QA-mDeBERTa-v3-large-binary

This model is a fine-tuned version of microsoft/mdeberta-v3-base on the saiteki-kai/Beavertails-it dataset. It achieves the following results on the evaluation set:

• Loss: 0.3241
• Accuracy: 0.8609
• Unsafe Precision: 0.8855
• Unsafe Recall: 0.8615
• Unsafe F1: 0.8733
• Unsafe Fpr: 0.1398
• Unsafe Aucpr: 0.9529
• Safe Precision: 0.8320
• Safe Recall: 0.8602
• Safe F1: 0.8458
• Safe Fpr: 0.1385
• Safe Aucpr: 0.9143

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: 6e-06
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Unsafe Precision	Unsafe Recall	Unsafe F1	Unsafe Fpr	Unsafe Aucpr	Safe Precision	Safe Recall	Safe F1	Safe Fpr	Safe Aucpr
0.3283	0.2501	2114	0.4026	0.8181	0.9079	0.7491	0.8209	0.0954	0.9303	0.7418	0.9046	0.8152	0.2509	0.8630
0.353	0.5001	4228	0.3597	0.8432	0.8658	0.8499	0.8578	0.1653	0.9402	0.8159	0.8347	0.8252	0.1501	0.8882
0.3248	0.7502	6342	0.3476	0.8484	0.8928	0.8268	0.8585	0.1245	0.9450	0.8011	0.8755	0.8367	0.1732	0.8972
0.3591	1.0002	8456	0.3414	0.8512	0.8822	0.8455	0.8635	0.1416	0.9468	0.8158	0.8584	0.8365	0.1545	0.9011
0.3167	1.2503	10570	0.3428	0.8537	0.8837	0.8488	0.8659	0.1402	0.9482	0.8192	0.8598	0.8390	0.1512	0.9046
0.2944	1.5004	12684	0.3422	0.8549	0.8864	0.8480	0.8668	0.1364	0.9491	0.8191	0.8636	0.8408	0.1520	0.9072
0.2875	1.7504	14798	0.3345	0.8577	0.8800	0.8617	0.8708	0.1474	0.9504	0.8309	0.8526	0.8416	0.1383	0.9101
0.346	2.0005	16912	0.3275	0.8583	0.8899	0.8507	0.8698	0.1321	0.9517	0.8225	0.8679	0.8446	0.1493	0.9122
0.3394	2.2505	19026	0.3280	0.8579	0.8770	0.8661	0.8715	0.1524	0.9516	0.8346	0.8476	0.8410	0.1339	0.9121
0.2935	2.5006	21140	0.3269	0.8589	0.8761	0.8693	0.8727	0.1543	0.9522	0.8376	0.8457	0.8417	0.1307	0.9133
0.2999	2.7507	23254	0.3241	0.8609	0.8855	0.8615	0.8733	0.1398	0.9529	0.8320	0.8602	0.8458	0.1385	0.9143
0.3016	3.0007	25368	0.3243	0.8608	0.8889	0.8570	0.8727	0.1344	0.9530	0.8283	0.8656	0.8465	0.1430	0.9153
0.3097	3.2508	27482	0.3271	0.8601	0.8832	0.8628	0.8729	0.1432	0.9525	0.8327	0.8568	0.8446	0.1372	0.9142
0.2794	3.5008	29596	0.3300	0.8604	0.8942	0.8496	0.8713	0.1261	0.9531	0.8224	0.8739	0.8474	0.1504	0.9156
0.3129	3.7509	31710	0.3282	0.8585	0.9033	0.8351	0.8679	0.1121	0.9532	0.8111	0.8879	0.8477	0.1649	0.9161

Framework versions

• Transformers 4.57.3
• Pytorch 2.7.1+cu118
• Datasets 4.4.1
• Tokenizers 0.22.1