whisper-base-ckb

This model is a fine-tuned version of openai/whisper-base on the razhan/common_voice_ckb_16 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0641
• Wer: 0.1262

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: 192
• eval_batch_size: 128
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• total_train_batch_size: 1152
• total_eval_batch_size: 768
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 2300
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3434	1.09	100	0.3840	0.6054
0.2089	2.17	200	0.2654	0.4740
0.167	3.26	300	0.2246	0.4190
0.1452	4.35	400	0.1964	0.3803
0.1287	5.43	500	0.1788	0.3542
0.1163	6.52	600	0.1650	0.3326
0.1068	7.61	700	0.1560	0.3155
0.1015	8.7	800	0.1489	0.3059
0.0968	9.78	900	0.1440	0.2954
0.0939	10.87	1000	0.1420	0.2918
0.0919	11.96	1100	0.1315	0.2742
0.0839	13.04	1200	0.1217	0.2597
0.0713	14.13	1300	0.1132	0.2371
0.0687	15.22	1400	0.1091	0.2372
0.0647	16.3	1500	0.1022	0.2173
0.059	17.39	1600	0.0967	0.2043
0.0539	18.48	1700	0.0897	0.1929
0.0518	19.57	1800	0.0827	0.1718
0.0495	20.65	1900	0.0787	0.1667
0.0444	21.74	2000	0.0718	0.1469
0.0392	22.83	2100	0.0671	0.1368
0.0335	23.91	2200	0.0645	0.1263
0.0292	25.0	2300	0.0641	0.1262

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.0