Views
No views yet
1 title = "{SQ}u{AD}: 100,000+ Questions for Machine Comprehension of Text",
2 author = "Rajpurkar, Pranav and
3 Zhang, Jian and
4 Lopyrev, Konstantin and
5 Liang, Percy",
6 editor = "Su, Jian and
7 Duh, Kevin and
8 Carreras, Xavier",
9 booktitle = "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
10 month = nov,
11 year = "2016",
12 address = "Austin, Texas",
13 publisher = "Association for Computational Linguistics",
14 url = "https://aclanthology.org/D16-1264",
15 doi = "10.18653/v1/D16-1264",
16 pages = "2383--2392",
17 eprint={1606.05250},
18 archivePrefix={arXiv},
19 primaryClass={cs.CL},
20}
21
22
23
24## Training and evaluation data
25
26Trained on [squad](https://huggingface.co/datasets/rajpurkar/squad) (train).
27Evaluated on its validation split.
28
29## Training procedure
30
31### Training hyperparameters
32
33The following hyperparameters were used during training:
34- learning_rate: 2e-05
35- train_batch_size: 16
36- eval_batch_size: 16
37- seed: 42
38- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
39- lr_scheduler_type: linear
40- num_epochs: 3
41
42### Training results
43
44| Training Loss | Epoch | Step | Validation Loss |
45|:-------------:|:-----:|:-----:|:---------------:|
46| 1.1995 | 1.0 | 5475 | 1.1494 |
47| 0.9689 | 2.0 | 10950 | 1.0921 |
48| 0.7334 | 3.0 | 16425 | 1.1407 |
49
50
51### Framework versions
52
53- Transformers 5.0.0.dev0
54- Pytorch 2.8.0+cu126
55- Datasets 4.0.0
56- Tokenizers 0.22.1