Modelpublic

markjosims/wav2vec2-large-xls-r-300m-tira-colab

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes7downloads

Model Card

wav2vec2-large-xls-r-300m-tira-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.2681
—Wer: 0.2787

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: 0.0003
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2937	1.45	400	1.0460	0.9297
0.9157	2.9	800	0.5732	0.6728
0.6258	4.35	1200	0.4319	0.5434
0.5114	5.8	1600	0.3822	0.5465
0.4059	7.25	2000	0.3439	0.4700
0.3407	8.7	2400	0.2997	0.4778
0.2938	10.14	2800	0.2956	0.4121
0.2465	11.59	3200	0.2834	0.3537
0.2148	13.04	3600	0.2662	0.3779
0.1711	14.49	4000	0.2724	0.3160
0.1621	15.94	4400	0.2452	0.3571
0.1301	17.39	4800	0.2638	0.2927
0.1119	18.84	5200	0.2724	0.2765
0.1026	20.29	5600	0.2703	0.2986
0.0906	21.74	6000	0.2642	0.2638
0.0785	23.19	6400	0.2653	0.2709
0.0648	24.64	6800	0.2644	0.2669
0.0578	26.09	7200	0.2712	0.3123
0.0514	27.54	7600	0.2703	0.2672
0.0459	28.99	8000	0.2681	0.2787

Framework versions

—Transformers 4.30.2
—Pytorch 2.0.1+cu118
—Datasets 2.13.1
—Tokenizers 0.13.3