dipeshpandit/distilbert-toxic-comment

checkpoints-toxic

This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0510
• Macro F1: 0.6969
• Micro F1: 0.7858
• Precision: 0.6672
• Recall: 0.7298
• Hamming Loss: 0.0163
• Roc Auc Macro: 0.9885
• Roc Auc Micro: 0.9900

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: 3.2719182895564304e-05
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Macro F1	Micro F1	Precision	Recall	Hamming Loss	Roc Auc Macro	Roc Auc Micro
0.0402	1.0	2244	0.0387	0.6473	0.7850	0.6806	0.6427	0.0160	0.9901	0.9920
0.0336	2.0	4488	0.0385	0.6798	0.7912	0.6727	0.6974	0.0161	0.9898	0.9924
0.0259	3.0	6732	0.0414	0.6896	0.7872	0.6520	0.7360	0.0165	0.9900	0.9918
0.0199	4.0	8976	0.0463	0.6761	0.7852	0.6525	0.7041	0.0165	0.9887	0.9905
0.0151	5.0	11220	0.0510	0.6969	0.7858	0.6672	0.7298	0.0163	0.9885	0.9900

Model Evaluation

Per Label Threshold

Label ID	Label Name	Optimal Threshold	F1 Score (Validation)
0	toxic	0.69	0.8312
1	severe_toxic	0.44	0.5564
2	obscene	0.60	0.8383
3	threat	0.47	0.6275
4	insult	0.32	0.7636
5	identity_hate	0.55	0.6179

Model Performance on Test Set

Metric	Value
F1 Score (Micro)	0.6668
F1 Score (Macro)	0.6083
Precision (Micro)	0.5642
Recall (Micro)	0.8152
Precision (Macro)	0.5264
Recall (Macro)	0.7475

Framework versions

• Transformers 4.56.0
• Pytorch 2.8.0+cu129
• Datasets 4.4.1
• Tokenizers 0.22.0