trocr_output_tibetan

This model is a fine-tuned version of microsoft/trocr-base-stage1 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3780
• Cer: 0.0914

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: 24
• eval_batch_size: 24
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
4.1754	0.0570	1000	4.0170	0.6859
2.1358	0.1140	2000	1.9342	0.3521
1.5611	0.1709	3000	1.3803	0.2512
1.3302	0.2279	4000	1.1365	0.2227
1.1461	0.2849	5000	0.9782	0.1976
1.051	0.3419	6000	0.8757	0.1837
0.9177	0.3989	7000	0.7911	0.1694
0.8402	0.4558	8000	0.7497	0.1555
0.7918	0.5128	9000	0.6986	0.1461
0.7888	0.5698	10000	0.6830	0.1464
0.7612	0.6268	11000	0.6241	0.1379
0.6737	0.6838	12000	0.6116	0.1306
0.6175	0.7407	13000	0.5877	0.1288
0.6349	0.7977	14000	0.5801	0.1281
0.6229	0.8547	15000	0.5456	0.1191
0.6241	0.9117	16000	0.5391	0.1194
0.5518	0.9687	17000	0.5126	0.1163
0.5301	1.0256	18000	0.5026	0.1149
0.4924	1.0826	19000	0.4964	0.1143
0.5149	1.1396	20000	0.4740	0.1071
0.5005	1.1966	21000	0.4699	0.1060
0.5093	1.2536	22000	0.4516	0.1056
0.4885	1.3105	23000	0.4488	0.1060
0.4964	1.3675	24000	0.4400	0.1023
0.4859	1.4245	25000	0.4330	0.1012
0.4654	1.4815	26000	0.4220	0.1023
0.4825	1.5385	27000	0.4150	0.1009
0.4226	1.5954	28000	0.4138	0.0973
0.4564	1.6524	29000	0.4078	0.0973
0.4458	1.7094	30000	0.3985	0.0948
0.4532	1.7664	31000	0.3916	0.0928
0.4492	1.8234	32000	0.3896	0.0932
0.4503	1.8803	33000	0.3836	0.0927
0.4321	1.9373	34000	0.3823	0.0915
0.4113	1.9943	35000	0.3780	0.0914

Framework versions

• Transformers 4.45.2
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.20.3