacts from this dataset, and cloning external corpora.