kiranpantha/w2v-bert-2.0-nepali

Wave2Vec2-Bert2.0 - Kiran Pantha

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the OpenSLR54 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2212
• Wer: 0.2525
• Cer: 0.0565

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: 5e-05
• train_batch_size: 8
• 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: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.4436	0.0900	300	0.5638	0.5560	0.1447
0.5495	0.1800	600	0.6876	0.6171	0.1641
0.6148	0.2699	900	0.6872	0.6211	0.1724
0.564	0.3599	1200	0.5503	0.5162	0.1326
0.4964	0.4499	1500	0.5831	0.5319	0.1318
0.4437	0.5399	1800	0.4913	0.4935	0.1202
0.4441	0.6299	2100	0.4754	0.4764	0.1193
0.3861	0.7199	2400	0.4357	0.4361	0.1055
0.3811	0.8098	2700	0.4282	0.4137	0.0976
0.3754	0.8998	3000	0.3905	0.4069	0.0975
0.3511	0.9898	3300	0.3547	0.3692	0.0863
0.2496	1.0798	3600	0.3297	0.3433	0.0796
0.242	1.1698	3900	0.3125	0.3315	0.0770
0.2378	1.2597	4200	0.3158	0.3336	0.0757
0.2274	1.3497	4500	0.2871	0.3097	0.0722
0.2142	1.4397	4800	0.3010	0.3058	0.0712
0.1949	1.5297	5100	0.2767	0.2944	0.0678
0.198	1.6197	5400	0.2487	0.2824	0.0639
0.1806	1.7097	5700	0.2376	0.2674	0.0612
0.1675	1.7996	6000	0.2293	0.2630	0.0595
0.1671	1.8896	6300	0.2248	0.2581	0.0576
0.1526	1.9796	6600	0.2212	0.2525	0.0565

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1