blip2-lora-finetune

This model is a fine-tuned version of Salesforce/blip2-opt-2.7b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8155

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: 2e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 0.05
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss
83.2389	0.5979	50	10.0021
18.5524	1.1913	100	1.9922
9.6493	1.7892	150	1.0606
8.5570	2.3827	200	0.9584
7.3530	2.9806	250	0.9102
8.6532	3.5740	300	0.8919
6.9255	4.1674	350	0.8629
7.2563	4.7653	400	0.8558
7.6426	5.3587	450	0.8349
7.3618	5.9567	500	0.8488
7.1585	6.5501	550	0.8302
8.3336	7.1435	600	0.8298
6.9589	7.7414	650	0.8193
7.0150	8.3348	700	0.8237
7.3567	8.9327	750	0.8238
8.1129	9.5262	800	0.8183
6.8732	10.1196	850	0.8128
6.4406	10.7175	900	0.8155
6.8474	11.3109	950	0.8236
6.5659	11.9088	1000	0.8157
7.1021	12.5022	1050	0.8156
7.2930	13.0957	1100	0.8174
6.9147	13.6936	1150	0.8145
7.0703	14.2870	1200	0.8173
6.6065	14.8849	1250	0.8155

Framework versions

• PEFT 0.18.1
• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2