distilbert-base-uncased_fine_tuned
This model is a fine-tuned version of distilbert-base-uncased on an reddit dataset -for NSFW classification. It was trained on titles + body_text of submissions. It achieves the following results on the evaluation set:
- Loss: 1.0159
- Accuracy: {'accuracy': 0.9095537914043252}
- Recall: {'recall': 0.8936873290793071}
- Precision: {'precision': 0.916024293389395}
- F1: {'f1': 0.9047179605490829}
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: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 15
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Recall | Precision | F1 |
|---|---|---|---|---|---|---|---|
| 0.256 | 1.0 | 2284 | 0.2569 | {'accuracy': 0.9085683000273748} | {'recall': 0.8976754785779398} | {'precision': 0.9107514450867052} | {'f1': 0.9041661884540342} |
| 0.1948 | 2.0 | 4568 | 0.2471 | {'accuracy': 0.9138242540377771} | {'recall': 0.8644029170464904} | {'precision': 0.9518193224592221} | {'f1': 0.9060074047533739} |
| 0.1318 | 3.0 | 6852 | 0.3057 | {'accuracy': 0.914207500684369} | {'recall': 0.8977894257064722} | {'precision': 0.9216282606152767} | {'f1': 0.9095526695526697} |
| 0.0865 | 4.0 | 9136 | 0.4174 | {'accuracy': 0.9047358335614564} | {'recall': 0.8697584320875114} | {'precision': 0.9274605103280681} | {'f1': 0.8976831706456546} |
| 0.0545 | 5.0 | 11420 | 0.4635 | {'accuracy': 0.9095537914043252} | {'recall': 0.8849134001823155} | {'precision': 0.9236441484300666} | {'f1': 0.9038640595903165} |
| 0.0359 | 6.0 | 13704 | 0.5654 | {'accuracy': 0.9071448124828908} | {'recall': 0.8919781221513218} | {'precision': 0.9127798507462687} | {'f1': 0.9022591055786076} |
| 0.0262 | 7.0 | 15988 | 0.5568 | {'accuracy': 0.8994251300301123} | {'recall': 0.900865998176846} | {'precision': 0.8910176941282543} | {'f1': 0.8959147827072356} |
| 0.0181 | 8.0 | 18272 | 0.6846 | {'accuracy': 0.9042430878729811} | {'recall': 0.9026891522333638} | {'precision': 0.898491550413973} | {'f1': 0.9005854601261866} |
| 0.0121 | 9.0 | 20556 | 0.7516 | {'accuracy': 0.9071448124828908} | {'recall': 0.8990428441203282} | {'precision': 0.906896551724138} | {'f1': 0.9029526207370108} |
| 0.0119 | 10.0 | 22840 | 0.8614 | {'accuracy': 0.9050095811661648} | {'recall': 0.9002962625341842} | {'precision': 0.9018376897614427} | {'f1': 0.9010663169299197} |
| 0.0105 | 11.0 | 25124 | 0.7298 | {'accuracy': 0.9105940323022174} | {'recall': 0.8907247037374658} | {'precision': 0.9206218348839948} | {'f1': 0.9054265361672554} |
| 0.0049 | 12.0 | 27408 | 0.9237 | {'accuracy': 0.9101560361346839} | {'recall': 0.8828623518687329} | {'precision': 0.9266834110752302} | {'f1': 0.9042422827799498} |
| 0.0026 | 13.0 | 29692 | 0.9489 | {'accuracy': 0.9066520667944156} | {'recall': 0.8988149498632635} | {'precision': 0.9061458931648478} | {'f1': 0.9024655340083519} |
| 0.0016 | 14.0 | 31976 | 1.0045 | {'accuracy': 0.9099917875718587} | {'recall': 0.8963081130355515} | {'precision': 0.9146511627906977} | {'f1': 0.9053867403314917} |
| 0.0022 | 15.0 | 34260 | 1.0159 | {'accuracy': 0.9095537914043252} | {'recall': 0.8936873290793071} | {'precision': 0.916024293389395} | {'f1': 0.9047179605490829} |
Framework versions
- Transformers 4.21.0
- Pytorch 1.12.0+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1
- Downloads last month
- 12