torgo_tiny_finetune_M03_frozen_encoder

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

• Loss: 0.3051
• Wer: 41.5959

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7806	0.85	500	0.2631	52.1222
0.0945	1.71	1000	0.2804	34.4652
0.071	2.56	1500	0.2464	22.5806
0.0455	3.41	2000	0.2476	21.3073
0.0335	4.27	2500	0.2581	21.2224
0.0253	5.12	3000	0.2617	25.0424
0.0177	5.97	3500	0.2898	26.4007
0.0127	6.83	4000	0.3068	24.5331
0.0111	7.68	4500	0.2925	41.9355
0.0087	8.53	5000	0.3179	23.2598
0.0064	9.39	5500	0.2884	29.8812
0.0056	10.24	6000	0.2952	35.4839
0.0037	11.09	6500	0.2956	26.4007
0.0035	11.95	7000	0.2839	27.3345
0.0028	12.8	7500	0.2975	28.3531
0.0019	13.65	8000	0.3129	42.3599
0.0018	14.51	8500	0.2932	31.5789
0.0015	15.36	9000	0.3047	32.0883
0.0008	16.21	9500	0.3071	37.4363
0.0008	17.06	10000	0.3081	39.8981
0.0006	17.92	10500	0.3064	39.5586
0.0003	18.77	11000	0.3052	40.2377
0.0002	19.62	11500	0.3051	41.5959

Framework versions

• Transformers 4.32.0
• Pytorch 2.1.0+cu121
• Datasets 2.14.7
• Tokenizers 0.13.3