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

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5288
• Wer: 0.3953

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
6.1396	0.41	400	1.6561	0.9954
0.8075	0.82	800	0.7072	0.7451
0.5006	1.23	1200	0.5985	0.6364
0.4299	1.65	1600	0.5317	0.5837
0.4067	2.06	2000	0.5414	0.5810
0.3401	2.47	2400	0.5073	0.5575
0.3227	2.88	2800	0.4859	0.5252
0.2943	3.29	3200	0.5026	0.5230
0.2842	3.7	3600	0.5288	0.5414
0.2799	4.12	4000	0.5075	0.5207
0.2595	4.53	4400	0.4868	0.5083
0.2595	4.94	4800	0.4971	0.5084
0.2408	5.35	5200	0.5484	0.5214
0.234	5.76	5600	0.5050	0.5274
0.2239	6.17	6000	0.4970	0.4930
0.2152	6.58	6400	0.4761	0.4911
0.2199	7.0	6800	0.4871	0.4889
0.1998	7.41	7200	0.5132	0.4866
0.1924	7.82	7600	0.5250	0.5082
0.1972	8.23	8000	0.4799	0.4793
0.1868	8.64	8400	0.4827	0.4713
0.1853	9.05	8800	0.5079	0.4828
0.1775	9.47	9200	0.5172	0.4923
0.1808	9.88	9600	0.4908	0.4903
0.1663	10.29	10000	0.5111	0.4750
0.1614	10.7	10400	0.5058	0.4710
0.1631	11.11	10800	0.5221	0.4695
0.1541	11.52	11200	0.4975	0.4712
0.1552	11.93	11600	0.4975	0.4734
0.1433	12.35	12000	0.5132	0.4637
0.1464	12.76	12400	0.5267	0.4787
0.1449	13.17	12800	0.4908	0.4548
0.1368	13.58	13200	0.4947	0.4632
0.1392	13.99	13600	0.5301	0.4617
0.1244	14.4	14000	0.5264	0.4611
0.1296	14.81	14400	0.4925	0.4598
0.1254	15.23	14800	0.4840	0.4447
0.12	15.64	15200	0.4828	0.4464
0.1181	16.05	15600	0.5159	0.4893
0.1124	16.46	16000	0.5209	0.4608
0.1154	16.87	16400	0.5097	0.4517
0.1074	17.28	16800	0.5215	0.4383
0.1038	17.7	17200	0.5044	0.4348
0.1073	18.11	17600	0.5017	0.4410
0.0996	18.52	18000	0.5106	0.4445
0.0982	18.93	18400	0.4867	0.4399
0.0927	19.34	18800	0.5314	0.4412
0.0918	19.75	19200	0.4925	0.4268
0.0881	20.16	19600	0.5249	0.4344
0.0856	20.58	20000	0.4953	0.4309
0.0845	20.99	20400	0.5042	0.4257
0.0823	21.4	20800	0.5149	0.4266
0.0817	21.81	21200	0.5095	0.4187
0.0719	22.22	21600	0.5257	0.4249
0.0773	22.63	22000	0.5090	0.4097
0.0724	23.05	22400	0.5340	0.4209
0.0692	23.46	22800	0.5279	0.4148
0.0695	23.87	23200	0.5224	0.4082
0.0681	24.28	23600	0.5344	0.4117
0.0625	24.69	24000	0.5352	0.4040
0.0626	25.1	24400	0.5410	0.4134
0.0599	25.51	24800	0.5344	0.4142
0.0639	25.93	25200	0.5293	0.4020
0.0559	26.34	25600	0.5449	0.4100
0.0581	26.75	26000	0.5245	0.4044
0.0544	27.16	26400	0.5377	0.4034
0.0504	27.57	26800	0.5354	0.4038
0.0521	27.98	27200	0.5295	0.4012
0.0495	28.4	27600	0.5481	0.4014
0.0493	28.81	28000	0.5387	0.3987
0.0512	29.22	28400	0.5311	0.3994
0.049	29.63	28800	0.5288	0.3953

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2