finetune_colpali_v1_2-german_ver3-4bit

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: 0.0992
• Model Preparation Time: 0.007

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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time
No log	0.0146	1	0.3248	0.007
1.8501	0.1460	10	0.3090	0.007
1.3107	0.2920	20	0.2642	0.007
0.9799	0.4380	30	0.2404	0.007
0.6309	0.5839	40	0.2107	0.007
0.9043	0.7299	50	0.1864	0.007
1.015	0.8759	60	0.1617	0.007
0.9035	1.0219	70	0.1585	0.007
0.6689	1.1679	80	0.1586	0.007
0.3336	1.3139	90	0.1477	0.007
0.377	1.4599	100	0.1408	0.007
0.5013	1.6058	110	0.1442	0.007
0.2791	1.7518	120	0.1277	0.007
0.3665	1.8978	130	0.1164	0.007
0.4709	2.0438	140	0.1106	0.007
0.2456	2.1898	150	0.1061	0.007
0.152	2.3358	160	0.1045	0.007
0.1813	2.4818	170	0.1007	0.007
0.1594	2.6277	180	0.1010	0.007
0.1856	2.7737	190	0.1005	0.007
0.1788	2.9197	200	0.0992	0.007

Framework versions

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