wh-ft-lr5e6-dtstf5-adm-ga1ba16-st15k-v2-evalstp100-pat15-trainvalch

This model is a fine-tuned version of HouraMor/wh-ft-lr5e6-dtstf5-adm-ga1ba16-st15k-v2-evalstp500-pat5 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7545
• Wer: 0.2875
• Cer: 0.2170

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-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch 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: 750
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.2915	0.2008	100	0.5596	0.2856	0.2155
0.2523	0.4016	200	0.5651	0.2701	0.2039
0.2447	0.6024	300	0.5693	0.2784	0.2093
0.2297	0.8032	400	0.5770	0.2784	0.2120
0.3404	1.0040	500	0.5713	0.2770	0.2097
0.1959	1.2048	600	0.6140	0.2955	0.2145
0.1946	1.4056	700	0.6199	0.3093	0.2413
0.1835	1.6064	800	0.6318	0.2903	0.2193
0.212	1.8072	900	0.6096	0.3030	0.2325
0.2341	2.0080	1000	0.6090	0.3507	0.2680
0.122	2.2088	1100	0.6882	0.2816	0.2099
0.1162	2.4096	1200	0.6873	0.4320	0.3549
0.1073	2.6104	1300	0.7016	0.2987	0.2279
0.1109	2.8112	1400	0.6768	0.3260	0.2485
0.091	3.0120	1500	0.6904	0.2987	0.2315
0.053	3.2129	1600	0.7545	0.2875	0.2170

Framework versions

• Transformers 4.55.2
• Pytorch 2.7.0+cu118
• Datasets 2.21.0
• Tokenizers 0.21.4