Downloads · 30 days
8
24% of all-time downloads
20split_dataset
Billwzl/20split_dataset
20split_dataset is a fill-mask model from Billwzl. Use it when you need the model to fill a missing word. It is set up for transformers. The card lists the license as apache-2.0.
should probably proofread and complete it, then remove this comment. --
All-time downloads
34
Public
Repo size
876 MB
Likes
1
Public
Repo makeup
Click a slice to open those files.
.bin438 MB · 100%
At a glance
- Task
- Fill-Mask
- Library
- transformers
- License
- apache-2.0
- Model type
- bert
- Access
- Public
- Created
- Jul 9, 2022
- Updated
- Jul 14, 2022
- SHA
- 5e544b6d
Try a prompt
Model card
From the Hugging Face model README
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
20split_dataset
This model is a fine-tuned version of bert-base-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 2.0446
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 64
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 6
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.5971 | 1.0 | 11851 | 2.3479 |
| 2.3773 | 2.0 | 23702 | 2.2446 |
| 2.2663 | 3.0 | 35553 | 2.1630 |
| 2.1842 | 4.0 | 47404 | 2.1059 |
| 2.1145 | 5.0 | 59255 | 2.0626 |
| 2.0652 | 6.0 | 71106 | 2.0446 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.12.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1
- Task
- Fill-Mask
- Library
- transformers
- Type
- bert
- License
- apache-2.0
- Created
- Jul 9, 2022
- Updated
- Jul 14, 2022