wav2vec2-xslr-tr-testv2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the GENBED - BEM dataset. It achieves the following results on the evaluation set:

• Loss: 0.2789
• Wer: 0.4783

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.1375	100	2.9646	1.0
No log	0.2749	200	0.9689	0.9848
No log	0.4124	300	0.8561	0.9170
No log	0.5498	400	0.7970	0.912
1.9898	0.6873	500	0.8464	0.9258
1.9898	0.8247	600	0.7358	0.8872
1.9898	0.9622	700	0.6374	0.8608
1.9898	1.0997	800	0.5180	0.7297
1.9898	1.2371	900	0.4852	0.7212
0.663	1.3746	1000	0.4840	0.7278
0.663	1.5120	1100	0.4626	0.7135
0.663	1.6495	1200	0.4493	0.676
0.663	1.7869	1300	0.4419	0.6813
0.663	1.9244	1400	0.4306	0.6749
0.5455	2.0619	1500	0.4329	0.6846
0.5455	2.1993	1600	0.4227	0.6685
0.5455	2.3368	1700	0.4097	0.6472
0.5455	2.4742	1800	0.4035	0.6343
0.5455	2.6117	1900	0.4041	0.6304
0.433	2.7491	2000	0.3962	0.6542
0.433	2.8866	2100	0.3601	0.6041
0.433	3.0241	2200	0.3473	0.5864
0.433	3.1615	2300	0.3456	0.5723
0.433	3.2990	2400	0.3380	0.5617
0.3509	3.4364	2500	0.3267	0.5563
0.3509	3.5739	2600	0.3208	0.5570
0.3509	3.7113	2700	0.3124	0.5397
0.3509	3.8488	2800	0.3038	0.5272
0.3509	3.9863	2900	0.2994	0.5254
0.2871	4.1237	3000	0.3073	0.5247
0.2871	4.2612	3100	0.3009	0.5122
0.2871	4.3986	3200	0.2975	0.4953
0.2871	4.5361	3300	0.2898	0.4938
0.2871	4.6735	3400	0.2835	0.4902
0.2198	4.8110	3500	0.2804	0.4802
0.2198	4.9485	3600	0.2789	0.4783

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0