johnlockejrr/GLM-OCR-samaritan

GLM-OCR

👋 Join our WeChat and Discord community
📍 Use GLM-OCR's API

This model is a fine-tuned version of GLM-OCR on the sam_44_mss_pango dataset. It achieves the following results on the evaluation set:

• Loss: 0.2622

Model description

The model was finetuned on 44 medieval Samaritan Hebrew/Aramaic manuscripts.

Training and evaluation data

Trained on a private dataset

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• 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: cosine
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
1.5390	0.7594	2000	1.4977
0.5723	1.5187	4000	0.5357
0.3869	2.2779	6000	0.3955
0.3180	3.0372	8000	0.3480
0.2683	3.7966	10000	0.3113
0.2507	4.5559	12000	0.2898
0.1941	5.3151	14000	0.2783
0.1738	6.0744	16000	0.2722
0.1792	6.8338	18000	0.2622
0.1260	7.5931	20000	0.2662
0.0956	8.3523	22000	0.2667
0.1271	9.1116	24000	0.2672

Framework versions

• PEFT 0.18.1
• Transformers 5.1.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2