debugger123 commited on
Commit
74a5ccc
·
verified ·
1 Parent(s): 773af7a

Simplify README to Aug and Gen record counts

Browse files
Files changed (2) hide show
  1. README.md +3 -68
  2. SHA256SUMS +1 -1
README.md CHANGED
@@ -1,81 +1,16 @@
1
  ---
2
  license: cc-by-nc-sa-4.0
3
- language:
4
- - en
5
- task_categories:
6
- - text-generation
7
- tags:
8
- - text-to-sql
9
- - sql
10
- - synthetic-data
11
- size_categories:
12
- - 100K<n<1M
13
  configs:
14
  - config_name: aug
15
  data_files:
16
  - split: train
17
- path:
18
- - data/aug/spider-*.parquet
19
- - data/aug/bird-*.parquet
20
- - data/aug/ehrsql-*.parquet
21
  - config_name: gen
22
  data_files:
23
  - split: train
24
- path: data/gen/gen-*.parquet
25
  ---
26
 
27
  # DataFlow-SQL-150k
28
 
29
- This package contains the **DataFlow-SQL-Aug** and **DataFlow-SQL-Gen** snapshots corresponding to the authors' November 2026 TKDE revision. The combined release contains **150,764 records**. The branch names and counts follow the manuscript; the files preserve the supplied records without further filtering or deduplication.
30
-
31
- | Configuration | Source | Records |
32
- |---|---|---:|
33
- | `aug` | SQLFlow / Spider | 30,502 |
34
- | `aug` | SQLFlow / BIRD | 33,163 |
35
- | `aug` | SQLFlow / EHRSQL | 11,721 |
36
- | **`aug` subtotal** | **DataFlow-SQL-Aug** | **75,386** |
37
- | `gen` | Author-supplied `75378.jsonl` | 75,378 |
38
- | **Total** | **DataFlow-SQL-150k** | **150,764** |
39
-
40
- ## Load the data
41
-
42
- Replace `YOUR_USERNAME/DataFlow-SQL-150k` with the actual dataset repository ID after upload.
43
-
44
- ```python
45
- from datasets import load_dataset
46
-
47
- repo_id = "YOUR_USERNAME/DataFlow-SQL-150k"
48
- aug = load_dataset(repo_id, "aug", split="train")
49
- gen = load_dataset(repo_id, "gen", split="train")
50
-
51
- # Augmentation sources can be selected using the original data_source field.
52
- spider_aug = aug.filter(lambda x: x["data_source"] == "spider")
53
- ```
54
-
55
- The two configurations preserve different original schemas: Aug uses `SQL`, `evidence`, and `cot_reasoning`; Gen uses `sql`, `external_knowledge`, and `cot`. See [DATA_SCHEMA.md](DATA_SCHEMA.md). An application combining the branches should map these fields explicitly.
56
-
57
- `train` is the packaging split name for each complete supplied branch. This release does not create validation or test splits, establish a new experimental partition, or guarantee separation from external benchmark evaluation data. Within each source file, record order is retained. Source-to-shard row ranges are provided in [source_manifest.json](source_manifest.json).
58
-
59
- ## Contents and provenance
60
-
61
- - **Aug:** the three SQLFlow JSON files from [debugger123/SQLFlow](https://huggingface.co/datasets/debugger123/SQLFlow), with upstream metadata revision `887581f9b2baff78ea9b9347461d3bd8ecc0929c`. The original README and snapshot metadata are preserved under `provenance/`. The local file SHA-256 hashes identify the exact inputs used here.
62
- - **Gen:** the authors supplied a file named `75378.jsonl` and identified it as the 75,378-record generation release. Its content hash is recorded in the source manifest. No additional upstream provenance is inferred from database identifiers.
63
- - **Code entry points:** [Text2SQL-Flow](https://github.com/TechNomad-ds/Text2SQL-Flow) and [DataFlow](https://github.com/OpenDCAI/DataFlow).
64
-
65
- The packaging operation converts the source JSON/JSONL to Parquet with Zstandard compression. It does not rewrite questions, SQL, reasoning, or metadata, and does not add or remove record fields. Empty strings remain empty strings. All original fields are strings. The package does not include database files or a separate schema catalog; Gen records do not have a dedicated schema field. Executing SQL requires the appropriate external database resources.
66
-
67
- ## Verification and limitations
68
-
69
- All 150,764 records were checked for JSON validity, consistent branch schemas, expected row counts, and exact field-value preservation after Parquet conversion. No exact full-record duplicates were found within either branch. Source and output file hashes, field statistics, empty-string counts, and original category distributions are included in [dataset_statistics.json](dataset_statistics.json) and [source_manifest.json](source_manifest.json).
70
-
71
- These packaging checks do **not** establish SQL correctness, execution validity, question–SQL alignment, reasoning faithfulness, or semantic uniqueness. Original complexity/style/variation labels are retained as source metadata and have not been reassessed. The manuscript reports quality limitations in sampled records and a conservative audit identifying one Aug query matching an EHRSQL development query under its stated criterion; this packaging step does not remove that record or revise historical results. Users constructing evaluation protocols should account for benchmark overlap and the relevant database resources.
72
-
73
- ## License
74
-
75
- **Aug:** the upstream SQLFlow dataset card declares **CC-BY-NC-SA-4.0**. Its original attribution and license declaration are preserved under `provenance/`.
76
-
77
- **Gen:** the author confirmed **CC-BY-NC-SA-4.0** for this release on November 15, 2026. Both configurations are therefore released under **CC-BY-NC-SA-4.0**, as declared in the repository YAML metadata and [LICENSE](LICENSE). See [LICENSE_STATUS.md](LICENSE_STATUS.md) for the declaration sources.
78
-
79
- ## Paper reference
80
-
81
- This local package was prepared against the authors' current TKDE revision dated November 2026. The publisher DOI, public manuscript URL, and final BibTeX citation have not been supplied and are not invented here. Add the chosen public paper reference before publication.
 
