trocr-bigram6

This model is a fine-tuned version of cyttic/exp2-frozen-benyehuda-cont on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4464
• Cer: 0.0247
• Wer: 0.0676

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 4650
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
5.3019	0.1290	2000	2.3519	0.2345	0.4579
3.9805	0.2581	4000	1.7519	0.1602	0.3280
2.9960	0.3871	6000	1.3825	0.1162	0.2584
2.3755	0.5161	8000	1.1416	0.0890	0.2057
2.1391	0.6452	10000	1.0513	0.0765	0.1769
1.9699	0.7742	12000	0.8668	0.0592	0.1504
1.8506	0.9032	14000	0.7908	0.0535	0.1356
1.2421	1.0323	16000	0.7315	0.0470	0.1221
1.2665	1.1613	18000	0.6848	0.0439	0.1131
1.1811	1.2903	20000	0.6578	0.0412	0.1070
1.1397	1.4194	22000	0.6194	0.0375	0.1012
1.1381	1.5484	24000	0.5918	0.0347	0.0929
1.0301	1.6774	26000	0.5624	0.0321	0.0868
1.0940	1.8065	28000	0.5368	0.0299	0.0824
0.9872	1.9355	30000	0.5151	0.0287	0.0796
0.6936	2.0645	32000	0.5026	0.0287	0.0780
0.7706	2.1935	34000	0.4897	0.0275	0.0758
0.7341	2.3226	36000	0.4800	0.0281	0.0738
0.7068	2.4516	38000	0.4682	0.0267	0.0718
0.7357	2.5806	40000	0.4607	0.0259	0.0698
0.6858	2.7097	42000	0.4544	0.0258	0.0697
0.7552	2.8387	44000	0.4483	0.0245	0.0675
0.7638	2.9677	46000	0.4461	0.0249	0.0681
0.7467	3.0	46500	0.4464	0.0247	0.0676

Framework versions

• Transformers 5.15.0
• Pytorch 2.11.0+cu128
• Datasets 5.0.1
• Tokenizers 0.22.2