Matteoooo46 commited on
Commit
c50a9fc
·
verified ·
1 Parent(s): 47bdcc6

Update README

Browse files
Files changed (1) hide show
  1. README.md +17 -15
README.md CHANGED
@@ -2,28 +2,30 @@
2
  language:
3
  - zh
4
  - en
5
- configs:
6
- - config_name: default
7
- data_files:
8
- - split: train
9
- path: original/llama_factory_rewriter_train.jsonl
10
  ---
11
 
12
  # ReCoEdit-rewriter-sft-data
13
 
14
- Original annotations used by the local ReCoEdit training configuration. The source file is preserved without changes.
15
 
16
- - Records: 1,118,003
17
- - Unique referenced images: 685,652
18
- - Referenced image bytes: 288,802,866,736
19
- - Missing local images at audit time: 0
20
 
21
- ## Format
 
22
 
23
- `instruction`: instruction/system text; `input`: original user instruction; `output`: target rewritten prompt; `images`: reference image paths.
24
 
25
- ## Upload status
 
 
26
 
27
- The original annotation file has been uploaded and verified against its local SHA-256 hash. Image files are not included yet: the full image collection exceeds the current private storage allocation. Image upload is pending a storage/visibility decision.
28
 
29
- Image paths retain their original absolute filesystem locations. Downloading this repository currently provides the annotations only, not a standalone runnable training dataset. No train/validation split has been added.
 
 
 
 
 
 
 
2
  language:
3
  - zh
4
  - en
5
+ license: cc-by-4.0
 
 
 
 
6
  ---
7
 
8
  # ReCoEdit-rewriter-sft-data
9
 
10
+ ReCoEdit training data — image assets and filtered annotations.
11
 
12
+ ## Filtering applied
 
 
 
13
 
14
+ Not included in this release (available or restricted elsewhere):
15
+ - **UltraEdit-500K** subset — see original public release [`BAAI/UltraEdit`](https://huggingface.co/datasets/BAAI/UltraEdit)
16
 
17
+ ## Files
18
 
19
+ - `annotations.jsonl` — filtered training records. Image paths point into `images/<sha256[:2]>/<sha256>.<ext>` inside the tar shards.
20
+ - `images-XXXXX.tar` — content-addressed image shards (~4 GB each).
21
+ - `mapping.jsonl` — audit mapping of original filesystem path → content-addressed archive name.
22
 
23
+ ## Download & extract
24
 
25
+ ```bash
26
+ hf download Matteoooo46/ReCoEdit-rewriter-sft-data --repo-type dataset --local-dir data
27
+ cd data
28
+ for shard in images-*.tar; do tar -xf "$shard"; done
29
+ ```
30
+
31
+ After extraction, images live under `images/<xx>/<sha>.<ext>` — the paths in `annotations.jsonl` resolve directly.