whisper-large-v3-turbo-verbatim-3-lora

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

Loss: 0.1459
Wer: 7.7269
Cer: 3.2519
Decode Runtime: 111.0004
Wer Runtime: 0.0705
Cer Runtime: 0.0932

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: 5e-05
train_batch_size: 4
eval_batch_size: 32
seed: 42
gradient_accumulation_steps: 16
total_train_batch_size: 64
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
No log	0	0	2.2169	32.7209	17.9205	106.5404	0.0825	0.1203
0.1681	0.1	100	0.1998	9.9454	4.1038	108.1653	0.0730	0.0960
0.1025	0.2	200	0.1693	8.6885	3.7458	109.6779	0.0707	0.0957
0.2508	0.3	300	0.1590	8.3897	3.4931	110.3209	0.0716	0.0947
0.1446	1.088	400	0.1571	8.2626	3.4939	110.1930	0.0718	0.0951
0.1833	1.188	500	0.1505	8.0463	3.4298	110.3821	0.0709	0.0950
0.1409	1.288	600	0.1489	7.9948	3.3401	110.6880	0.0709	0.0939
0.1184	2.076	700	0.1492	7.9124	3.3181	110.6153	0.0728	0.0946
0.1737	2.176	800	0.1468	7.8128	3.2583	110.7120	0.0714	0.0947
0.1522	2.276	900	0.1462	7.7887	3.2604	110.7694	0.0710	0.0937
0.1077	3.064	1000	0.1459	7.7269	3.2519	111.0004	0.0705	0.0932

Framework versions

PEFT 0.14.0
Transformers 4.48.0
Pytorch 2.4.1+cu124
Datasets 3.2.0
Tokenizers 0.21.0