whisper-large-v2-phase2

This model is a fine-tuned version of openai/whisper-large-v2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6045
• Cer: 15.0612
• Wer: 25.5404

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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
1.0856	1.0	1056	0.4609	28.0711	44.4346
0.586	2.0	2112	0.4488	22.5153	36.5941
0.4423	3.0	3168	0.4476	22.8514	37.4503
0.3508	4.0	4224	0.4556	18.8734	31.5375
0.2819	5.0	5280	0.4645	17.8030	29.7958
0.229	6.0	6336	0.4680	17.3063	29.2004
0.1869	7.0	7392	0.4826	17.4992	29.2425
0.1519	8.0	8448	0.4988	16.4930	27.7301
0.1263	9.0	9504	0.5140	15.9626	27.0750
0.105	10.0	10560	0.5260	15.4014	26.3931
0.0888	11.0	11616	0.5388	16.1010	27.3195
0.0748	12.0	12672	0.5461	15.5794	26.5042
0.0656	13.0	13728	0.5575	15.3598	26.1159
0.0581	14.0	14784	0.5625	15.5086	26.2820
0.0509	15.0	15840	0.5690	15.5266	26.3849
0.0451	16.0	16896	0.5827	15.6307	26.4481
0.0405	17.0	17952	0.5834	15.3782	26.1755
0.0373	18.0	19008	0.5840	15.3487	26.0820
0.0346	19.0	20064	0.5952	15.2204	25.9463
0.0316	20.0	21120	0.5997	15.2929	26.0247
0.0295	21.0	22176	0.5972	15.2908	25.8855
0.027	22.0	23232	0.6036	15.1715	25.7077
0.0251	23.0	24288	0.6007	15.1156	25.6316
0.0238	24.0	25344	0.6025	15.0851	25.5673
0.0229	25.0	26400	0.6045	15.0612	25.5404

Framework versions

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