paddleocr-nepali-stage1-fourthrun

This model is a fine-tuned version of strangervisionhf/PaddleOCR-VL-1.5-hf-transformers-v5.2.0.dev0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7513

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• 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: cosine
• lr_scheduler_warmup_steps: 3376
• training_steps: 33768

Training results

Training Loss	Epoch	Step	Validation Loss
0.6124	0.8889	1000	0.2542
0.3151	1.7778	2000	0.1858
0.4257	2.6667	3000	0.1601
0.2122	3.5556	4000	0.1526
0.1546	4.4444	5000	0.1481
0.0744	5.0027	5628	0.1476
0.2220	5.3333	6000	0.1535
0.2524	6.2222	7000	0.1541
0.1360	7.1111	8000	0.1520
0.1078	8.0	9000	0.1484
0.2665	8.8889	10000	0.1480
0.1625	9.7778	11000	0.1563
0.3061	10.6667	12000	0.1567
0.1176	11.5556	13000	0.1616
0.0747	12.4444	14000	0.1657
0.0904	13.3333	15000	0.1672
0.5847	14.2222	16000	0.7794
0.4700	15.1111	17000	0.7652
0.1566	16.0	18000	0.7683
0.4304	16.8889	19000	0.7401
0.3640	17.7778	20000	0.7317
0.4795	18.6667	21000	0.7369
0.2979	19.5556	22000	0.7518
0.2953	20.4444	23000	0.7513

Framework versions

• PEFT 0.19.1
• Transformers 5.7.0
• Pytorch 2.11.0+cu130
• Tokenizers 0.22.2