tinyllama-lora-malicious-classifier

This model is a fine-tuned version of TinyLlama/TinyLlama_v1.1 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4833
• Accuracy: 0.8289
• Precision Weighted: 0.8220
• Recall Weighted: 0.8289
• F1 Weighted: 0.8239
• Mcc: 0.7203
• Balanced Accuracy: 0.7724
• Macro Fnr: 0.2276
• Macro Fpr: 0.0897
• Macro Specificity: 0.9103
• Per Class: {'jailbreaking': {'TP': 228, 'FP': 103, 'FN': 161, 'TN': 1437, 'FNR': 0.4138817480719794, 'FPR': 0.06688311688311688, 'Specificity': 0.9331168831168831}, 'prompt injection': {'TP': 439, 'FP': 112, 'FN': 131, 'TN': 1247, 'FNR': 0.22982456140350876, 'FPR': 0.08241353936718175, 'Specificity': 0.9175864606328182}, 'unharmful': {'TP': 932, 'FP': 115, 'FN': 38, 'TN': 844, 'FNR': 0.03917525773195876, 'FPR': 0.11991657977059438, 'Specificity': 0.8800834202294057}}

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
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision Weighted	Recall Weighted	F1 Weighted	Mcc	Balanced Accuracy	Macro Fnr	Macro Fpr	Macro Specificity	Per Class
1.0323	1.0	1107	1.0327	0.5702	0.5595	0.5702	0.5636	0.2924	0.5112	0.4888	0.2364	0.7636	{'jailbreaking': {'TP': 113, 'FP': 209, 'FN': 276, 'TN': 1331, 'FNR': 0.7095115681233933, 'FPR': 0.1357142857142857, 'Specificity': 0.8642857142857143}, 'prompt injection': {'TP': 312, 'FP': 238, 'FN': 258, 'TN': 1121, 'FNR': 0.45263157894736844, 'FPR': 0.1751287711552612, 'Specificity': 0.8248712288447387}, 'unharmful': {'TP': 675, 'FP': 382, 'FN': 295, 'TN': 577, 'FNR': 0.30412371134020616, 'FPR': 0.3983315954118874, 'Specificity': 0.6016684045881127}}
0.7496	2.0	2214	0.7073	0.7252	0.7112	0.7252	0.7144	0.5463	0.6529	0.3471	0.1502	0.8498	{'jailbreaking': {'TP': 154, 'FP': 132, 'FN': 235, 'TN': 1408, 'FNR': 0.6041131105398457, 'FPR': 0.08571428571428572, 'Specificity': 0.9142857142857143}, 'prompt injection': {'TP': 386, 'FP': 163, 'FN': 184, 'TN': 1196, 'FNR': 0.32280701754385965, 'FPR': 0.1199411331861663, 'Specificity': 0.8800588668138337}, 'unharmful': {'TP': 859, 'FP': 235, 'FN': 111, 'TN': 724, 'FNR': 0.11443298969072165, 'FPR': 0.24504692387904067, 'Specificity': 0.7549530761209593}}
0.5695	3.0	3321	0.6096	0.7766	0.7636	0.7766	0.7654	0.6323	0.7031	0.2969	0.1215	0.8785	{'jailbreaking': {'TP': 171, 'FP': 106, 'FN': 218, 'TN': 1434, 'FNR': 0.5604113110539846, 'FPR': 0.06883116883116883, 'Specificity': 0.9311688311688312}, 'prompt injection': {'TP': 417, 'FP': 141, 'FN': 153, 'TN': 1218, 'FNR': 0.26842105263157895, 'FPR': 0.10375275938189846, 'Specificity': 0.8962472406181016}, 'unharmful': {'TP': 910, 'FP': 184, 'FN': 60, 'TN': 775, 'FNR': 0.061855670103092786, 'FPR': 0.19186652763295098, 'Specificity': 0.808133472367049}}
0.5059	4.0	4428	0.5686	0.7926	0.7817	0.7926	0.7843	0.6596	0.7245	0.2755	0.1107	0.8893	{'jailbreaking': {'TP': 191, 'FP': 114, 'FN': 198, 'TN': 1426, 'FNR': 0.5089974293059126, 'FPR': 0.07402597402597402, 'Specificity': 0.925974025974026}, 'prompt injection': {'TP': 419, 'FP': 131, 'FN': 151, 'TN': 1228, 'FNR': 0.2649122807017544, 'FPR': 0.0963944076526858, 'Specificity': 0.9036055923473142}, 'unharmful': {'TP': 919, 'FP': 155, 'FN': 51, 'TN': 804, 'FNR': 0.05257731958762887, 'FPR': 0.1616266944734098, 'Specificity': 0.8383733055265902}}
