./whisper-large-cit-synth-do015-wd0-lr1e-06-1000

This model is a fine-tuned version of openai/whisper-large-v3 on the SF 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3706
• Wer: 23.6647

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
No log	0.4444	25	0.7983	35.9064
0.967	0.8889	50	0.6724	32.3977
0.967	1.3333	75	0.5459	30.7602
0.6804	1.7778	100	0.4692	27.4854
0.6804	2.2222	125	0.4341	26.3548
0.5145	2.6667	150	0.4143	25.5361
0.5145	3.1111	175	0.4019	25.4191
0.4614	3.5556	200	0.3914	25.0292
0.4614	4.0	225	0.3879	24.4444
0.3891	4.4444	250	0.3835	24.6784
0.3891	4.8889	275	0.3794	24.6004
0.3765	5.3333	300	0.3772	24.0156
0.3765	5.7778	325	0.3745	23.4308
0.3511	6.2222	350	0.3726	23.5478
0.3511	6.6667	375	0.3713	23.5867
0.3307	7.1111	400	0.3706	23.4308
0.3307	7.5556	425	0.3699	23.1189
0.3176	8.0	450	0.3706	23.3918
0.3176	8.4444	475	0.3708	23.6647
0.31	8.8889	500	0.3706	23.6647

Framework versions

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