breco/whisper-tiny-all-audios

Whisper Tiny All Audios - vfranchis

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

• Loss: 0.0368
• Wer: 1.9658

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: 650
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4661	0.05	25	0.5654	20.5212
0.2642	0.1	50	0.1588	8.6859
0.1282	0.15	75	0.1102	6.3494
0.0861	0.2	100	0.0901	4.9068
0.0652	0.25	125	0.0784	4.0738
0.0676	0.3	150	0.0695	3.4490
0.0865	0.35	175	0.0649	3.4185
0.0454	0.4	200	0.0610	3.0477
0.0517	0.45	225	0.0567	2.9664
0.0471	0.5	250	0.0548	2.8344
0.0394	0.55	275	0.0521	2.8648
0.0347	0.6	300	0.0488	2.4585
0.0596	0.65	325	0.0477	2.4483
0.0426	0.7	350	0.0452	2.7836
0.0428	0.75	375	0.0436	2.2401
0.0518	0.8	400	0.0417	2.1181
0.0379	0.85	425	0.0407	2.0928
0.0259	0.9	450	0.0399	1.9861
0.0691	0.95	475	0.0394	2.2096
0.0382	1.0	500	0.0384	2.1131
0.0311	1.05	525	0.0377	1.9810
0.0301	1.1	550	0.0375	1.9404
0.021	1.15	575	0.0371	1.9505
0.0205	1.2	600	0.0369	1.9404
0.0163	1.25	625	0.0369	1.9505
0.018	1.3	650	0.0368	1.9658

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1