Makkoen/whisper-large-v3-cit-do005-wd0-lr5e-06-steps2000-FULL6

FULL6

This model is a fine-tuned version of openai/whisper-large-v3 on the 9712 FULL-2024-10-24 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3684
• Wer Ortho: 20.2615
• Wer: 14.4085

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-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6536	0.3661	200	0.4514	25.3730	18.9684
0.5136	0.7323	400	0.4055	22.8309	16.6362
0.4422	1.0984	600	0.3855	21.7595	15.9420
0.3691	1.4645	800	0.3767	22.0561	15.9885
0.3633	1.8307	1000	0.3694	21.1391	15.3002
0.3282	2.1968	1200	0.3701	21.1633	15.3641
0.284	2.5629	1400	0.3657	20.5429	14.6728
0.2819	2.9291	1600	0.3640	20.4431	14.6031
0.2449	3.2952	1800	0.3672	20.4067	14.5247
0.2382	3.6613	2000	0.3684	20.2615	14.4085

Framework versions

• Transformers 4.45.1
• Pytorch 1.13.1+cu117
• Datasets 3.0.1
• Tokenizers 0.20.0