./openai/whisper-medium.en-cit-do015-wd0-lr1e-06-1000

This model is a fine-tuned version of openai/whisper-medium.en on the SF 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6953
• Wer Ortho: 26.2768
• Wer: 14.7572

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-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• 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: 100
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
No log	0.4444	25	1.5811	45.2632	31.9044
1.7463	0.8889	50	1.3848	39.1033	27.0106
1.7463	1.3333	75	1.2178	35.7505	23.0273
1.3387	1.7778	100	1.0166	36.1014	23.4446
1.3387	2.2222	125	0.8784	31.9298	19.1958
0.988	2.6667	150	0.8340	30.8382	18.4750
0.988	3.1111	175	0.8027	30.3314	17.7162
0.8856	3.5556	200	0.7812	29.6686	17.4127
0.8856	4.0	225	0.7651	30.1365	17.6783
0.7927	4.4444	250	0.7515	29.2008	16.8816
0.7927	4.8889	275	0.7402	28.2651	15.6677
0.7482	5.3333	300	0.7300	27.9922	15.5159
0.7482	5.7778	325	0.7217	27.8752	15.6677
0.7275	6.2222	350	0.7153	27.4854	15.4021
0.7275	6.6667	375	0.7085	27.3684	15.3642
0.7003	7.1111	400	0.7041	26.6277	14.6813
0.7003	7.5556	425	0.7002	26.3158	14.7572
0.6763	8.0	450	0.6973	26.2378	14.6055
0.6763	8.4444	475	0.6963	26.4327	14.7951
0.6687	8.8889	500	0.6953	26.2768	14.7572

Framework versions

• Transformers 4.42.3
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1