xiaotianhan nielsr HF Staff commited on
Commit
c962e85
·
verified ·
1 Parent(s): 3efcc8d

Update task category, add tags, and include survey paper link for InfiMM-WebMath-40B (#4)

Browse files

- Update task category, add tags, and include survey paper link for InfiMM-WebMath-40B (2e66d95e186d5762ad0f417ea9965b23939e6af7)


Co-authored-by: Niels Rogge <[email protected]>

Files changed (1) hide show
  1. README.md +12 -8
README.md CHANGED
@@ -1,21 +1,26 @@
1
  ---
2
- license: odc-by
3
- task_categories:
4
- - text-generation
5
  language:
6
  - en
7
  - zh
 
 
 
 
 
8
  tags:
9
  - pretrain
10
  - multi-modal
11
- size_categories:
12
- - 10B<n<100B
13
  ---
14
 
15
  # InfiMM-WebMath-40B Dataset
16
 
17
  [ArXiv](https://arxiv.org/abs/2409.12568)| [PDF](https://arxiv.org/pdf/2409.12568)
18
 
 
 
 
19
  **InfiMM-WebMath-40B** is a large-scale, open-source multimodal dataset specifically designed for mathematical reasoning tasks. It incorporates both text and images, extracted from web documents, to advance the pre-training of Multimodal Large Language Models (MLLMs). The dataset is tailored to support sophisticated reasoning tasks that involve understanding both text and visual elements like diagrams, figures, and geometric plots.
20
 
21
  ## Dataset Overview
@@ -41,7 +46,7 @@ The dataset is organized in a format that captures both text and images in their
41
  "metadata": { # Metadata containing information about the extraction process (e.g., processing details, timestamps).
42
  "ft_lang_label", # Type of languages detected by fastText
43
  "ft_lang_prob", # Probability of type of language detected by fastText
44
- "math_prob", # First round math content detection with high recal fastText model
45
  "size",
46
  "snap", # Timestamp of Common Crawl snapshot
47
  "text_gpt3_token_len",
@@ -51,12 +56,11 @@ The dataset is organized in a format that captures both text and images in their
51
  "punctuation_ratio",
52
  "nsfw_num_words", # Number of words which are NSFW
53
  "has_unicode_error", # If there's any unicode error exists
54
- "math_prob_llama3", # Probability of second round math detection with high precision fastText model
55
  }
56
  }
57
  ```
58
 
59
-
60
  ### Interleaved Text and Images
61
 
62
  The **text_list** and **image_list** are designed as parallel arrays, maintaining the sequence of the document. This interleaving structure allows models to reconstruct the flow of the original document:
 
1
  ---
 
 
 
2
  language:
3
  - en
4
  - zh
5
+ license: odc-by
6
+ size_categories:
7
+ - 10B<n<100B
8
+ task_categories:
9
+ - image-text-to-text
10
  tags:
11
  - pretrain
12
  - multi-modal
13
+ - mathematical-reasoning
14
+ - geometry
15
  ---
16
 
17
  # InfiMM-WebMath-40B Dataset
18
 
19
  [ArXiv](https://arxiv.org/abs/2409.12568)| [PDF](https://arxiv.org/pdf/2409.12568)
20
 
21
+ This dataset is also discussed in the survey paper [A Survey of Deep Learning for Geometry Problem Solving](https://huggingface.co/papers/2507.11936).
22
+ The accompanying reading list/code for the survey can be found at: https://github.com/majianz/gps-survey
23
+
24
  **InfiMM-WebMath-40B** is a large-scale, open-source multimodal dataset specifically designed for mathematical reasoning tasks. It incorporates both text and images, extracted from web documents, to advance the pre-training of Multimodal Large Language Models (MLLMs). The dataset is tailored to support sophisticated reasoning tasks that involve understanding both text and visual elements like diagrams, figures, and geometric plots.
25
 
26
  ## Dataset Overview
 
46
  "metadata": { # Metadata containing information about the extraction process (e.g., processing details, timestamps).
47
  "ft_lang_label", # Type of languages detected by fastText
48
  "ft_lang_prob", # Probability of type of language detected by fastText
49
+ "math_prob", # First round math content detection with high recal FastText model
50
  "size",
51
  "snap", # Timestamp of Common Crawl snapshot
52
  "text_gpt3_token_len",
 
56
  "punctuation_ratio",
57
  "nsfw_num_words", # Number of words which are NSFW
58
  "has_unicode_error", # If there's any unicode error exists
59
+ "math_prob_llama3", # Probability of second round math detection with high precision FastText model
60
  }
61
  }
62
  ```
63
 
 
64
  ### Interleaved Text and Images
65
 
66
  The **text_list** and **image_list** are designed as parallel arrays, maintaining the sequence of the document. This interleaving structure allows models to reconstruct the flow of the original document: