SmolVLM-Base-ocr-isl

This model is a fine-tuned version of HuggingFaceTB/SmolVLM-Base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0420
• Wer: 0.4108
• Cer: 0.4556
• Exact Match: 0.0
• Special Char Acc: 0.0084
• Seq Acc 5: 0.0
• Seq Acc 10: 0.0

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT 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: 50
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Exact Match	Special Char Acc	Seq Acc 5	Seq Acc 10
0.1564	0.1245	500	0.1101	0.4213	0.5448	0.0	0.0140	0.0	0.0
0.0866	0.2490	1000	0.0791	0.3409	0.4947	0.0	0.0112	0.0	0.0
0.1093	0.3735	1500	0.0646	0.4073	0.4989	0.0	0.0140	0.0	0.0
0.1016	0.4979	2000	0.0570	0.3951	0.4507	0.0	0.0056	0.0	0.0
0.1	0.6224	2500	0.0504	0.4318	0.5059	0.0	0.0169	0.0	0.0
0.0777	0.7469	3000	0.0415	0.4248	0.4692	0.0	0.0140	0.0	0.0
0.107	0.8714	3500	0.0427	0.4021	0.4732	0.0	0.0140	0.0	0.0
0.1286	0.9959	4000	0.0420	0.4108	0.4556	0.0	0.0084	0.0	0.0

Framework versions

• PEFT 0.17.1
• Transformers 4.56.2
• Pytorch 2.8.0+cu128
• Datasets 4.1.0
• Tokenizers 0.22.1