dit-tiny_rvl_cdip_100_examples_per_class_kd_MSE_lr_fix

This model is a fine-tuned version of microsoft/dit-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.4358
• Accuracy: 0.195
• Brier Loss: 0.9035
• Nll: 12.0550
• F1 Micro: 0.195
• F1 Macro: 0.1471
• Ece: 0.1675
• Aurc: 0.6988

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Brier Loss	Nll	F1 Micro	F1 Macro	Ece	Aurc
No log	1.0	25	1.5167	0.07	0.9368	20.8948	0.07	0.0305	0.1106	0.8850
No log	2.0	50	1.5246	0.08	0.9362	21.4368	0.08	0.0346	0.1200	0.8659
No log	3.0	75	1.5053	0.1	0.9340	23.7241	0.1000	0.0522	0.1280	0.8087
No log	4.0	100	1.5097	0.0975	0.9322	17.3004	0.0975	0.0487	0.1220	0.8220
No log	5.0	125	1.4926	0.12	0.9296	16.3893	0.12	0.0600	0.1284	0.7752
No log	6.0	150	1.4838	0.105	0.9273	19.3692	0.1050	0.0356	0.1254	0.7955
No log	7.0	175	1.4729	0.0975	0.9229	18.6899	0.0975	0.0411	0.1134	0.7963
No log	8.0	200	1.4754	0.125	0.9196	17.7842	0.125	0.0676	0.1238	0.7778
No log	9.0	225	1.4725	0.1125	0.9193	16.6572	0.1125	0.0505	0.1254	0.7839
No log	10.0	250	1.4702	0.1175	0.9168	16.3975	0.1175	0.0556	0.1183	0.7638
No log	11.0	275	1.4648	0.1175	0.9169	18.4274	0.1175	0.0558	0.1219	0.7806
No log	12.0	300	1.4660	0.155	0.9166	15.6492	0.155	0.0791	0.1411	0.7512
No log	13.0	325	1.4684	0.16	0.9164	17.1698	0.16	0.1140	0.1519	0.7285
No log	14.0	350	1.4662	0.1175	0.9158	17.6999	0.1175	0.0501	0.1269	0.7637
No log	15.0	375	1.4602	0.1675	0.9143	13.2540	0.1675	0.1153	0.1515	0.7223
No log	16.0	400	1.4556	0.1325	0.9138	13.3868	0.1325	0.0881	0.1323	0.7558
No log	17.0	425	1.4527	0.175	0.9128	11.1983	0.175	0.1334	0.1596	0.7153
No log	18.0	450	1.4535	0.1625	0.9111	17.6046	0.1625	0.1021	0.1435	0.7379
No log	19.0	475	1.4453	0.1825	0.9086	11.8948	0.1825	0.1228	0.1594	0.7098
1.4614	20.0	500	1.4431	0.1525	0.9078	14.2631	0.1525	0.1115	0.1410	0.7293
1.4614	21.0	525	1.4392	0.1825	0.9063	10.7664	0.1825	0.1378	0.1567	0.7058
1.4614	22.0	550	1.4469	0.1775	0.9055	13.4724	0.1775	0.1212	0.1483	0.7107
1.4614	23.0	575	1.4356	0.17	0.9039	11.8141	0.17	0.1232	0.1515	0.7091
1.4614	24.0	600	1.4370	0.1875	0.9039	12.9338	0.1875	0.1384	0.1539	0.7017
1.4614	25.0	625	1.4358	0.195	0.9035	12.0550	0.195	0.1471	0.1675	0.6988

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.12.0
• Tokenizers 0.12.1