beit-base-finetuned-ade-640-640_alpha0.7_temp3.0_t2

This model is a fine-tuned version of c14kevincardenas/ClimBEiT-t2 on the c14kevincardenas/beta_caller_284_person_crop_seq_withlimb dataset. It achieves the following results on the evaluation set:

• Loss: 0.5170
• Accuracy: 0.8310

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: 5e-05
• 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: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.1461	1.0	180	1.3108	0.3923
0.6458	2.0	360	0.8457	0.6690
0.49	3.0	540	0.7169	0.7441
0.3569	4.0	720	0.6869	0.7569
0.2876	5.0	900	0.6191	0.7757
0.2172	6.0	1080	0.6253	0.7905
0.2017	7.0	1260	0.5801	0.8034
0.187	8.0	1440	0.6530	0.7836
0.1894	9.0	1620	0.6205	0.7984
0.1884	10.0	1800	0.5659	0.8103
0.1529	11.0	1980	0.5854	0.8123
0.1473	12.0	2160	0.5478	0.8192
0.1369	13.0	2340	0.5504	0.8211
0.1334	14.0	2520	0.5204	0.8300
0.1294	15.0	2700	0.5254	0.8330
0.1192	16.0	2880	0.5247	0.8251
0.1183	17.0	3060	0.5274	0.8231
0.1199	18.0	3240	0.5264	0.8271
0.1132	19.0	3420	0.5209	0.8320
0.109	20.0	3600	0.5170	0.8310

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1