finetune_colpali-v1_3-4bit_v2

This model is a fine-tuned version of vidore/colpaligemma-3b-pt-448-base on the RowekBrah/ColPali_ann_rep_v2_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0670
• Model Preparation Time: 0.0056

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 OptimizerNames.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: 1.5

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time
No log	0.0012	1	0.2103	0.0056
0.4896	0.1238	100	0.1170	0.0056
0.5575	0.2476	200	0.0940	0.0056
0.3973	0.3714	300	0.0920	0.0056
0.4478	0.4952	400	0.0836	0.0056
0.2364	0.6190	500	0.0808	0.0056
0.2158	0.7428	600	0.0742	0.0056
0.339	0.8666	700	0.0700	0.0056
0.2052	0.9904	800	0.0704	0.0056
0.1546	1.1151	900	0.0672	0.0056
0.2003	1.2389	1000	0.0672	0.0056
0.1242	1.3627	1100	0.0676	0.0056
0.283	1.4865	1200	0.0674	0.0056

Framework versions

• Transformers 4.50.0
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1