paligemma-imgtojson

This model is a fine-tuned version of google/paligemma-3b-pt-224 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4194
• Json Validity: 0.0
• Field Match Avg: 0.0

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: 0.0002
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Json Validity	Field Match Avg
1.6843	0.04	4	1.6791	0.0	0.0
1.2435	0.08	8	1.1588	0.0	0.0
0.9726	0.12	12	0.8497	0.0	0.0
0.3672	0.16	16	0.6707	0.0	0.0
0.8473	0.2	20	0.5506	0.0	0.0
0.3753	0.24	24	0.4935	0.0	0.0
0.5084	0.28	28	0.4973	0.0	0.0
0.3072	0.32	32	0.4974	0.0	0.0
0.2135	0.36	36	0.4716	0.0	0.0
0.3368	0.4	40	0.4641	0.0	0.0
0.4291	0.44	44	0.4585	0.0	0.0
0.334	0.48	48	0.4150	0.0	0.0
0.2791	0.52	52	0.4408	0.0	0.0
0.287	0.56	56	0.4097	0.0	0.0
0.2269	0.6	60	0.3962	0.0	0.0
0.3634	0.64	64	0.3881	0.0	0.0
0.352	0.68	68	0.3956	0.0	0.0
0.2697	0.72	72	0.3953	0.0	0.0
0.3239	0.76	76	0.3821	0.0	0.0
0.2503	0.8	80	0.3809	0.0	0.0
0.278	0.84	84	0.3980	0.2	1.0
0.2364	0.88	88	0.4126	0.0	0.0
0.1774	0.92	92	0.4199	0.0	0.0
0.5384	0.96	96	0.4227	0.0	0.0
0.21	1.0	100	0.4168	0.0	0.0
0.2008	1.04	104	0.3820	0.0	0.0
0.0569	1.08	108	0.3619	0.0	0.0
0.1999	1.12	112	0.3687	0.0	0.0
0.1869	1.16	116	0.3742	0.0	0.0
0.2146	1.2	120	0.3774	0.0	0.0
0.2165	1.24	124	0.3805	0.0	0.0
0.1742	1.28	128	0.3623	0.0	0.0
0.392	1.32	132	0.3592	0.0	0.0
0.2822	1.3600	136	0.3600	0.0	0.0
0.1651	1.4	140	0.3579	0.0	0.0
0.259	1.44	144	0.3640	0.0	0.0
0.2883	1.48	148	0.3653	0.0	0.0
0.0865	1.52	152	0.3669	0.0	0.0
0.1842	1.56	156	0.3732	0.0	0.0
0.1547	1.6	160	0.3852	0.0	0.0
0.1551	1.6400	164	0.3732	0.0	0.0
0.0667	1.6800	168	0.3655	0.0	0.0
0.0763	1.72	172	0.3654	0.0	0.0
0.1045	1.76	176	0.3698	0.0	0.0
0.1016	1.8	180	0.3681	0.0	0.0
0.1273	1.8400	184	0.3711	0.0	0.0
0.0772	1.88	188	0.3717	0.0	0.0
0.123	1.92	192	0.3775	0.0	0.0
0.1533	1.96	196	0.3606	0.0	0.0
0.0748	2.0	200	0.3529	0.0	0.0
0.1028	2.04	204	0.3554	0.0	0.0
0.0557	2.08	208	0.3690	0.0	0.0
0.134	2.12	212	0.3848	0.0	0.0
0.0785	2.16	216	0.3964	0.0	0.0
0.0602	2.2	220	0.4010	0.0	0.0
0.0426	2.24	224	0.4038	0.0	0.0
0.0211	2.2800	228	0.4033	0.0	0.0
0.0632	2.32	232	0.4067	0.0	0.0
0.1438	2.36	236	0.4087	0.0	0.0
0.0917	2.4	240	0.4094	0.0	0.0
0.0292	2.44	244	0.4128	0.0	0.0
0.0339	2.48	248	0.4189	0.0	0.0
0.0619	2.52	252	0.4250	0.0	0.0
0.0252	2.56	256	0.4229	0.0	0.0
0.1563	2.6	260	0.4226	0.0	0.0
0.0258	2.64	264	0.4277	0.0	0.0
0.1214	2.68	268	0.4225	0.0	0.0
0.0562	2.7200	272	0.4237	0.0	0.0
0.1413	2.76	276	0.4195	0.0	0.0
0.0922	2.8	280	0.4172	0.0	0.0
0.0061	2.84	284	0.4155	0.0	0.0
0.0863	2.88	288	0.4220	0.0	0.0
0.059	2.92	292	0.4154	0.0	0.0
0.0188	2.96	296	0.4162	0.0	0.0
0.2014	3.0	300	0.4194	0.0	0.0

Framework versions

• PEFT 0.15.2
• Transformers 4.47.1
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1