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

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.4075
• Wer: 0.2074

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
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.4642	1.07	400	1.4491	0.8681
0.9543	2.14	800	0.5998	0.4982
0.5364	3.21	1200	0.4400	0.3549
0.4236	4.28	1600	0.4348	0.3476
0.3345	5.35	2000	0.3897	0.3000
0.2938	6.42	2400	0.3893	0.3176
0.2502	7.49	2800	0.4306	0.3000
0.2376	8.56	3200	0.4023	0.2939
0.1999	9.63	3600	0.3973	0.2652
0.1859	10.7	4000	0.3701	0.2773
0.1673	11.76	4400	0.4047	0.2661
0.1555	12.83	4800	0.4207	0.2670
0.1385	13.9	5200	0.4110	0.2700
0.13	14.97	5600	0.4209	0.2575
0.1185	16.04	6000	0.4385	0.2582
0.11	17.11	6400	0.4334	0.2461
0.1016	18.18	6800	0.4058	0.2450
0.0913	19.25	7200	0.3923	0.2439
0.0843	20.32	7600	0.4139	0.2434
0.0782	21.39	8000	0.4111	0.2397
0.0732	22.46	8400	0.4116	0.2327
0.0644	23.53	8800	0.4041	0.2327
0.0603	24.6	9200	0.4065	0.2232
0.0553	25.67	9600	0.4198	0.2198
0.0502	26.74	10000	0.4137	0.2172
0.0472	27.81	10400	0.4084	0.2148
0.0455	28.88	10800	0.4116	0.2109
0.0417	29.95	11200	0.4075	0.2074

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2