parambharat/whisper-tiny-kn

Whisper Tiny Kn - Bharat Ramanathan

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3057
• Wer: 46.3937

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: 150
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4091	0.1	300	1.4915	101.5026
1.1294	0.2	600	1.2845	94.7408
0.5426	0.3	900	0.4621	64.2374
0.4128	1.02	1200	0.3695	54.6582
0.3629	1.12	1500	0.3414	52.9677
0.3321	1.22	1800	0.3249	50.3005
0.3066	1.32	2100	0.3181	48.9106
0.2958	2.03	2400	0.3136	47.7836
0.2883	2.13	2700	0.3055	46.6191
0.2857	2.23	3000	0.3057	46.3937

Framework versions

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