Nayana-IR-colpali_v1_3-kn-12k-4bit-LoRA

This model is a fine-tuned version of vidore/colpaligemma-3b-pt-448-base on the Nayana-cognitivelab/Nayana-IR-DescVQA-finetune-kn-47k dataset. It achieves the following results on the evaluation set:

• Loss: 0.2632
• Model Preparation Time: 0.005

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	0.0013	1	1.0454	0.005
0.4978	0.128	100	0.5114	0.005
0.4475	0.256	200	0.4115	0.005
0.3772	0.384	300	0.3764	0.005
0.3981	0.512	400	0.3713	0.005
0.3479	0.64	500	0.3283	0.005
0.2673	0.768	600	0.3042	0.005
0.3274	0.896	700	0.2806	0.005
0.1974	1.0230	800	0.2655	0.005
0.2274	1.1510	900	0.2612	0.005
0.1932	1.2790	1000	0.2690	0.005
0.2611	1.4070	1100	0.2658	0.005

Framework versions

• Transformers 4.47.1
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0