medsiglip-448-vindr-lp

This model is a fine-tuned version of google/medsiglip-448 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7781

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• 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: 5
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.0787	0.1975	40	0.9601
0.9313	0.3951	80	0.9440
0.8649	0.5926	120	0.8996
0.8648	0.7901	160	0.8341
0.8421	0.9877	200	0.8553
0.7576	1.1827	240	0.8185
0.7483	1.3802	280	0.8028
0.7361	1.5778	320	0.7917
0.7264	1.7753	360	0.7943
0.7602	1.9728	400	0.7573
0.6597	2.1679	440	0.7896
0.6092	2.3654	480	0.7708
0.6076	2.5630	520	0.7760
0.5964	2.7605	560	0.7813
0.6047	2.9580	600	0.7781

Framework versions

• Transformers 4.55.0
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.21.4