whisper-large-v2-atco2-asr

This model is a fine-tuned version of openai/whisper-large-v2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7915
• Wer: 18.7722

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: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 2800

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1333	3.57	100	0.5298	21.8861
0.0338	7.14	200	0.5430	18.8167
0.0132	10.71	300	0.5830	17.9270
0.0067	14.29	400	0.6011	17.6157
0.0009	17.86	500	0.6582	18.8167
0.0004	21.43	600	0.6743	18.7722
0.0003	25.0	700	0.6919	18.4609
0.0004	28.57	800	0.6943	26.6459
0.0004	32.14	900	0.7090	18.5053
0.0002	35.71	1000	0.7212	18.8167
0.0001	39.29	1100	0.7305	18.8612
0.0001	42.86	1200	0.7383	18.6388
0.0001	46.43	1300	0.7451	18.5498
0.0001	50.0	1400	0.7515	18.5498
0.0001	53.57	1500	0.7573	18.5498
0.0001	57.14	1600	0.7622	18.5943
0.0001	60.71	1700	0.7666	18.5943
0.0001	64.29	1800	0.7705	18.5498
0.0001	67.86	1900	0.7744	18.6833
0.0001	71.43	2000	0.7778	18.6833
0.0001	75.0	2100	0.7808	18.7278
0.0001	78.57	2200	0.7837	18.6833
0.0001	82.14	2300	0.7856	18.6388
0.0001	85.71	2400	0.7881	18.6833
0.0001	89.29	2500	0.7896	18.6388
0.0001	92.86	2600	0.7905	18.7278
0.0001	96.43	2700	0.7915	18.8167
0.0001	100.0	2800	0.7915	18.7722

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3