ihanif/whisper-small-pashto

Whisper Small Pashto

This model is a fine-tuned version of openai/whisper-small on the google/fleurs ps_af dataset. It achieves the following results on the evaluation set:

• Loss: 1.1800
• Wer: 63.1053

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: 3e-07
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• training_steps: 5200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.0871	14.29	100	2.0102	230.2739
1.465	28.57	200	1.4969	137.2427
1.1617	42.86	300	1.2716	76.3242
1.0019	57.14	400	1.1645	71.3756
0.9052	71.43	500	1.1051	69.7866
0.8334	85.71	600	1.0691	68.2657
0.7838	100.0	700	1.0483	67.1686
0.7539	114.29	800	1.0363	66.4195
0.7377	128.57	900	1.0297	66.2001
0.7325	142.86	1000	1.0277	66.0033
0.6952	157.14	1100	1.0122	65.0575
0.6531	171.43	1200	1.0014	64.4219
0.6189	185.71	1300	0.9945	63.7939
0.5993	200.0	1400	0.9896	63.3550
0.5757	214.29	1500	0.9864	63.2264
0.5601	228.57	1600	0.9845	62.9162
0.5482	242.86	1700	0.9833	62.8178
0.5382	257.14	1800	0.9827	62.8405
0.5325	271.43	1900	0.9823	62.7648
0.5287	285.71	2000	0.9822	62.8178
0.3494	357.14	2500	1.0026	61.6147
0.2287	428.57	3000	1.0533	61.5163
0.1525	500.0	3500	1.1041	62.0536
0.1089	571.43	4000	1.1451	62.5076
0.0871	642.86	4500	1.1704	62.9313
0.0797	714.29	5000	1.1791	63.1659
0.0799	728.57	5100	1.1800	63.1053
0.0791	742.86	5200	1.1803	63.1129

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2