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

Loss: 0.3202
Accuracy: 0.8617
Unsafe Precision: 0.8690
Unsafe Recall: 0.8848
Unsafe F1: 0.8769
Unsafe Fpr: 0.1673
Unsafe Aucpr: 0.9555
Safe Precision: 0.8521
Safe Recall: 0.8327
Safe F1: 0.8423
Safe Fpr: 0.1152
Safe Aucpr: 0.9212

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: 128
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
0.3088	0.2501	2114	0.3633	0.8474	0.8856	0.8334	0.8587	0.1351	0.9413	0.8054	0.8649	0.8341	0.1666	0.8880
0.3276	0.5001	4228	0.3342	0.8548	0.8646	0.8764	0.8705	0.1722	0.9483	0.8422	0.8278	0.8349	0.1236	0.9065
0.3059	0.7502	6342	0.3256	0.8578	0.8966	0.8416	0.8682	0.1218	0.9514	0.8155	0.8782	0.8457	0.1584	0.9119
0.3413	1.0002	8456	0.3251	0.8586	0.8651	0.8837	0.8743	0.1729	0.9525	0.8501	0.8271	0.8384	0.1163	0.9136
0.3035	1.2503	10570	0.3247	0.8611	0.8775	0.8721	0.8748	0.1528	0.9529	0.8408	0.8472	0.8440	0.1279	0.9160
0.2828	1.5004	12684	0.3307	0.8614	0.8721	0.8800	0.8760	0.1620	0.9539	0.8477	0.8380	0.8428	0.1200	0.9178
0.2864	1.7504	14798	0.3233	0.8623	0.8758	0.8769	0.8763	0.1560	0.9548	0.8453	0.8440	0.8446	0.1231	0.9185
0.3083	2.0005	16912	0.3236	0.8627	0.8966	0.8516	0.8735	0.1232	0.9551	0.8248	0.8768	0.8500	0.1484	0.9140
0.3109	2.2505	19026	0.3227	0.8596	0.8737	0.8742	0.8739	0.1586	0.9539	0.8420	0.8414	0.8417	0.1258	0.9199
0.2648	2.5006	21140	0.3256	0.8614	0.8694	0.8838	0.8765	0.1666	0.9547	0.8511	0.8334	0.8422	0.1162	0.9196
0.2712	2.7507	23254	0.3202	0.8617	0.8690	0.8848	0.8769	0.1673	0.9555	0.8521	0.8327	0.8423	0.1152	0.9212
0.2809	3.0007	25368	0.3266	0.8614	0.8803	0.8691	0.8746	0.1483	0.9555	0.8383	0.8517	0.8450	0.1309	0.9220
0.2753	3.2508	27482	0.3326	0.8600	0.8818	0.8643	0.8729	0.1454	0.9546	0.8339	0.8546	0.8441	0.1357	0.9181
0.2408	3.5008	29596	0.3384	0.8596	0.8812	0.8642	0.8726	0.1462	0.9544	0.8337	0.8538	0.8436	0.1358	0.9177
0.2915	3.7509	31710	0.3398	0.8589	0.8959	0.8446	0.8695	0.1232	0.9548	0.8181	0.8768	0.8465	0.1554	0.9186

Framework versions

Transformers 4.57.3
Pytorch 2.7.1+cu118
Datasets 4.4.1
Tokenizers 0.22.1

saiteki-kai/QA-DeBERTa-v3-large-qa_bi_cross_attn_cls-binary

QA-DeBERTa-v3-large-qa_bi_cross_attn_cls-binary

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

saiteki-kai/
QA-DeBERTa-v3-large-qa_bi_cross_attn_cls-binary