Modelpublic

leobianco/bosch_RM_google_S130104_LLM_false_STRUCT_false_epo15_lr6.2e-04_r8_2608191212

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face

0likes50downloads

Model Card

boschRMgoogleS130104LLMfalseSTRUCTfalseepo15lr6.2e-04r8_2608191212

This model is a fine-tuned version of google/gemma-4-E4B-it on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 1.9397
—Roc Auc: 0.8799
—Best Threshold: 0.9997
—Tpr At Best Threshold: 0.8028
—Fpr At Best Threshold: 0.1818
—Accuracy At Best Threshold: 0.8065
—Avg Score True Positives: 0.8987
—Avg Score True Negatives: 0.3415

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: 0.0006174293063066378
—trainbatchsize: 16
—evalbatchsize: 32
—seed: 130104
—distributed_type: multi-GPU
—num_devices: 2
—totaltrainbatch_size: 32
—totalevalbatch_size: 64
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_steps: 0.1
—num_epochs: 15.0

Training results

Training Loss	Epoch	Step	Validation Loss	Roc Auc	Best Threshold	Tpr At Best Threshold	Fpr At Best Threshold	Accuracy At Best Threshold	Avg Score True Positives	Avg Score True Negatives
No log	0	0	1.2592	0.4586	0.5026	0.9296	0.8409	0.7473	0.5864	0.5913
0.4487	0.5882	20	0.9393	0.7517	0.6516	0.9789	0.5455	0.8548	0.8450	0.7068
0.6523	1.1765	40	0.8086	0.8510	0.7592	0.8451	0.2273	0.8280	0.8708	0.4310
1.8125	1.7647	60	1.2938	0.8765	0.9923	0.7535	0.1136	0.7849	0.9834	0.7823
0.1971	2.3529	80	1.1667	0.8339	0.5687	0.8169	0.25	0.8011	0.7598	0.3157
1.0371	2.9412	100	0.9195	0.8094	0.7896	0.8099	0.2727	0.7903	0.8775	0.5184
0.6328	3.5294	120	0.9378	0.8971	0.9775	0.8310	0.1364	0.8387	0.9533	0.5406
0.1332	4.1176	140	0.9581	0.8857	0.7680	0.8310	0.2045	0.8226	0.8508	0.2809
0.0580	4.7059	160	1.3505	0.8752	0.3698	0.8310	0.1818	0.8280	0.7819	0.2023
0.0025	5.2941	180	2.1579	0.8796	0.9866	0.9648	0.3409	0.8925	0.9941	0.7516
0.2951	5.8824	200	0.9745	0.8783	0.9627	0.7535	0.1364	0.7796	0.8705	0.3184
0.2604	6.4706	220	1.6551	0.8939	0.8193	0.8028	0.1364	0.8172	0.8181	0.1970
0.0080	7.0588	240	1.8263	0.8525	0.9585	0.7606	0.1818	0.7742	0.8370	0.2950
0.0196	7.6471	260	1.8269	0.8600	0.9979	0.7887	0.2045	0.7903	0.9214	0.4739
0.0439	8.2353	280	2.1334	0.8856	0.9970	0.9014	0.2045	0.8763	0.9296	0.3125
1.0725	8.8235	300	1.8254	0.8812	0.9934	0.8169	0.1818	0.8172	0.8934	0.3415
0.0004	9.4118	320	1.7214	0.8600	0.9944	0.8169	0.2273	0.8065	0.9004	0.4122
0.0000	10.0	340	1.9074	0.8671	0.9954	0.8592	0.2273	0.8387	0.9302	0.4206
0.0000	10.5882	360	1.9057	0.8768	0.9696	0.8521	0.2273	0.8333	0.8926	0.3417
0.0000	11.1765	380	1.9765	0.8827	0.9953	0.8028	0.1818	0.8065	0.8625	0.2983
0.0001	11.7647	400	1.9267	0.8837	0.9989	0.8028	0.1818	0.8065	0.8820	0.3254
0.0000	12.3529	420	1.9024	0.8834	0.9995	0.8028	0.1818	0.8065	0.8923	0.3368
0.0092	12.9412	440	1.9313	0.8800	0.9995	0.8028	0.1818	0.8065	0.8915	0.3337
0.0013	13.5294	460	1.9567	0.8786	0.9997	0.8028	0.1818	0.8065	0.8991	0.3441
0.0000	14.1176	480	1.9513	0.8800	0.9995	0.8028	0.1818	0.8065	0.8981	0.3403
0.0001	14.7059	500	1.9504	0.8795	0.9997	0.8028	0.1818	0.8065	0.8981	0.3409
0.0000	15.0	510	1.9397	0.8799	0.9997	0.8028	0.1818	0.8065	0.8987	0.3415

Framework versions

—PEFT 0.20.0
—Transformers 5.14.1
—Pytorch 2.11.0+cu130
—Datasets 5.0.1
—Tokenizers 0.22.2