zuazo/whisper-large-v2-gl-train

Whisper Large-V2 Galician

This model is a fine-tuned version of openai/whisper-large-v2 on the mozilla-foundation/common_voice_13_0 gl dataset. It achieves the following results on the evaluation set:

• Loss: 0.2981
• Wer: 5.7016

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
• gradient_accumulation_steps: 2
• 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: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0659	5.83	1000	0.1663	5.9593
0.0164	11.66	2000	0.1947	5.5044
0.0069	17.49	3000	0.2165	5.7500
0.0052	23.32	4000	0.2292	5.9212
0.0032	29.15	5000	0.2320	5.8884
0.0037	34.99	6000	0.2434	6.0647
0.0022	40.82	7000	0.2465	6.1114
0.0019	46.65	8000	0.2531	5.8590
0.0009	52.48	9000	0.2567	5.8451
0.001	58.31	10000	0.2718	5.7673
0.0011	64.14	11000	0.2659	6.1045
0.0008	69.97	12000	0.2765	6.0405
0.0006	75.8	13000	0.2793	6.0250
0.0004	81.63	14000	0.2848	6.0025
0.0005	87.46	15000	0.2790	5.9454
0.0002	93.29	16000	0.2884	5.8175
0.0002	99.13	17000	0.2913	5.7898
0.0001	104.96	18000	0.2901	5.7258
0.0001	110.79	19000	0.2991	5.7050
0.0001	116.62	20000	0.2981	5.7016

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1