0.4583	5.0	5535	0.5413	0.8056	0.7953	0.8056	0.7977	0.6812	0.7389	0.2611	0.1032	0.8968	{'jailbreaking': {'TP': 199, 'FP': 109, 'FN': 190, 'TN': 1431, 'FNR': 0.4884318766066838, 'FPR': 0.07077922077922078, 'Specificity': 0.9292207792207792}, 'prompt injection': {'TP': 426, 'FP': 126, 'FN': 144, 'TN': 1233, 'FNR': 0.25263157894736843, 'FPR': 0.09271523178807947, 'Specificity': 0.9072847682119205}, 'unharmful': {'TP': 929, 'FP': 140, 'FN': 41, 'TN': 819, 'FNR': 0.042268041237113405, 'FPR': 0.145985401459854, 'Specificity': 0.8540145985401459}}
0.4763	6.0	6642	0.5243	0.8113	0.8028	0.8113	0.8052	0.6910	0.7498	0.2502	0.0994	0.9006	{'jailbreaking': {'TP': 212, 'FP': 113, 'FN': 177, 'TN': 1427, 'FNR': 0.455012853470437, 'FPR': 0.07337662337662337, 'Specificity': 0.9266233766233766}, 'prompt injection': {'TP': 428, 'FP': 120, 'FN': 142, 'TN': 1239, 'FNR': 0.24912280701754386, 'FPR': 0.08830022075055188, 'Specificity': 0.9116997792494481}, 'unharmful': {'TP': 925, 'FP': 131, 'FN': 45, 'TN': 828, 'FNR': 0.04639175257731959, 'FPR': 0.13660062565172054, 'Specificity': 0.8633993743482794}}
0.4283	7.0	7749	0.5095	0.8170	0.8083	0.8170	0.8104	0.7003	0.7546	0.2454	0.0969	0.9031	{'jailbreaking': {'TP': 213, 'FP': 105, 'FN': 176, 'TN': 1435, 'FNR': 0.4524421593830334, 'FPR': 0.06818181818181818, 'Specificity': 0.9318181818181818}, 'prompt injection': {'TP': 430, 'FP': 118, 'FN': 140, 'TN': 1241, 'FNR': 0.24561403508771928, 'FPR': 0.08682855040470934, 'Specificity': 0.9131714495952906}, 'unharmful': {'TP': 933, 'FP': 130, 'FN': 37, 'TN': 829, 'FNR': 0.03814432989690722, 'FPR': 0.13555787278415016, 'Specificity': 0.8644421272158499}}
0.4119	8.0	8856	0.5033	0.8191	0.8116	0.8191	0.8135	0.7041	0.7592	0.2408	0.0951	0.9049	{'jailbreaking': {'TP': 223, 'FP': 118, 'FN': 166, 'TN': 1422, 'FNR': 0.4267352185089974, 'FPR': 0.07662337662337662, 'Specificity': 0.9233766233766234}, 'prompt injection': {'TP': 422, 'FP': 105, 'FN': 148, 'TN': 1254, 'FNR': 0.2596491228070175, 'FPR': 0.0772626931567329, 'Specificity': 0.9227373068432672}, 'unharmful': {'TP': 935, 'FP': 126, 'FN': 35, 'TN': 833, 'FNR': 0.03608247422680412, 'FPR': 0.13138686131386862, 'Specificity': 0.8686131386861314}}
0.412	9.0	9963	0.4955	0.8253	0.8178	0.8253	0.8199	0.7143	0.7671	0.2329	0.0916	0.9084	{'jailbreaking': {'TP': 222, 'FP': 103, 'FN': 167, 'TN': 1437, 'FNR': 0.42930591259640105, 'FPR': 0.06688311688311688, 'Specificity': 0.9331168831168831}, 'prompt injection': {'TP': 440, 'FP': 118, 'FN': 130, 'TN': 1241, 'FNR': 0.22807017543859648, 'FPR': 0.08682855040470934, 'Specificity': 0.9131714495952906}, 'unharmful': {'TP': 930, 'FP': 116, 'FN': 40, 'TN': 843, 'FNR': 0.041237113402061855, 'FPR': 0.12095933263816476, 'Specificity': 0.8790406673618353}}
0.496	10.0	11070	0.4926	0.8289	0.8214	0.8289	0.8232	0.7202	0.7703	0.2297	0.0903	0.9097	{'jailbreaking': {'TP': 224, 'FP': 98, 'FN': 165, 'TN': 1442, 'FNR': 0.4241645244215938, 'FPR': 0.06363636363636363, 'Specificity': 0.9363636363636364}, 'prompt injection': {'TP': 439, 'FP': 113, 'FN': 131, 'TN': 1246, 'FNR': 0.22982456140350876, 'FPR': 0.08314937454010302, 'Specificity': 0.9168506254598969}, 'unharmful': {'TP': 936, 'FP': 119, 'FN': 34, 'TN': 840, 'FNR': 0.03505154639175258, 'FPR': 0.12408759124087591, 'Specificity': 0.8759124087591241}}
