alakxender/whisper-large-v3-lg-dv

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1718
• Wer: 59.3103

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: 36
• 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7327	0.0213	300	0.5098	93.9655
0.0845	0.0425	600	0.3085	75.8621
0.0512	0.0213	900	0.2694	70.8621
0.0333	0.0425	1200	0.2489	66.0345
0.0435	0.0638	1500	0.2286	64.4828
0.037	0.0851	1800	0.2187	64.4828
0.0337	0.1063	2100	0.2117	62.7586
0.0303	0.1276	2400	0.2036	60.5172
0.0282	0.1489	2700	0.1898	59.3103
0.0267	0.1701	3000	0.1864	59.8276
0.0272	0.1914	3300	0.1788	59.4828
0.0239	0.2126	3600	0.1759	58.4483
0.0229	0.2339	3900	0.1718	59.3103

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1