whisper-large-v3-atco2-asr

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

• Loss: 0.7578
• Wer: 29.0480

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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1323	3.57	100	0.5386	20.4626
0.0207	7.14	200	0.5952	40.3025
0.0125	10.71	300	0.5767	25.9342
0.0056	14.29	400	0.6133	20.6406
0.0022	17.86	500	0.6367	30.2936
0.0005	21.43	600	0.6670	21.6637
0.0002	25.0	700	0.6841	22.2420
0.0002	28.57	800	0.6948	23.4431
0.0001	32.14	900	0.7026	23.6210
0.0001	35.71	1000	0.7095	26.0676
0.0001	39.29	1100	0.7153	25.9786
0.0001	42.86	1200	0.7202	25.1335
0.0001	46.43	1300	0.7251	25.3559
0.0001	50.0	1400	0.7295	29.4929
0.0001	53.57	1500	0.7334	25.9786
0.0001	57.14	1600	0.7373	28.6032
0.0001	60.71	1700	0.7402	28.9146
0.0	64.29	1800	0.7427	29.4484
0.0001	67.86	1900	0.7461	29.4484
0.0	71.43	2000	0.7480	32.2509
0.0	75.0	2100	0.7505	32.2064
0.0001	78.57	2200	0.7524	32.2064
0.0	82.14	2300	0.7539	32.2509
0.0	85.71	2400	0.7549	32.3843
0.0	89.29	2500	0.7563	32.2954
0.0	92.86	2600	0.7573	32.3399
0.0	96.43	2700	0.7578	29.0480
0.0	100.0	2800	0.7578	29.0480

Framework versions

• Transformers 4.37.0
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2