whisper-large-v2-30s-final

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: 0.5711
• Cer: 14.4843
• Wer: 25.0120

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: 8
• 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: 1000
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.2819	1.0	1737	0.5189	23.9878	39.7700
0.7333	2.0	3474	0.5002	22.7616	36.0189
0.5886	3.0	5211	0.4789	21.2654	34.8689
0.4846	4.0	6948	0.4797	18.3889	30.3922
0.4034	5.0	8685	0.4723	21.4274	33.6368
0.3401	6.0	10422	0.4861	16.6427	28.2360
0.2898	7.0	12159	0.4987	15.9506	27.2914
0.2442	8.0	13896	0.5033	15.9706	27.7637
0.2083	9.0	15633	0.5140	15.2464	26.1003
0.1797	10.0	17370	0.5105	15.3605	25.9840
0.1551	11.0	19107	0.5205	15.0444	25.8402
0.1334	12.0	20844	0.5297	14.8864	25.5459
0.1169	13.0	22581	0.5394	15.0624	26.1209
0.1008	14.0	24318	0.5416	15.2704	26.0730
0.0895	15.0	26055	0.5511	14.8824	25.5938
0.0802	16.0	27792	0.5500	15.0644	26.2920
0.0721	17.0	29529	0.5600	14.6583	25.2721
0.0651	18.0	31266	0.5627	15.0064	25.7376
0.0592	19.0	33003	0.5649	14.9904	25.9634
0.0547	20.0	34740	0.5644	14.5583	25.1352
0.0509	21.0	36477	0.5662	14.6303	25.0873
0.0469	22.0	38214	0.5705	14.8204	25.2721
0.0444	23.0	39951	0.5711	14.4843	25.0120
0.0425	24.0	41688	0.5729	14.6563	25.1968
0.0422	25.0	43425	0.5718	14.5823	25.0667

Framework versions

• Transformers 4.53.3
• Pytorch 2.7.1+cu118
• Datasets 3.6.0
• Tokenizers 0.21.2