columns list | data list |
|---|---|
[
"question",
"answerable",
"label",
"answers",
"text"
] | [["who portrayed dorothy in 1925 wizard of oz?","answerable","Dorothy Dwan","Dorothy Dwan","<|begin_(...TRUNCATED) |
[
"question",
"answerable",
"label",
"answers",
"text"
] | [["who portrayed dorothy in 1925 wizard of oz?","answerable","Dorothy Dwan","Dorothy Dwan","<|begin_(...TRUNCATED) |
[
"question",
"answerable",
"label",
"answers",
"text"
] | [["who portrayed dorothy in 1925 wizard of oz?","answerable","Dorothy Dwan","Dorothy Dwan","<|begin_(...TRUNCATED) |
[
"question",
"answerable",
"label",
"answers",
"text"
] | [["who portrayed dorothy in 1925 wizard of oz?","answerable","Dorothy Dwan","Dorothy Dwan","<|begin_(...TRUNCATED) |
[
"question",
"answerable",
"label",
"text_for_cls"
] | [["\"who sang the superbowl national anthem in 2017\"","\"answerable\"","1","\"####\\nDocuments:\\n-(...TRUNCATED) |
[
"question",
"answerable",
"label",
"text_for_cls"
] | [["\"who sang the superbowl national anthem in 2017\"","\"answerable\"","1","\"####\\nDocuments:\\n-(...TRUNCATED) |
[
"question",
"answerable",
"label",
"text_for_cls"
] | [["\"who sang the superbowl national anthem in 2017\"","\"answerable\"","0","\"####\\nDocuments:\\n-(...TRUNCATED) |
[
"question",
"answerable",
"label",
"text_for_cls"
] | [["\"who sang the superbowl national anthem in 2017\"","\"answerable\"","0","\"####\\nDocuments:\\n-(...TRUNCATED) |
Robust RAG repository backup
Destination: Jongbin-kr/backup-for-robust-rag (public Hugging Face Dataset repository).
This backup contains the complete regular-file contents of outputs/ and wandb/.
It preserves classifier checkpoints, LoRA adapters, optimizer/scheduler/RNG state,
W&B run records, per-example predictions, metrics, and run configurations.
Symlinks, empty directories, and original permission bits are recorded in manifest.json.
The manifest contains per-file sizes, SHA256 digests, and Git blob hashes.
W&B debug-log symlinks into external cache directories are recorded with their original
targets; the external cache contents are outside this backup and these links are not
recreated automatically.
Prepare and upload
Run from the ICL repository root using the existing Hugging Face authentication:
python scripts/backup_robust_rag.py prepare
python scripts/backup_robust_rag.py upload --disable_xet
Preparation scans for credential literals and creates a hardlinked staging tree in
.backup-for-robust-rag/. It does not duplicate the checkpoint data. Keep source files
unchanged throughout upload. Uploads are resumable through the Hugging Face upload cache.
The HTTP fallback above avoids the repeated Xet request timeouts encountered on this server.
After upload, the script verifies every source file's size and content hash at one
immutable Hub revision and writes verification.json next to this README.
The backup is complete only when that verification file reports verified.
Restore
Download the Dataset repository to a new directory:
hf download Jongbin-kr/backup-for-robust-rag --repo-type dataset --local-dir /path/to/restore
python /path/to/restore/backup_robust_rag.py restore_links --root /path/to/restore
For a reproducible restore, pass --revision with the commit from verification.json
to hf download. The restore command recreates empty directories, file permissions,
and symlinks using relative targets inside the downloaded backup.
Git archive
The Git archive also includes the previously ignored legacy source files, the original
Pareto input CSV, the recovered Table 15--17/Figure 4 CSVs, and the small Arrow dataset
under experiments/0308_reproduce/. Hardcoded credentials in three legacy scripts were
replaced with environment-based authentication. .env and Python caches stay ignored.
The bulk outputs/ and wandb/ trees remain excluded from Git.
- Downloads last month
- 31