1
  ---
2
  license: cc-by-nc-sa-4.0
 
 
 
 
 
 
 
 
 
 
3
  configs:
4
  - config_name: aug
5
  data_files:
6
  - split: train
7
+ path: data/aug/*.parquet
 
 
 
8
  - config_name: gen
9
  data_files:
10
  - split: train
11
+ path: data/gen/*.parquet
12
  ---
13
 
14
  # DataFlow-SQL-150k
15
 
16
+ This dataset contains **Aug (75,386 examples)** and **Gen (75,378 examples)**, for a total of **150,764 examples**.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
SHA256SUMS CHANGED
@@ -2,7 +2,7 @@
2
  c35328c0d3d29a09208b74671422caa1753290360e1f79e17a93b34e234a3215 DATA_SCHEMA.md
3
  cb80253379d575d0fffda41fed50234ddad5906219588e9883ecd957aad9efac LICENSE
4
  d2cb4d7fb456049061d028fc1da224e512d889ca8e2511abcfb7760c07f51d5c LICENSE_STATUS.md
5
- 57bf13ccc87bb23a21f05419517f46feee80085902a6f3514ed2b31b8415deda README.md
6
  e68bcb1de3f8fde0b1c3215c0080f38e5be7decc09d30b56a54772dd76bdd88b data/aug/bird-00000.parquet
7
  db6053666d1d52156db05db3f1cdf4abefeeba43720411ad7bca4bcfffb9ab11 data/aug/bird-00001.parquet
8
  b0b8db80d0aa792d739fc008ff970fee045c69d7e901bffe574dbe894f51a386 data/aug/ehrsql-00000.parquet
 
2
  c35328c0d3d29a09208b74671422caa1753290360e1f79e17a93b34e234a3215 DATA_SCHEMA.md
3
  cb80253379d575d0fffda41fed50234ddad5906219588e9883ecd957aad9efac LICENSE
4
  d2cb4d7fb456049061d028fc1da224e512d889ca8e2511abcfb7760c07f51d5c LICENSE_STATUS.md
5
+ a16252e6120e89b277cf0a8d2a661c7534dbb15b6e6931ca6266487c3ba76e96 README.md
6
  e68bcb1de3f8fde0b1c3215c0080f38e5be7decc09d30b56a54772dd76bdd88b data/aug/bird-00000.parquet
7
  db6053666d1d52156db05db3f1cdf4abefeeba43720411ad7bca4bcfffb9ab11 data/aug/bird-00001.parquet
8
  b0b8db80d0aa792d739fc008ff970fee045c69d7e901bffe574dbe894f51a386 data/aug/ehrsql-00000.parquet