--- license: other license_name: mixed-see-readme tags: [ocr, archive] --- # OCR projects archive This is a private archive of the smaller OCR projects from `~/ocr` on the OCR server, migrated on 16 Nov 2026. | Folder | What | |---|---| | `OCR_INDRA/` | Indra OCR investigation: code, docs, results, reports, and the git history (sharded). **Indra weights (`models/`) and Indra fine-tunes (`runs/`) are left out** because the AI PUBS OpenRAIL-M licence has commercial limits; get the base model from Datalab instead. The 10.5 GB of installed packages (`.deps`, `.torch-cu126`) are also left out. | | `ocr-tech/` | OCR for the languages of Nepal (Limbu Sirijonga, Newar, Nepali), including git history. Also on GitHub as `Ampixa/ocr-tech`. | | `ocr-tech-mps-integration/` | The MPS integration branch of ocr-tech | | `k2-ampixa-demo-ocr/` | Demo app workers and their model cache | | `ocr-deva-eval-20260924/` | Newar and smoke evaluation sets with predictions | | `jobs/`, `logs/` | Job scripts and logs | | `indra_nepali_exp.bundle` | A git bundle of the OCR_INDRA experiment code | Also left out: `paddlex-official-models/` (official PaddleX downloads: PaddleOCR-VL-1.6, PP-DocLayoutV3, PP-OCRv5/v6 det, UVDoc, devanagari_PP-OCRv5_mobile_rec, PP-LCNet orientation models) and `bin/` (the `uv` tool). Folders with many small files are packed into tar shards, each with a sha256 index (`*.index.json`). **Git histories.** Hugging Face does not allow paths under a `.git/` folder, so each project's `.git` is stored as `/_git/shard-*.tar`. To restore one: ```bash mkdir -p ocr-tech/.git && for t in ocr-tech/_git/shard-*.tar; do tar -xf "$t" -C ocr-tech/.git; done cd ocr-tech && git status # working tree files are already in the repo ```