whisper-small-CV_Fleurs_AMMI_ALFFA-sw-5hrs-v1

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9973
• Wer: 0.3738
• Cer: 0.1499

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-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_hf with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
4.7293	1.0	179	0.2498	1.0815	0.7089
1.5994	2.0	358	0.2279	0.7995	0.5767
0.9176	3.0	537	0.1536	0.6981	0.4359
0.4962	4.0	716	0.1715	0.6860	0.4339
0.2817	5.0	895	0.1346	0.7140	0.3806
0.2062	6.0	1074	0.1892	0.7249	0.4400
0.1654	7.0	1253	0.1203	0.7535	0.3534
0.1504	8.0	1432	0.1317	0.7843	0.3656
0.1499	9.0	1611	0.1677	0.7990	0.4059
0.1459	10.0	1790	0.1339	0.8195	0.3706
0.1409	11.0	1969	0.1334	0.8332	0.3632
0.1191	12.0	2148	0.1154	0.8286	0.3334
0.1046	13.0	2327	0.1385	0.8617	0.3722
0.0965	14.0	2506	0.1605	0.8756	0.4000
0.086	15.0	2685	0.1258	0.8674	0.3489
0.0747	16.0	2864	0.1688	0.8831	0.3948
0.0682	17.0	3043	0.1257	0.8814	0.3398
0.0635	18.0	3222	0.1366	0.9215	0.3569
0.0606	19.0	3401	0.1271	0.9136	0.3477
0.0477	20.0	3580	0.1273	0.9359	0.3456
0.0439	21.0	3759	0.1286	0.9150	0.3441
0.048	22.0	3938	0.1226	0.9309	0.3392
0.0429	23.0	4117	0.9718	0.3416	0.1249
0.0418	24.0	4296	0.9585	0.3389	0.1204
0.042	25.0	4475	0.9693	0.3636	0.1361
0.0353	26.0	4654	0.9719	0.3518	0.1309
0.0329	27.0	4833	0.9545	0.3617	0.1424
0.0288	28.0	5012	0.9910	0.3389	0.1257
0.0298	29.0	5191	0.9686	0.3380	0.1264
0.0242	30.0	5370	0.9915	0.3634	0.1488
0.0225	31.0	5549	0.9842	0.3349	0.1279
0.019	32.0	5728	0.9973	0.3738	0.1499

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1