wav2vec2-bert-turkish-2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5150

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.1947	0.1724	1000	1.4438
0.8195	0.3448	2000	1.0187
0.6782	0.5172	3000	0.9245
0.5903	0.6895	4000	0.8630
0.5485	0.8619	5000	0.8007
0.4942	1.0343	6000	0.8140
0.4387	1.2067	7000	0.8115
0.4253	1.3791	8000	0.7229
0.4093	1.5515	9000	0.6622
0.3921	1.7238	10000	0.7313
0.3811	1.8962	11000	0.7775
0.3526	2.0686	12000	0.7081
0.3036	2.2410	13000	0.6260
0.2946	2.4134	14000	0.6579
0.3077	2.5858	15000	0.6672
0.3028	2.7581	16000	0.5932
0.2926	2.9305	17000	0.6578
0.2631	3.1029	18000	0.5469
0.2403	3.2753	19000	0.5683
0.2472	3.4477	20000	0.6180
0.2429	3.6201	21000	0.5719
0.2443	3.7924	22000	0.5890
0.2387	3.9648	23000	0.5849
0.204	4.1372	24000	0.5565
0.1961	4.3096	25000	0.5805
0.1985	4.4820	26000	0.5714
0.2017	4.6544	27000	0.5723
0.2007	4.8268	28000	0.5288
0.2012	4.9991	29000	0.5250
0.1642	5.1715	30000	0.5875
0.1641	5.3439	31000	0.5434
0.1608	5.5163	32000	0.5618
0.1751	5.6887	33000	0.5782
0.1709	5.8611	34000	0.5322
0.1578	6.0334	35000	0.4982
0.1371	6.2058	36000	0.5446
0.1423	6.3782	37000	0.5506
0.1475	6.5506	38000	0.5275
0.1449	6.7230	39000	0.5302
0.1534	6.8954	40000	0.4791
0.1365	7.0677	41000	0.5311
0.1197	7.2401	42000	0.4958
0.1226	7.4125	43000	0.4881
0.1344	7.5849	44000	0.5065
0.1276	7.7573	45000	0.4973
0.1267	7.9297	46000	0.5196
0.1126	8.1021	47000	0.5186
0.1084	8.2744	48000	0.5125
0.1142	8.4468	49000	0.4895
0.1174	8.6192	50000	0.5017
0.1158	8.7916	51000	0.4935
0.1131	8.9640	52000	0.4811
0.0971	9.1364	53000	0.5146
0.098	9.3087	54000	0.5397
0.0999	9.4811	55000	0.5150

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1