whisper-large-v3-phase2-test

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6966
• Cer: 16.8717
• Wer: 28.3301

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: 1e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: 1000
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.8187	1.0	2111	0.6914	45.3063	66.9946
0.7214	2.0	4222	0.6309	37.9743	55.6871
0.5298	3.0	6333	0.6026	29.0135	45.6557
0.4164	4.0	8444	0.6037	35.3396	54.7338
0.3355	5.0	10555	0.5956	27.3975	42.6625
0.2702	6.0	12666	0.6047	26.9102	42.2039
0.2202	7.0	14777	0.6023	21.7689	35.8946
0.1806	8.0	16888	0.6053	20.1071	32.7984
0.1496	9.0	18999	0.6262	20.3211	33.2826
0.1227	10.0	21110	0.6374	19.5237	31.9854
0.1013	11.0	23221	0.6532	18.4214	30.6836
0.0859	12.0	25332	0.6505	18.6292	30.9128
0.0728	13.0	27443	0.6658	19.0582	31.7761
0.0629	14.0	29554	0.6691	17.9456	30.1198
0.0549	15.0	31665	0.6693	17.5997	29.5186
0.0479	16.0	33776	0.6894	18.0434	30.0882
0.043	17.0	35887	0.6831	17.4846	29.3805
0.0385	18.0	37998	0.6906	17.9625	29.9607
0.0344	19.0	40109	0.6914	16.9491	28.6015
0.0315	20.0	42220	0.6968	16.9293	28.3968
0.029	21.0	44331	0.6966	16.8717	28.3301
0.0254	22.0	46442	0.6997	16.9834	28.4600
0.0236	23.0	48553	0.6967	16.9067	28.3442
0.0227	24.0	50664	0.7046	16.9137	28.2646
0.0212	25.0	52775	0.7053	16.8825	28.2705

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1