ihanif/content

Whisper Base Pashto - Augmented

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

• Loss: 0.8723
• Wer: 57.6120
• Cer: 26.6468

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 30
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.9708	2.38	100	0.8821	64.0133	27.3253
0.7477	4.75	200	0.8062	59.9576	26.4079
0.6229	7.14	300	0.7855	58.3081	26.3193
0.4833	9.52	400	0.7870	57.5288	24.8855
0.4084	11.89	500	0.7980	56.5224	25.2214
0.3323	14.28	600	0.8201	56.6662	25.3317
0.283	16.66	700	0.8406	57.7406	26.8674
0.2598	19.05	800	0.8538	57.2866	26.0386
0.2235	21.42	900	0.8697	58.2703	26.6819
0.2202	23.8	1000	0.8723	57.6120	26.6468

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu116
• Datasets 2.8.1.dev0
• Tokenizers 0.13.2