finetune_colpali_v1_2-german-4bit

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

• Loss: 0.1100
• Model Preparation Time: 0.008

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time
No log	0.0533	1	0.3717	0.008
1.1358	0.5333	10	0.3356	0.008
1.2182	1.0667	20	0.2811	0.008
0.844	1.6	30	0.2365	0.008
0.7722	2.1333	40	0.1990	0.008
0.4823	2.6667	50	0.1758	0.008
0.46	3.2	60	0.1451	0.008
0.1477	3.7333	70	0.1252	0.008
0.1764	4.2667	80	0.1258	0.008
0.2329	4.8	90	0.1100	0.008

Framework versions

• Transformers 4.46.1
• Pytorch 2.3.1
• Datasets 3.1.0
• Tokenizers 0.20.1