Modelpublic

Makkoen/whisper-large-cit-synth-do0.15-wd0-lr1e-06

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes5downloads

Model Card

./whisper-large-cit-synth-do0.15-wd0-lr1e-06

This model is a fine-tuned version of openai/whisper-large-v3 on the SF 200 synth 2000 dataset. It achieves the following results on the evaluation set:

—Loss: 1.5469
—Wer: 52.8152

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
—trainbatchsize: 4
—evalbatchsize: 8
—seed: 42
—distributed_type: multi-GPU
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 16
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 100
—training_steps: 200
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.3143	0.0808	10	2.3730	62.3073
2.3217	0.1616	20	2.2910	60.9606
2.4278	0.2424	30	2.1875	59.9059
2.0697	0.3232	40	2.1152	59.2082
1.9675	0.4040	50	2.0410	59.1270
1.891	0.4848	60	1.9668	57.3098
1.7818	0.5657	70	1.8887	56.4660
1.7774	0.6465	80	1.8145	55.1030
1.8454	0.7273	90	1.7588	53.7238
1.6584	0.8081	100	1.7129	53.0586
2.0202	0.8889	110	1.6738	52.4258
1.6375	0.9697	120	1.6416	51.9552
1.3514	1.0505	130	1.6172	51.6307
1.5016	1.1313	140	1.5977	51.1601
1.8013	1.2121	150	1.5811	50.9330
1.4723	1.2929	160	1.5693	52.9937
1.5952	1.3737	170	1.5605	52.8152
1.4724	1.4545	180	1.5527	52.8476
1.4115	1.5354	190	1.5488	52.7990
1.5161	1.6162	200	1.5469	52.8152

Framework versions

—Transformers 4.41.2
—Pytorch 1.13.1+cu117
—Datasets 2.19.2
—Tokenizers 0.19.1