v2-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: 0.3972
• Wer: 0.2154

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
0.3672	1.07	400	0.4010	0.3241
0.3467	2.14	800	0.4687	0.3642
0.3223	3.21	1200	0.4495	0.3248
0.3236	4.28	1600	0.4325	0.3289
0.2764	5.35	2000	0.4101	0.3005
0.2554	6.42	2400	0.4211	0.3148
0.2198	7.49	2800	0.4217	0.2946
0.2112	8.56	3200	0.4217	0.2930
0.1813	9.63	3600	0.4110	0.2682
0.1727	10.7	4000	0.3908	0.2791
0.1543	11.76	4400	0.4284	0.2746
0.154	12.83	4800	0.4096	0.2743
0.134	13.9	5200	0.4157	0.2582
0.1207	14.97	5600	0.4057	0.2525
0.1145	16.04	6000	0.4255	0.2498
0.0996	17.11	6400	0.4282	0.2488
0.0971	18.18	6800	0.3763	0.2427
0.0886	19.25	7200	0.3833	0.2473
0.082	20.32	7600	0.3849	0.2402
0.0765	21.39	8000	0.4083	0.2327
0.07	22.46	8400	0.4132	0.2355
0.0601	23.53	8800	0.4124	0.2332
0.0583	24.6	9200	0.3956	0.2248
0.0537	25.67	9600	0.4103	0.2289
0.0487	26.74	10000	0.4050	0.2266
0.0459	27.81	10400	0.3827	0.2173
0.0426	28.88	10800	0.3943	0.2152
0.0386	29.95	11200	0.3972	0.2154

Framework versions

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