whisper-medium-Split-Sentences

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: 0.7461
• Cer: 16.0707
• Wer: 28.2750

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
2.4779	1.0	1353	0.7097	28.0595	46.8534
0.937	2.0	2706	0.6523	23.8101	38.7592
0.7346	3.0	4059	0.6169	23.7184	39.5741
0.6084	4.0	5412	0.6107	21.8429	36.2939
0.5137	5.0	6765	0.6078	22.2236	35.1640
0.4376	6.0	8118	0.6032	22.2375	34.7942
0.3744	7.0	9471	0.6082	19.0684	31.7127
0.321	8.0	10824	0.6157	19.2498	32.3358
0.2739	9.0	12177	0.6352	17.0055	29.7199
0.2351	10.0	13530	0.6357	17.4420	29.5761
0.2016	11.0	14883	0.6552	17.0513	29.4734
0.1731	12.0	16236	0.6640	16.8102	29.4734
0.148	13.0	17589	0.6769	17.2686	30.1171
0.1286	14.0	18942	0.6896	16.5830	28.8297
0.1101	15.0	20295	0.6986	16.5431	28.9872
0.0964	16.0	21648	0.7064	16.3219	28.6380
0.085	17.0	23001	0.7250	16.3956	28.3983
0.0749	18.0	24354	0.7278	16.3418	28.5832
0.0667	19.0	25707	0.7328	16.2282	28.4736
0.06	20.0	27060	0.7461	16.0707	28.2750
0.0552	21.0	28413	0.7518	16.3238	28.5421
0.0503	22.0	29766	0.7600	16.4614	28.5147
0.0467	23.0	31119	0.7666	17.1669	29.6377
0.0445	24.0	32472	0.7667	16.7225	28.9598
0.0428	25.0	33825	0.7691	16.5291	28.7407

Framework versions

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