zuazo/whisper-large-v3-gl-train

Whisper Large-V3 Galician

This model is a fine-tuned version of openai/whisper-large-v3 on the mozilla-foundation/common_voice_13_0 gl dataset. It achieves the following results on the evaluation set:

• Loss: 0.2735
• Wer: 5.3090

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
0.0761	5.83	1000	0.1531	6.0959
0.0148	11.66	2000	0.1874	5.7327
0.0076	17.49	3000	0.2062	5.7587
0.0035	23.32	4000	0.2196	5.4491
0.0029	29.15	5000	0.2265	5.5892
0.0027	34.99	6000	0.2376	5.8365
0.0028	40.82	7000	0.2396	5.6964
0.0021	46.65	8000	0.2449	5.4820
0.0012	52.48	9000	0.2438	5.4491
0.0014	58.31	10000	0.2490	5.5581
0.0009	64.14	11000	0.2462	5.3696
0.0006	69.97	12000	0.2598	5.6307
0.0008	75.8	13000	0.2543	5.6013
0.0003	81.63	14000	0.2582	5.3609
0.0003	87.46	15000	0.2591	5.3402
0.0003	93.29	16000	0.2657	5.3609
0.0002	99.13	17000	0.2661	5.3869
0.0001	104.96	18000	0.2704	5.3177
0.0001	110.79	19000	0.2750	5.3159
0.0001	116.62	20000	0.2735	5.3090

Framework versions

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