ymoslem/whisper-base-ga2en-v1.2

Whisper Base GA-EN Speech Translation

This model is a fine-tuned version of openai/whisper-base on an unknown dataset. The best model (this version) is at checkpoint 1000, epoch 2.54, and it achieves the following results on the evaluation set:

• Loss: 1.9005
• Bleu: 21.83
• Chrf: 37.13
• Wer: 80.4593

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Experiment

• Data (v1.1: IWSLT2023-GA-EN; v1.2: +FLEURS-GA-EN)

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.03
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
2.6826	0.25	100	2.0993	7.23	22.29	100.7654
2.1287	0.51	200	1.9451	9.37	27.74	125.9343
1.8482	0.76	300	1.8356	13.11	30.65	103.5570
1.2977	1.02	400	1.8643	10.56	30.86	128.5907
0.8068	1.27	500	1.8658	18.23	35.17	82.6204
0.7257	1.52	600	1.8493	17.81	34.13	90.7249
0.6202	1.78	700	1.8312	17.6	35.19	92.2107
0.4348	2.03	800	1.8771	17.9	35.66	91.9856
0.2566	2.28	900	1.9088	20.14	36.79	81.4498
0.2301	2.54	1000	1.9005	21.83	37.13	80.4593

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2