whisper-medium-SplitEndMovie-specAug

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

• Loss: 0.7327
• Cer: 20.4118
• Wer: 32.4937

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
2.7084	1.0	1277	0.8031	39.2754	59.3129
1.0556	2.0	2554	0.7228	36.8260	54.1170
0.8311	3.0	3831	0.6839	31.9316	50.1252
0.7048	4.0	5108	0.6720	28.7325	44.4101
0.6069	5.0	6385	0.6596	30.0426	44.3696
0.532	6.0	7662	0.6573	26.4811	40.8860
0.4694	7.0	8939	0.6568	28.1342	43.4747
0.413	8.0	10216	0.6549	25.1352	38.5734
0.3685	9.0	11493	0.6674	24.3855	38.2825
0.3248	10.0	12770	0.6708	25.1352	39.1847
0.2911	11.0	14047	0.6807	22.8741	35.9074
0.2595	12.0	15324	0.6819	22.0941	34.8063
0.2355	13.0	16601	0.6879	22.4868	35.8042
0.2117	14.0	17878	0.6964	21.6051	33.9557
0.194	15.0	19155	0.7067	21.5586	33.7237
0.1765	16.0	20432	0.7162	21.7814	34.0035
0.1634	17.0	21709	0.7281	21.6678	33.3112
0.1502	18.0	22986	0.7353	20.8575	32.9099
0.1405	19.0	24263	0.7327	20.4118	32.4937
0.1305	20.0	25540	0.7430	21.1409	33.4769
0.1243	21.0	26817	0.7497	21.1637	33.2486
0.1191	22.0	28094	0.7489	20.5708	32.5968
0.1143	23.0	29371	0.7519	20.9808	33.0424
0.1098	24.0	30648	0.7525	20.8932	32.8509

Framework versions

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