--- language: - zh - en license: cc-by-4.0 configs: - config_name: default data_files: - split: train path: annotations.jsonl --- # ReCoEdit-rewriter-sft-data ReCoEdit training data — image assets and filtered annotations. ## Files - `annotations.jsonl` — filtered training records. Image paths point into `images//.` inside the tar shards. - `images-XXXXX.tar` — content-addressed image shards (~4 GB each). - `mapping.jsonl` — audit mapping of original filesystem path → content-addressed archive name. The default dataset configuration loads only `annotations.jsonl`. Files under `original/` and `mapping.jsonl` retain original local paths for provenance and are not training inputs. Use the relative image paths in `annotations.jsonl` after extracting the shards. ## Download & extract ```bash hf download Matteoooo46/ReCoEdit-rewriter-sft-data --repo-type dataset --local-dir data cd data for shard in images-*.tar; do tar -xf "$shard"; done ``` After extraction, images live under `images//.` — the paths in `annotations.jsonl` resolve directly.