layoutlmv2-large-uncased-finetuned-vi-infovqa
This model is a fine-tuned version of microsoft/layoutlmv2-large-uncased on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 8.5806
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: 2e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 250500
- 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 |
---|---|---|---|
No log | 0.17 | 100 | 4.6181 |
No log | 0.33 | 200 | 4.3357 |
No log | 0.5 | 300 | 4.3897 |
No log | 0.66 | 400 | 4.8238 |
4.4277 | 0.83 | 500 | 3.9088 |
4.4277 | 0.99 | 600 | 3.6063 |
4.4277 | 1.16 | 700 | 3.4278 |
4.4277 | 1.32 | 800 | 3.5428 |
4.4277 | 1.49 | 900 | 3.4331 |
3.0413 | 1.65 | 1000 | 3.3699 |
3.0413 | 1.82 | 1100 | 3.3622 |
3.0413 | 1.98 | 1200 | 3.5294 |
3.0413 | 2.15 | 1300 | 3.7918 |
3.0413 | 2.31 | 1400 | 3.4007 |
2.0843 | 2.48 | 1500 | 4.0296 |
2.0843 | 2.64 | 1600 | 4.1852 |
2.0843 | 2.81 | 1700 | 3.6690 |
2.0843 | 2.97 | 1800 | 3.6089 |
2.0843 | 3.14 | 1900 | 5.5534 |
1.7527 | 3.3 | 2000 | 4.7498 |
1.7527 | 3.47 | 2100 | 5.2691 |
1.7527 | 3.63 | 2200 | 5.1324 |
1.7527 | 3.8 | 2300 | 4.5912 |
1.7527 | 3.96 | 2400 | 4.1727 |
1.2037 | 4.13 | 2500 | 6.1174 |
1.2037 | 4.29 | 2600 | 5.7172 |
1.2037 | 4.46 | 2700 | 5.8843 |
1.2037 | 4.62 | 2800 | 6.4232 |
1.2037 | 4.79 | 2900 | 7.4486 |
0.8386 | 4.95 | 3000 | 7.1946 |
0.8386 | 5.12 | 3100 | 7.9869 |
0.8386 | 5.28 | 3200 | 8.0310 |
0.8386 | 5.45 | 3300 | 8.2954 |
0.8386 | 5.61 | 3400 | 8.5361 |
0.4389 | 5.78 | 3500 | 8.6040 |
0.4389 | 5.94 | 3600 | 8.5806 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.8.0+cu101
- Datasets 1.17.0
- Tokenizers 0.10.3
- Downloads last month
- 25
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.