Nayana-IR-colpali_v1_3-hi-47k-4bit-LoRA

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

• Loss: 0.1117
• Model Preparation Time: 0.0054

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	0.0004	1	0.4548	0.0054
0.3422	0.0362	100	0.3221	0.0054
0.3327	0.0724	200	0.2744	0.0054
0.2215	0.1085	300	0.2383	0.0054
0.2231	0.1447	400	0.2206	0.0054
0.2427	0.1809	500	0.2126	0.0054
0.1809	0.2171	600	0.1962	0.0054
0.2156	0.2532	700	0.1819	0.0054
0.1941	0.2894	800	0.1720	0.0054
0.147	0.3256	900	0.1681	0.0054
0.1656	0.3618	1000	0.1596	0.0054
0.1854	0.3979	1100	0.1525	0.0054
0.1315	0.4341	1200	0.1434	0.0054
0.1623	0.4703	1300	0.1487	0.0054
0.1645	0.5065	1400	0.1386	0.0054
0.1616	0.5426	1500	0.1351	0.0054
0.1398	0.5788	1600	0.1293	0.0054
0.131	0.6150	1700	0.1243	0.0054
0.1183	0.6512	1800	0.1256	0.0054
0.1529	0.6873	1900	0.1230	0.0054
0.1409	0.7235	2000	0.1354	0.0054
0.1385	0.7597	2100	0.1253	0.0054
0.0969	0.7959	2200	0.1192	0.0054
0.0966	0.8321	2300	0.1174	0.0054
0.1085	0.8682	2400	0.1161	0.0054
0.1455	0.9044	2500	0.1172	0.0054
0.1459	0.9406	2600	0.1132	0.0054
0.1224	0.9768	2700	0.1087	0.0054
0.095	1.0127	2800	0.1106	0.0054
0.0692	1.0488	2900	0.1108	0.0054
0.0866	1.0850	3000	0.1135	0.0054
0.109	1.1212	3100	0.1155	0.0054
0.0928	1.1574	3200	0.1164	0.0054
0.0669	1.1935	3300	0.1159	0.0054
0.073	1.2297	3400	0.1142	0.0054
0.0783	1.2659	3500	0.1126	0.0054
0.0721	1.3021	3600	0.1123	0.0054
0.0742	1.3382	3700	0.1134	0.0054
0.1083	1.3744	3800	0.1129	0.0054
0.056	1.4106	3900	0.1122	0.0054
0.1057	1.4468	4000	0.1120	0.0054
0.1095	1.4830	4100	0.1117	0.0054

Framework versions

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