Modelpublic

avsolatorio/doc-topic-model_eval-04_train-00

sourceHugging Facemitupdated 2y agoView on Hugging Face

0likes7downloads

Model Card

doc-topic-modeleval-04train-00

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

—Loss: 0.0377
—Accuracy: 0.9880
—F1: 0.6385
—Precision: 0.7205
—Recall: 0.5733

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: 4
—evalbatchsize: 256
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—num_epochs: 100
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.0929	0.4931	1000	0.0906	0.9815	0.0	0.0	0.0
0.0785	0.9862	2000	0.0704	0.9815	0.0	0.0	0.0
0.0622	1.4793	3000	0.0572	0.9823	0.1022	0.8442	0.0544
0.0542	1.9724	4000	0.0499	0.9843	0.3390	0.7662	0.2176
0.048	2.4655	5000	0.0459	0.9853	0.4278	0.7709	0.2960
0.0436	2.9586	6000	0.0429	0.9863	0.5135	0.7466	0.3913
0.0384	3.4517	7000	0.0411	0.9868	0.5512	0.7418	0.4386
0.0385	3.9448	8000	0.0396	0.9868	0.5391	0.7659	0.4159
0.0343	4.4379	9000	0.0392	0.9870	0.5622	0.7475	0.4505
0.0343	4.9310	10000	0.0383	0.9872	0.5747	0.7490	0.4662
0.0304	5.4241	11000	0.0381	0.9873	0.5883	0.7375	0.4894
0.0299	5.9172	12000	0.0367	0.9877	0.6116	0.7341	0.5242
0.0265	6.4103	13000	0.0374	0.9876	0.6157	0.7219	0.5367
0.0261	6.9034	14000	0.0365	0.9879	0.6179	0.7448	0.5279
0.0236	7.3964	15000	0.0374	0.9877	0.6228	0.7218	0.5476
0.0236	7.8895	16000	0.0372	0.9880	0.6263	0.7356	0.5453
0.0215	8.3826	17000	0.0376	0.9879	0.6326	0.7199	0.5642
0.0216	8.8757	18000	0.0381	0.9878	0.6322	0.7149	0.5666
0.0177	9.3688	19000	0.0377	0.9880	0.6385	0.7205	0.5733

Framework versions

—Transformers 4.44.2
—Pytorch 2.4.1+cu121
—Datasets 2.21.0
—Tokenizers 0.19.1