3sara/BD_10_epochs_colpali_v1.3

BD_3_epochs_colpali_v1.3

This model is a fine-tuned version of vidore/colpaligemma-3b-pt-448-base on the 3sara/bounding_doc_clean dataset. It achieves the following results on the evaluation set:

• Loss: 0.3903

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0040	1	0.9861
1.6717	0.4012	100	0.4507
1.4046	0.8024	200	0.3875
1.2309	1.2046	300	0.3866
1.3361	1.6058	400	0.3793
1.3497	2.0080	500	0.3778
1.2154	2.4092	600	0.3653
1.2651	2.8104	700	0.3554
1.1827	3.2126	800	0.3432
1.2514	3.6138	900	0.3469
1.0762	4.0160	1000	0.3704
1.1595	4.4173	1100	0.3552
1.2052	4.8185	1200	0.3791
1.2966	5.2207	1300	0.3668
1.1034	5.6219	1400	0.3693
1.0709	6.0241	1500	0.3788
1.1456	6.4253	1600	0.3635
0.9316	6.8265	1700	0.3847
1.1855	7.2287	1800	0.3846
1.246	7.6299	1900	0.3711
0.962	8.0321	2000	0.3791
1.056	8.4333	2100	0.3928
1.0595	8.8345	2200	0.3907
0.9379	9.2367	2300	0.3890
1.1661	9.6379	2400	0.3903

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1