breco/whisper-tiny-1000-diverse-audios

Whisper Tiny 1000 Diverse Audios - vfranchis

This model is a fine-tuned version of openai/whisper-tiny on the 1000 diverse audios 1.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1835
• Wer: 42.9577

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 25
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.7684	0.4425	25	2.2485	135.5131
1.5347	0.8850	50	0.9286	75.5533
0.8425	1.3274	75	0.5561	56.4386
0.5722	1.7699	100	0.4103	43.4608
0.3867	2.2124	125	0.3423	40.5433
0.3107	2.6549	150	0.2967	51.0060
0.2931	3.0973	175	0.2656	78.8732
0.2031	3.5398	200	0.2421	57.8471
0.2004	3.9823	225	0.2305	51.8109
0.1254	4.4248	250	0.2198	22.4346
0.1332	4.8673	275	0.2070	22.2334
0.1089	5.3097	300	0.2049	51.4085
0.0627	5.7522	325	0.1988	28.5714
0.0959	6.1947	350	0.1948	31.6901
0.0794	6.6372	375	0.1910	28.8732
0.0696	7.0796	400	0.1879	43.5614
0.0458	7.5221	425	0.1861	43.4608
0.0524	7.9646	450	0.1841	53.5211
0.0453	8.4071	475	0.1832	40.4427
0.0485	8.8496	500	0.1835	42.9577

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1