Update README.md
Browse files
README.md
CHANGED
|
@@ -18,7 +18,7 @@ size_categories:
|
|
| 18 |
Medium-scale LaTeX corpus from arXiv (math, CS, physics, statistics)
|
| 19 |
</p>
|
| 20 |
|
| 21 |
-
|
| 22 |
|
| 23 |
## 📚 Overview
|
| 24 |
|
|
@@ -41,7 +41,6 @@ This version (~15GB) is intended for:
|
|
| 41 |
* Ablation studies
|
| 42 |
* Resource-constrained training setups
|
| 43 |
|
| 44 |
-
---
|
| 45 |
|
| 46 |
## 🔗 Full Dataset
|
| 47 |
|
|
@@ -51,11 +50,9 @@ For the complete version (~80GB), see:
|
|
| 51 |
|
| 52 |
The 80GB version contains the full dataset without size reduction.
|
| 53 |
|
| 54 |
-
---
|
| 55 |
|
| 56 |
## 📦 Files
|
| 57 |
|
| 58 |
-
```
|
| 59 |
train.jsonl
|
| 60 |
val.jsonl
|
| 61 |
checkpoint.txt
|
|
@@ -74,8 +71,6 @@ Validation split in JSON Lines format.
|
|
| 74 |
Internal processing checkpoint file used during dataset creation.
|
| 75 |
Not required for training.
|
| 76 |
|
| 77 |
-
---
|
| 78 |
-
|
| 79 |
## 🧾 Data Format
|
| 80 |
|
| 81 |
Each line in `train.jsonl` and `val.jsonl` is a JSON object.
|
|
@@ -104,8 +99,6 @@ Example schema:
|
|
| 104 |
* `stat.TH`
|
| 105 |
* `latex` — extracted LaTeX source content
|
| 106 |
|
| 107 |
-
---
|
| 108 |
-
|
| 109 |
## 📊 Dataset Characteristics
|
| 110 |
|
| 111 |
* Size: ~15GB
|
|
@@ -116,8 +109,6 @@ Example schema:
|
|
| 116 |
|
| 117 |
This version is a reduced subset of the full dataset.
|
| 118 |
|
| 119 |
-
---
|
| 120 |
-
|
| 121 |
## 🧹 Processing Pipeline
|
| 122 |
|
| 123 |
The dataset was constructed by:
|
|
@@ -129,8 +120,6 @@ The dataset was constructed by:
|
|
| 129 |
5. Creating train/validation splits.
|
| 130 |
6. Subsampling to create the 15GB medium version.
|
| 131 |
|
| 132 |
-
---
|
| 133 |
-
|
| 134 |
## ⚖️ Licensing
|
| 135 |
|
| 136 |
This dataset contains LaTeX source files from papers hosted on arXiv.
|
|
@@ -142,8 +131,6 @@ Users are responsible for complying with the licensing terms of individual paper
|
|
| 142 |
For license information:
|
| 143 |
[https://arxiv.org/help/license](https://arxiv.org/help/license)
|
| 144 |
|
| 145 |
-
---
|
| 146 |
-
|
| 147 |
## 🎯 Intended Use
|
| 148 |
|
| 149 |
* Medium-scale pretraining
|
|
@@ -151,8 +138,6 @@ For license information:
|
|
| 151 |
* Research in mathematical reasoning
|
| 152 |
* LaTeX modeling and generation
|
| 153 |
|
| 154 |
-
---
|
| 155 |
-
|
| 156 |
## 🚫 Limitations
|
| 157 |
|
| 158 |
* LaTeX sources may contain compilation errors.
|
|
@@ -160,8 +145,6 @@ For license information:
|
|
| 160 |
* Licensing varies per paper.
|
| 161 |
* This dataset is a subset of the full corpus.
|
| 162 |
|
| 163 |
-
---
|
| 164 |
-
|
| 165 |
## 📌 Citation
|
| 166 |
|
| 167 |
If you use this dataset, please cite:
|
|
|
|
| 18 |
Medium-scale LaTeX corpus from arXiv (math, CS, physics, statistics)
|
| 19 |
</p>
|
| 20 |
|
| 21 |
+
📄 Paper: https://arxiv.org/abs/2602.17288
|
| 22 |
|
| 23 |
## 📚 Overview
|
| 24 |
|
|
|
|
| 41 |
* Ablation studies
|
| 42 |
* Resource-constrained training setups
|
| 43 |
|
|
|
|
| 44 |
|
| 45 |
## 🔗 Full Dataset
|
| 46 |
|
|
|
|
| 50 |
|
| 51 |
The 80GB version contains the full dataset without size reduction.
|
| 52 |
|
|
|
|
| 53 |
|
| 54 |
## 📦 Files
|
| 55 |
|
|
|
|
| 56 |
train.jsonl
|
| 57 |
val.jsonl
|
| 58 |
checkpoint.txt
|
|
|
|
| 71 |
Internal processing checkpoint file used during dataset creation.
|
| 72 |
Not required for training.
|
| 73 |
|
|
|
|
|
|
|
| 74 |
## 🧾 Data Format
|
| 75 |
|
| 76 |
Each line in `train.jsonl` and `val.jsonl` is a JSON object.
|
|
|
|
| 99 |
* `stat.TH`
|
| 100 |
* `latex` — extracted LaTeX source content
|
| 101 |
|
|
|
|
|
|
|
| 102 |
## 📊 Dataset Characteristics
|
| 103 |
|
| 104 |
* Size: ~15GB
|
|
|
|
| 109 |
|
| 110 |
This version is a reduced subset of the full dataset.
|
| 111 |
|
|
|
|
|
|
|
| 112 |
## 🧹 Processing Pipeline
|
| 113 |
|
| 114 |
The dataset was constructed by:
|
|
|
|
| 120 |
5. Creating train/validation splits.
|
| 121 |
6. Subsampling to create the 15GB medium version.
|
| 122 |
|
|
|
|
|
|
|
| 123 |
## ⚖️ Licensing
|
| 124 |
|
| 125 |
This dataset contains LaTeX source files from papers hosted on arXiv.
|
|
|
|
| 131 |
For license information:
|
| 132 |
[https://arxiv.org/help/license](https://arxiv.org/help/license)
|
| 133 |
|
|
|
|
|
|
|
| 134 |
## 🎯 Intended Use
|
| 135 |
|
| 136 |
* Medium-scale pretraining
|
|
|
|
| 138 |
* Research in mathematical reasoning
|
| 139 |
* LaTeX modeling and generation
|
| 140 |
|
|
|
|
|
|
|
| 141 |
## 🚫 Limitations
|
| 142 |
|
| 143 |
* LaTeX sources may contain compilation errors.
|
|
|
|
| 145 |
* Licensing varies per paper.
|
| 146 |
* This dataset is a subset of the full corpus.
|
| 147 |
|
|
|
|
|
|
|
| 148 |
## 📌 Citation
|
| 149 |
|
| 150 |
If you use this dataset, please cite:
|