saiteki-kai/QA-ModernBERT-large

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

QA-ModernBERT-large

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

• —Loss: 0.0826
• —Accuracy: 0.6608
• —Macro F1: 0.6703
• —Macro Precision: 0.6572
• —Macro Recall: 0.6909
• —Micro F1: 0.7482
• —Micro Precision: 0.7288
• —Micro Recall: 0.7687
• —Flagged/accuracy: 0.8493
• —Flagged/precision: 0.8469
• —Flagged/recall: 0.8900
• —Flagged/f1: 0.8679
• —Flagged/aucpr: 0.8991
• —Flagged/fpr: 0.2018
• —Animal Abuse/accuracy: 0.9946
• —Animal Abuse/precision: 0.7483
• —Animal Abuse/recall: 0.7951
• —Animal Abuse/f1: 0.7710
• —Animal Abuse/fpr: 0.0031
• —Animal Abuse/threshold: 0.2736
• —Child Abuse/accuracy: 0.9968
• —Child Abuse/precision: 0.75
• —Child Abuse/recall: 0.6396
• —Child Abuse/f1: 0.6904
• —Child Abuse/fpr: 0.0012
• —Child Abuse/threshold: 0.3748
• —Controversial Topics,politics/accuracy: 0.9679
• —Controversial Topics,politics/precision: 0.4800
• —Controversial Topics,politics/recall: 0.5597
• —Controversial Topics,politics/f1: 0.5168
• —Controversial Topics,politics/fpr: 0.0192
• —Controversial Topics,politics/threshold: 0.2822
• —Discrimination,stereotype,injustice/accuracy: 0.9498
• —Discrimination,stereotype,injustice/precision: 0.6553
• —Discrimination,stereotype,injustice/recall: 0.7770
• —Discrimination,stereotype,injustice/f1: 0.7109
• —Discrimination,stereotype,injustice/fpr: 0.0353
• —Discrimination,stereotype,injustice/threshold: 0.1871
• —Drug Abuse,weapons,banned Substance/accuracy: 0.9728
• —Drug Abuse,weapons,banned Substance/precision: 0.7361
• —Drug Abuse,weapons,banned Substance/recall: 0.8048
• —Drug Abuse,weapons,banned Substance/f1: 0.7689
• —Drug Abuse,weapons,banned Substance/fpr: 0.0172
• —Drug Abuse,weapons,banned Substance/threshold: 0.3478
• —Financial Crime,property Crime,theft/accuracy: 0.9593
• —Financial Crime,property Crime,theft/precision: 0.7569
• —Financial Crime,property Crime,theft/recall: 0.8564
• —Financial Crime,property Crime,theft/f1: 0.8036
• —Financial Crime,property Crime,theft/fpr: 0.0297
• —Financial Crime,property Crime,theft/threshold: 0.3831
• —Hate Speech,offensive Language/accuracy: 0.9504
• —Hate Speech,offensive Language/precision: 0.7616
• —Hate Speech,offensive Language/recall: 0.6499
• —Hate Speech,offensive Language/f1: 0.7013
• —Hate Speech,offensive Language/fpr: 0.0200
• —Hate Speech,offensive Language/threshold: 0.3886
• —Misinformation Regarding Ethics,laws And Safety/accuracy: 0.9792
• —Misinformation Regarding Ethics,laws And Safety/precision: 0.2061
• —Misinformation Regarding Ethics,laws And Safety/recall: 0.2503
• —Misinformation Regarding Ethics,laws And Safety/f1: 0.2261
• —Misinformation Regarding Ethics,laws And Safety/fpr: 0.0119
• —Misinformation Regarding Ethics,laws And Safety/threshold: 0.1871
• —Non Violent Unethical Behavior/accuracy: 0.8783
• —Non Violent Unethical Behavior/precision: 0.6977
• —Non Violent Unethical Behavior/recall: 0.6839
• —Non Violent Unethical Behavior/f1: 0.6907
• —Non Violent Unethical Behavior/fpr: 0.0735
• —Non Violent Unethical Behavior/threshold: 0.3478
• —Privacy Violation/accuracy: 0.9805
• —Privacy Violation/precision: 0.7845
• —Privacy Violation/recall: 0.8345
• —Privacy Violation/f1: 0.8087
• —Privacy Violation/fpr: 0.0119
• —Privacy Violation/threshold: 0.3469
• —Self Harm/accuracy: 0.9966
• —Self Harm/precision: 0.8035
• —Self Harm/recall: 0.6683
• —Self Harm/f1: 0.7297
• —Self Harm/fpr: 0.0011
• —Self Harm/threshold: 0.4879
• —Sexually Explicit,adult Content/accuracy: 0.9834
• —Sexually Explicit,adult Content/precision: 0.6342
• —Sexually Explicit,adult Content/recall: 0.7381
• —Sexually Explicit,adult Content/f1: 0.6822
• —Sexually Explicit,adult Content/fpr: 0.0105
• —Sexually Explicit,adult Content/threshold: 0.3124
• —Terrorism,organized Crime/accuracy: 0.9888
• —Terrorism,organized Crime/precision: 0.3634
• —Terrorism,organized Crime/recall: 0.5364
• —Terrorism,organized Crime/f1: 0.4332
• —Terrorism,organized Crime/fpr: 0.0076
• —Terrorism,organized Crime/threshold: 0.1733
• —Violence,aiding And Abetting,incitement/accuracy: 0.9177
• —Violence,aiding And Abetting,incitement/precision: 0.8232
• —Violence,aiding And Abetting,incitement/recall: 0.8793
• —Violence,aiding And Abetting,incitement/f1: 0.8503
• —Violence,aiding And Abetting,incitement/fpr: 0.0684
• —Violence,aiding And Abetting,incitement/threshold: 0.4301

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: 3e-05
• —trainbatchsize: 32
• —evalbatchsize: 16
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: cosine
• —lrschedulerwarmup_ratio: 0.03
• —num_epochs: 10

Training results

Framework versions

• —Transformers 4.57.1
• —Pytorch 2.7.1+cu118
• —Datasets 4.4.1
• —Tokenizers 0.22.1