arun100/whisper-small-tr-derived-1

Whisper Small Turkish Derived

This model is a fine-tuned version of alikanakar/whisper-synthesized-turkish-8-hour on the mozilla-foundation/common_voice_16_0 tr dataset. It achieves the following results on the evaluation set:

• Loss: 0.2859
• Wer: 18.6250

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: 5e-07
• train_batch_size: 16
• eval_batch_size: 32
• 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: 100
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3197	0.58	200	0.3517	21.5757
0.2672	1.16	400	0.3203	20.6724
0.2532	1.75	600	0.3065	19.6496
0.2382	2.33	800	0.2991	19.3420
0.2448	2.91	1000	0.2943	19.1276
0.2197	3.49	1200	0.2909	18.9712
0.2159	4.07	1400	0.2885	18.7340
0.2212	4.65	1600	0.2871	18.6988
0.2114	5.24	1800	0.2861	18.6320
0.2092	5.82	2000	0.2859	18.6250

Framework versions

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