0.428	11.0	12177	0.4890	0.8258	0.8191	0.8258	0.8207	0.7154	0.7677	0.2323	0.0913	0.9087	{'jailbreaking': {'TP': 230, 'FP': 117, 'FN': 159, 'TN': 1423, 'FNR': 0.4087403598971722, 'FPR': 0.07597402597402597, 'Specificity': 0.924025974025974}, 'prompt injection': {'TP': 424, 'FP': 99, 'FN': 146, 'TN': 1260, 'FNR': 0.256140350877193, 'FPR': 0.0728476821192053, 'Specificity': 0.9271523178807947}, 'unharmful': {'TP': 939, 'FP': 120, 'FN': 31, 'TN': 839, 'FNR': 0.031958762886597936, 'FPR': 0.1251303441084463, 'Specificity': 0.8748696558915537}}
0.4103	12.0	13284	0.4866	0.8269	0.8191	0.8269	0.8206	0.7166	0.7669	0.2331	0.0922	0.9078	{'jailbreaking': {'TP': 221, 'FP': 99, 'FN': 168, 'TN': 1441, 'FNR': 0.4318766066838046, 'FPR': 0.06428571428571428, 'Specificity': 0.9357142857142857}, 'prompt injection': {'TP': 437, 'FP': 107, 'FN': 133, 'TN': 1252, 'FNR': 0.23333333333333334, 'FPR': 0.07873436350257543, 'Specificity': 0.9212656364974245}, 'unharmful': {'TP': 937, 'FP': 128, 'FN': 33, 'TN': 831, 'FNR': 0.03402061855670103, 'FPR': 0.1334723670490094, 'Specificity': 0.8665276329509907}}
0.4009	13.0	14391	0.4833	0.8289	0.8220	0.8289	0.8239	0.7203	0.7724	0.2276	0.0897	0.9103	{'jailbreaking': {'TP': 228, 'FP': 103, 'FN': 161, 'TN': 1437, 'FNR': 0.4138817480719794, 'FPR': 0.06688311688311688, 'Specificity': 0.9331168831168831}, 'prompt injection': {'TP': 439, 'FP': 112, 'FN': 131, 'TN': 1247, 'FNR': 0.22982456140350876, 'FPR': 0.08241353936718175, 'Specificity': 0.9175864606328182}, 'unharmful': {'TP': 932, 'FP': 115, 'FN': 38, 'TN': 844, 'FNR': 0.03917525773195876, 'FPR': 0.11991657977059438, 'Specificity': 0.8800834202294057}}
0.4242	14.0	15498	0.4834	0.8284	0.8211	0.8284	0.8228	0.7193	0.7700	0.2300	0.0906	0.9094	{'jailbreaking': {'TP': 226, 'FP': 105, 'FN': 163, 'TN': 1435, 'FNR': 0.4190231362467866, 'FPR': 0.06818181818181818, 'Specificity': 0.9318181818181818}, 'prompt injection': {'TP': 435, 'FP': 104, 'FN': 135, 'TN': 1255, 'FNR': 0.23684210526315788, 'FPR': 0.07652685798381163, 'Specificity': 0.9234731420161884}, 'unharmful': {'TP': 937, 'FP': 122, 'FN': 33, 'TN': 837, 'FNR': 0.03402061855670103, 'FPR': 0.12721584984358708, 'Specificity': 0.872784150156413}}
0.3859	15.0	16605	0.4829	0.8269	0.8197	0.8269	0.8215	0.7168	0.7692	0.2308	0.0912	0.9088	{'jailbreaking': {'TP': 226, 'FP': 106, 'FN': 163, 'TN': 1434, 'FNR': 0.4190231362467866, 'FPR': 0.06883116883116883, 'Specificity': 0.9311688311688312}, 'prompt injection': {'TP': 436, 'FP': 107, 'FN': 134, 'TN': 1252, 'FNR': 0.23508771929824562, 'FPR': 0.07873436350257543, 'Specificity': 0.9212656364974245}, 'unharmful': {'TP': 933, 'FP': 121, 'FN': 37, 'TN': 838, 'FNR': 0.03814432989690722, 'FPR': 0.1261730969760167, 'Specificity': 0.8738269030239834}}

Framework versions

• PEFT 0.17.1
• Transformers 4.53.3
• Pytorch 2.6.0+cu124
• Datasets 4.3.0
• Tokenizers 0.21.4