blip2_lora_vqa_model

This model is a fine-tuned version of Salesforce/blip2-flan-t5-xl on the arrow dataset. It achieves the following results on the evaluation set:

• Loss: 0.0612
• Exact: 74.8351
• F1: 78.3950
• Total: 1061
• Hasans Exact: 74.8351
• Hasans F1: 78.3950
• Hasans Total: 1061
• Best Exact: 74.8351
• Best Exact Thresh: 0.0
• Best F1: 78.3950
• Best F1 Thresh: 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: 0.002
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• 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
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Exact	F1	Total	Hasans Exact	Hasans F1	Hasans Total	Best Exact	Best Exact Thresh	Best F1	Best F1 Thresh
No log	1.0	77	0.1769	57.3044	62.1696	1061	57.3044	62.1696	1061	57.3044	0.0	62.1696	0.0
1.8287	2.0	154	0.1209	63.8077	68.4910	1061	63.8077	68.4910	1061	63.8077	0.0	68.4910	0.0
0.1517	3.0	231	0.0959	65.0330	69.5182	1061	65.0330	69.5182	1061	65.0330	0.0	69.5182	0.0
0.1145	4.0	308	0.0863	67.9548	72.6603	1061	67.9548	72.6603	1061	67.9548	0.0	72.6603	0.0
0.1145	5.0	385	0.0788	70.8765	74.6248	1061	70.8765	74.6248	1061	70.8765	0.0	74.6248	0.0
0.0946	6.0	462	0.0697	73.1385	76.8465	1061	73.1385	76.8465	1061	73.1385	0.0	76.8465	0.0
0.0837	7.0	539	0.0724	72.3845	76.2186	1061	72.3845	76.2186	1061	72.3845	0.0	76.2186	0.0
0.069	8.0	616	0.0644	74.3638	77.8250	1061	74.3638	77.8250	1061	74.3638	0.0	77.8250	0.0
0.069	9.0	693	0.0632	73.7041	77.3865	1061	73.7041	77.3865	1061	73.7041	0.0	77.3865	0.0
0.0652	10.0	770	0.0612	74.8351	78.3950	1061	74.8351	78.3950	1061	74.8351	0.0	78.3950	0.0

Framework versions

• PEFT 0.13.2
• Transformers 4.46.3
• Pytorch 2.4.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3