Modelpublic

safepaca/absolute-harmfulness-predictor-redteam

sourceHugging Facemitupdated 3y agoView on Hugging Face

1likes35downloads

Model Card

safetune

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

—Loss: 1.1197
—Mse: 1.1197

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
—trainbatchsize: 8
—evalbatchsize: 4
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: constant
—num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Mse
2.2203	0.05	50	1.8289	1.8289
1.8997	0.1	100	1.7516	1.7516
1.4082	0.15	150	1.3950	1.3950
1.5899	0.2	200	1.9590	1.9590
1.3633	0.25	250	1.3316	1.3316
1.3758	0.29	300	1.2860	1.2860
1.3339	0.34	350	1.2694	1.2694
1.2831	0.39	400	1.3048	1.3048
1.2928	0.44	450	1.2395	1.2395
1.2506	0.49	500	1.4315	1.4315
1.204	0.54	550	1.1596	1.1596
1.1749	0.59	600	1.1995	1.1995
1.134	0.64	650	1.3782	1.3782
1.3097	0.69	700	1.1867	1.1867
1.29	0.74	750	1.2024	1.2024
1.1575	0.78	800	1.1197	1.1197
1.2148	0.83	850	1.1944	1.1944
1.1597	0.88	900	1.2023	1.2023
1.1422	0.93	950	1.1546	1.1546
1.0734	0.98	1000	1.2593	1.2593

Framework versions

—Transformers 4.30.1
—Pytorch 2.0.1+cu117
—Datasets 2.12.0
—Tokenizers 0.13.3