Datasets:
xiaotianhan
commited on
Commit
•
3efcc8d
1
Parent(s):
aae309d
Update README.md
Browse files
README.md
CHANGED
@@ -34,13 +34,29 @@ The dataset is organized in a format that captures both text and images in their
|
|
34 |
|
35 |
```json
|
36 |
{
|
37 |
-
"URL": "...",
|
38 |
"text_list": [...], # List of extracted text segments, None if the element is an image.
|
39 |
"image_list": [...], # List of image URLs, None if the element is a text segment.
|
40 |
"metadata": {...} # Metadata containing information about the extraction process (e.g., processing details, timestamps).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
}
|
42 |
```
|
43 |
|
|
|
44 |
### Interleaved Text and Images
|
45 |
|
46 |
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:
|
@@ -66,6 +82,10 @@ The **InfiMM-WebMath-40B** dataset was created through a comprehensive multi-sta
|
|
66 |
1. **Base Text Download**: The dataset is available for download as a set of web documents with interleaved text and image URLs.
|
67 |
2. **Image Download**: Users need to download images according to the image URLs provided.
|
68 |
|
|
|
|
|
|
|
|
|
69 |
# License
|
70 |
|
71 |
**InfiMM-WebMath-40B** is made available under an ODC-By 1.0 license; users should also abide by the CommonCrawl ToU: [https://commoncrawl.org/terms-of-use/](https://commoncrawl.org/terms-of-use/). We do not alter the license of any of the underlying data.
|
|
|
34 |
|
35 |
```json
|
36 |
{
|
37 |
+
"URL": "...", # The URL of the source document.
|
38 |
"text_list": [...], # List of extracted text segments, None if the element is an image.
|
39 |
"image_list": [...], # List of image URLs, None if the element is a text segment.
|
40 |
"metadata": {...} # Metadata containing information about the extraction process (e.g., processing details, timestamps).
|
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",
|
48 |
+
"char_repetition_ratio",
|
49 |
+
"word_repetition_ratio",
|
50 |
+
"special_character_ratio",
|
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:
|
|
|
82 |
1. **Base Text Download**: The dataset is available for download as a set of web documents with interleaved text and image URLs.
|
83 |
2. **Image Download**: Users need to download images according to the image URLs provided.
|
84 |
|
85 |
+
### Note
|
86 |
+
|
87 |
+
If you want more data with more precision, you can always use higher thresholds with `math_prob` and `math_prob_llama3` fields in `metadata`.
|
88 |
+
|
89 |
# License
|
90 |
|
91 |
**InfiMM-WebMath-40B** is made available under an ODC-By 1.0 license; users should also abide by the CommonCrawl ToU: [https://commoncrawl.org/terms-of-use/](https://commoncrawl.org/terms-of-use/). We do not alter the license of any of the underlying data.
|