aipib/LightOnOCR-1B-1025-ft-ja1-full

This model is a fine-tuned version of lightonai/LightOnOCR-1B-1025 on aipib/japanese-text-image-retrieval-small2 (little bit better japanese accuracy than ft-ja).

It achieves the following results on the evaluation set:

-VAL_Loss: 1.5539

Intended uses & limitations Please refer the original information

Training hyperparameters The following hyperparameters were used during training:

learning_rate: 6e-05 train_batch_size: 2 eval_batch_size: 4 seed: 42 gradient_accumulation_steps: 4 total_train_batch_size: 8 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: 10 training_steps: 100

Framework versions PEFT 0.18.0 Transformers 5.0.0.dev0 Pytorch 2.9.1 Datasets 4.4.1 Tokenizers 0.22.1