spermatogenesis-classifier-v2

This model is a fine-tuned version of facebook/dinov2-base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4213
• Accuracy: 0.8506
• F1: 0.8497
• Acc I-iv: 0.6970
• Acc Ix-x: 0.8571
• Acc V-vi: 0.8571
• Acc Vii-vii: 0.9574
• Acc Xi- xii: 0.8387

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Acc I-iv	Acc Ix-x	Acc V-vi	Acc Vii-vii	Acc Xi- xii
0.9761	1.0	25	1.0469	0.5977	0.5529	0.3030	0.0476	0.8571	0.9787	0.3548
0.8169	2.0	50	0.8854	0.6839	0.6571	0.2424	0.9048	0.8810	0.8936	0.4194
0.6579	3.0	75	0.9098	0.6782	0.6655	0.2727	0.9048	0.9762	0.6809	0.5484
0.3899	4.0	100	0.5647	0.7874	0.7882	0.6364	0.9048	0.7857	0.8936	0.7097
0.5074	5.0	125	1.0395	0.6609	0.6200	0.0909	1.0	0.9762	0.7447	0.4839
0.4786	6.0	150	0.5871	0.7816	0.7842	0.6364	0.9048	0.8333	0.8298	0.7097
0.3498	7.0	175	0.8465	0.6897	0.6908	0.7576	0.8095	0.5476	0.5745	0.9032
0.3699	8.0	200	0.4213	0.8506	0.8497	0.6970	0.8571	0.8571	0.9574	0.8387
0.5555	9.0	225	0.5174	0.7874	0.7892	0.6667	0.8571	0.8333	0.8298	0.7419
0.2530	10.0	250	0.7004	0.7989	0.7994	0.7576	0.8571	0.7857	0.9362	0.6129
0.2849	11.0	275	0.5274	0.8103	0.8078	0.6061	0.8095	0.9286	0.9149	0.7097
0.1722	12.0	300	0.5315	0.7931	0.7934	0.7879	0.8571	0.5714	0.9362	0.8387
0.1955	13.0	325	0.9076	0.7644	0.7517	0.4545	0.8571	0.9048	0.9787	0.5161

Framework versions

• Transformers 5.6.2
• Pytorch 2.11.0+cu130
• Tokenizers 0.22.2