Datasets:

Modalities:
Image
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
MatteoOmenetti commited on
Commit
0daf46e
·
verified ·
1 Parent(s): 01ccdd2

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +91 -0
README.md ADDED
@@ -0,0 +1,91 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cdla-permissive-2.0
3
+ task_categories:
4
+ - text-generation
5
+ tags:
6
+ - code
7
+ - ocr
8
+ size_categories:
9
+ - 1M<n<10M
10
+ ---
11
+ **SynthCodeNet** is a multimodal dataset created for training the **SmolDocling** model. It consists of over **9.3 million** synthetically generated image-text pairs, covering code snippets from **56** different programming languages. Text data was sourced from permissively licensed resources, while images were synthetically generated using LaTeX and Pygments to ensure visual diversity.
12
+
13
+ ---
14
+
15
+ ## Dataset Statistics
16
+
17
+ * **Total samples**: 9,334,257
18
+
19
+ * **Training set**: 8,400,838
20
+ * **Validation set**: 466,703
21
+ * **Test set**: 466,716
22
+
23
+ * **Modalities**: Image, Text
24
+
25
+ * **Image Generation**: Synthetic (LaTeX, Pygments)
26
+
27
+ ### Programming Languages & Sample Counts
28
+
29
+ | Language | Samples | Language | Samples | Language | Samples |
30
+ | -------- | ------- | ---------- | ------- | ----------- | --------- |
31
+ | Ada | 20,094 | Dart | 20,415 | Matlab | 1,170 |
32
+ | Awk | 22,334 | Dockerfile | 99,459 | MoonScript | 6,237 |
33
+ | Bash | 98,950 | Elixir | 20,387 | Nim | 37,236 |
34
+ | C | 599,096 | Erlang | 20,039 | OCaml | 32,297 |
35
+ | C# | 303,720 | FORTRAN | 34,023 | ObjectiveC | 158,398 |
36
+ | C++ | 698,870 | Forth | 5,548 | Octave | 2,537 |
37
+ | CMake | 19,910 | Go | 333,722 | PHP | 249,566 |
38
+ | COBOL | 5,153 | HTML | 245,228 | Pascal | 28,254 |
39
+ | CSS | 236,596 | Haskell | 39,848 | Perl | 33,938 |
40
+ | Ceylon | 8,369 | Haxe | 20,070 | Prolog | 2,058 |
41
+ | Clojure | 20,765 | Java | 698,421 | Python | 1,797,063 |
42
+ | Crystal | 24,720 | JavaScript | 530,899 | Racket | 4,340 |
43
+ | Cuda | 142,344 | Julia | 29,681 | Ruby | 348,976 |
44
+ | Cython | 22,136 | Kotlin | 292,986 | Rust | 344,491 |
45
+ | D | 20,338 | Lisp | 29,749 | SML | 19,333 |
46
+ | Lua | 25,328 | SQL | 493,412 | YAML | 249,011 |
47
+ | Scala | 273,825 | Scheme | 23,242 | VisualBasic | 13,908 |
48
+ | Swift | 25,374 | TypeScript | 255,475 | XML | 246,209 |
49
+ | bc | 249 | dc | 1,713 | | |
50
+
51
+ ---
52
+
53
+ ## Data Format
54
+
55
+ Each dataset entry is structured as follows:
56
+
57
+ ```json
58
+ {
59
+ "images": [PIL Image],
60
+ "texts": [
61
+ {
62
+ "assistant": "<loc_x0><loc_y0><loc_x1><loc_y1><_Language_>CODE_SNIPPET</code>",
63
+ "source": "SynthCodeNetNoImageTag",
64
+ "user": "<code>"
65
+ }
66
+ ]
67
+ }
68
+ ```
69
+
70
+ ---
71
+
72
+ ## Intended Use
73
+
74
+ * Training multimodal models for **document understanding**, specifically:
75
+
76
+ * **Code snippet extraction and transcription**
77
+
78
+ ---
79
+
80
+ ## Citation
81
+
82
+ If you use SynthCodeNet, please cite:
83
+
84
+ ```bibtex
85
+ @article{nassar2025smoldocling,
86
+ title={SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion},
87
+ author={Nassar, Ahmed and Marafioti, Andres and Omenetti, Matteo and Lysak, Maksym and Livathinos, Nikolaos and Auer, Christoph and Morin, Lucas and de Lima, Rafael Teixeira and Kim, Yusik and Gurbuz, A Said and others},
88
+ journal={arXiv preprint arXiv:2503.11576},
89
+ year={2025}
90
+ }
91
+ ```