glm-ocr-bnk-v3

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

glm-ocr-bnk-v3

This model is a fine-tuned version of zai-org/GLM-OCR on the bnk_ocr_train dataset. It achieves the following results on the evaluation set:

• Loss: 1.4526

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: 5e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 0.1
• num_epochs: 6.0
• label_smoothing_factor: 0.1

Training results

Framework versions

• PEFT 0.18.1
• Transformers 5.2.0
• Pytorch 2.4.1+cu124
• Datasets 4.0.0
• Tokenizers 0.22.2