URDU-ASR-25-EPOCH

This model is a fine-tuned version of Shehryar718/URDU-ASR on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7833
• Wer: 0.4924
• Cer: 0.2059

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: 0.00025
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.5981	1.0	341	0.7487	0.5453	0.2200
0.2559	2.0	683	0.7159	0.5086	0.2077
0.3018	3.0	1024	0.7059	0.5457	0.2325
0.2848	4.0	1366	0.6575	0.5464	0.2350
0.2599	5.0	1707	0.6924	0.5436	0.2346
0.2479	6.0	2049	0.6785	0.5372	0.2254
0.2363	7.0	2390	0.7261	0.5356	0.2284
0.2225	8.0	2732	0.7228	0.5199	0.2268
0.2038	9.0	3073	0.7688	0.5248	0.2218
0.1944	10.0	3415	0.7385	0.5384	0.2298
0.1908	11.0	3756	0.7569	0.5325	0.2283
0.1899	12.0	4098	0.7458	0.5088	0.2106
0.1728	13.0	4439	0.7386	0.5326	0.2236
0.1762	14.0	4781	0.7521	0.5297	0.2265
0.1762	15.0	5122	0.7338	0.5197	0.2184
0.1666	16.0	5464	0.7795	0.5294	0.2149
0.1605	17.0	5805	0.7622	0.5092	0.2211
0.1539	18.0	6147	0.7756	0.5144	0.2132
0.1472	19.0	6488	0.7522	0.4989	0.2094
0.1399	20.0	6830	0.7691	0.5144	0.2171
0.1341	21.0	7171	0.7673	0.4992	0.2079
0.1278	22.0	7513	0.7807	0.4889	0.2005
0.1235	23.0	7854	0.7924	0.4932	0.2060
0.1189	24.0	8196	0.7876	0.4954	0.2060
0.1167	24.96	8525	0.7833	0.4924	0.2059

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.0+cu121
• Datasets 2.14.4
• Tokenizers 0.14.1