tgrhn/wav2vec2-turkish-5

Wav2Vec2-XLS-TR

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Common Voice 17 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3285

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: 32
• eval_batch_size: 8
• seed: 42
• 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
5.5445	0.2757	400	1.2160
0.7341	0.5513	800	0.6683
0.5195	0.8270	1200	0.5313
0.4561	1.1027	1600	0.4837
0.4016	1.3784	2000	0.4725
0.3945	1.6540	2400	0.4570
0.3756	1.9297	2800	0.4284
0.3341	2.2054	3200	0.4283
0.3316	2.4810	3600	0.3945
0.3333	2.7567	4000	0.4283
0.3224	3.0324	4400	0.3993
0.2924	3.3081	4800	0.4216
0.3012	3.5837	5200	0.3729
0.2889	3.8594	5600	0.3962
0.2767	4.1351	6000	0.4037
0.2714	4.4108	6400	0.3740
0.2721	4.6864	6800	0.3821
0.2673	4.9621	7200	0.3580
0.2407	5.2378	7600	0.3758
0.2525	5.5134	8000	0.4067
0.2477	5.7891	8400	0.3675
0.2433	6.0648	8800	0.3653
0.229	6.3405	9200	0.3485
0.2326	6.6161	9600	0.3674
0.2288	6.8918	10000	0.3664
0.2199	7.1675	10400	0.3656
0.2094	7.4431	10800	0.3389
0.2191	7.7188	11200	0.3446
0.2149	7.9945	11600	0.3489
0.1967	8.2702	12000	0.3482
0.2042	8.5458	12400	0.3464
0.2043	8.8215	12800	0.3517
0.1919	9.0972	13200	0.3408
0.1844	9.3728	13600	0.3465
0.1906	9.6485	14000	0.3349
0.1868	9.9242	14400	0.3282
0.1732	10.1999	14800	0.3604
0.1715	10.4755	15200	0.3413
0.1734	10.7512	15600	0.3309
0.1785	11.0269	16000	0.3351
0.1643	11.3025	16400	0.3326
0.1603	11.5782	16800	0.3205
0.1662	11.8539	17200	0.3332
0.1561	12.1296	17600	0.3311
0.1512	12.4052	18000	0.3322
0.1509	12.6809	18400	0.3227
0.1516	12.9566	18800	0.3338
0.1493	13.2323	19200	0.3439
0.1426	13.5079	19600	0.3447
0.143	13.7836	20000	0.3299
0.1398	14.0593	20400	0.3273
0.1351	14.3349	20800	0.3281
0.1384	14.6106	21200	0.3333
0.1335	14.8863	21600	0.3311
0.1291	15.1620	22000	0.3230
0.1259	15.4376	22400	0.3301
0.1294	15.7133	22800	0.3446
0.1269	15.9890	23200	0.3271
0.1196	16.2646	23600	0.3204
0.1166	16.5403	24000	0.3031
0.12	16.8160	24400	0.3258
0.1163	17.0917	24800	0.3408
0.1101	17.3673	25200	0.3246
0.1142	17.6430	25600	0.3201
0.1121	17.9187	26000	0.3198
0.1044	18.1943	26400	0.3441
0.105	18.4700	26800	0.3441
0.1032	18.7457	27200	0.3252
0.104	19.0214	27600	0.3170
0.0968	19.2970	28000	0.3363
0.0946	19.5727	28400	0.3100
0.0974	19.8484	28800	0.3128
0.0889	20.1241	29200	0.3325
0.0887	20.3997	29600	0.3276
0.0891	20.6754	30000	0.3253
0.0937	20.9511	30400	0.3270
0.0854	21.2267	30800	0.3294
0.0864	21.5024	31200	0.3352
0.0864	21.7781	31600	0.3279
0.0829	22.0538	32000	0.3245
0.0799	22.3294	32400	0.3329
0.0811	22.6051	32800	0.3295
0.0777	22.8808	33200	0.3204
0.074	23.1564	33600	0.3286
0.0762	23.4321	34000	0.3326
0.0765	23.7078	34400	0.3061
0.0745	23.9835	34800	0.3313
0.0702	24.2591	35200	0.3131
0.0684	24.5348	35600	0.3236
0.0689	24.8105	36000	0.3152
0.0716	25.0861	36400	0.3272
0.0609	25.3618	36800	0.3294
0.0607	25.6375	37200	0.3366
0.0624	25.9132	37600	0.3215
0.0614	26.1888	38000	0.3236
0.0599	26.4645	38400	0.3321
0.0608	26.7402	38800	0.3311
0.0558	27.0159	39200	0.3368
0.0589	27.2915	39600	0.3217
0.0573	27.5672	40000	0.3320
0.0546	27.8429	40400	0.3316
0.0526	28.1185	40800	0.3285
0.0501	28.3942	41200	0.3287
0.0515	28.6699	41600	0.3261
0.0507	28.9456	42000	0.3316
0.0539	29.2212	42400	0.3285
0.0489	29.4969	42800	0.3319
0.053	29.7726	43200	0.3285

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1