simpragma/breeze-listen-dsw-base-kn

Breeze DSW Kannada - base

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

• Loss: 0.2258
• Wer: 30.6127

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: 32
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7196	1.03	100	0.5166	55.2130
0.2769	2.06	200	0.2532	36.1594
0.1896	4.02	300	0.2167	32.7298
0.1384	5.04	400	0.2037	31.8356
0.1099	7.0	500	0.2030	31.0560
0.0707	8.03	600	0.2153	31.2453
0.052	9.06	700	0.2258	30.6127
0.0375	11.02	800	0.2413	31.2204
0.0256	12.05	900	0.2507	31.0635
0.0245	14.01	1000	0.2549	31.1059

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0