w2v-bert-punjabi_v2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2031
• Wer: 0.1135

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 60000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4419	0.2174	2000	0.3828	0.2268
0.3492	0.4348	4000	0.3401	0.1836
0.3205	0.6522	6000	0.2932	0.1712
0.2813	0.8696	8000	0.2844	0.1590
0.255	1.0870	10000	0.2562	0.1469
0.2451	1.3043	12000	0.2431	0.1386
0.2305	1.5217	14000	0.2299	0.1312
0.2156	1.7391	16000	0.2191	0.1274
0.2119	1.9565	18000	0.2269	0.1205
0.182	2.1739	20000	0.2091	0.1181
0.1789	2.3913	22000	0.1980	0.1136
0.1766	2.6087	24000	0.1945	0.1092
0.1657	2.8261	26000	0.1881	0.1079
0.1461	3.0435	28000	0.1809	0.1050
0.1454	3.2609	30000	0.1810	0.1029
0.1697	3.4783	32000	0.2085	0.1210
0.1763	3.6957	34000	0.2017	0.1172
0.1642	3.9130	36000	0.2031	0.1135

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0