hd-0.2-model

This model is a fine-tuned version of anvitamanne/base-model on the None dataset. It achieves the following results on the evaluation set:

• Loss: 545.6776
• Wer: 0.3934
• Cer: 0.1653

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: AdamW with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
326.2242	0.86	1000	493.4137	0.3996	0.1642
310.8459	1.72	2000	511.3396	0.4079	0.1657
313.6741	2.58	3000	487.3476	0.3989	0.1634
307.0468	3.44	4000	493.8859	0.3970	0.1644
306.6389	4.3	5000	518.9114	0.4146	0.1702
322.4012	5.17	6000	507.5997	0.3936	0.1632
310.2188	6.03	7000	493.5461	0.3967	0.1640
311.2426	6.89	8000	504.9644	0.3973	0.1645
305.9267	7.75	9000	509.5207	0.3924	0.1630
285.1184	8.61	10000	526.2730	0.3976	0.1651
281.5382	9.47	11000	512.3894	0.3921	0.1636
291.3399	10.33	12000	529.6555	0.3951	0.1657
297.8667	11.19	13000	529.5459	0.3925	0.1640
294.463	12.05	14000	528.7574	0.3933	0.1640
282.9003	12.91	15000	536.5745	0.3905	0.1645
275.5425	13.78	16000	546.1980	0.3914	0.1648
273.5143	14.64	17000	545.6776	0.3934	0.1653

Framework versions

• Transformers 4.38.2
• Pytorch 2.7.1+cu118
• Datasets 3.6.0
• Tokenizers 0.15.2