ymoslem/whisper-medium-ga2en-v1.2

Whisper Medium GA-EN Speech Translation

This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, and SpokenWords dataset. It achieves the following results on the evaluation set:

• Loss: 1.2998
• Bleu: 27.06
• Chrf: 47.61
• Wer: 73.4804

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 0.03
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Bleu	Chrf	Validation Loss	Wer
2.5227	0.05	100	1.05	12.82	2.4253	343.2238
2.4775	0.11	200	10.04	24.39	2.0665	95.2724
2.114	0.16	300	8.79	28.6	1.9792	141.9181
1.9813	0.22	400	17.5	33.84	1.7596	82.8906
1.6979	0.27	500	13.89	33.51	1.6820	115.0383
1.7157	0.32	600	18.54	36.44	1.5795	91.4003
1.3845	0.38	700	19.51	39.03	1.4989	88.7888
1.3803	0.43	800	25.18	40.96	1.4176	69.5182
1.1	0.49	900	28.98	44.78	1.3666	65.9613
1.1843	0.54	1000	27.59	45.91	1.3298	70.4638
1.1317	0.59	1100	1.5018	20.22	41.14	86.9878
1.071	0.65	1200	1.4600	20.67	40.43	85.6371
1.1542	0.7	1300	1.4114	26.84	43.76	69.5182
1.0729	0.76	1400	1.4056	22.98	42.65	78.0729
0.8747	0.81	1500	1.3537	24.65	44.89	73.4804
0.8626	0.86	1600	1.3391	28.0	46.03	68.7978
0.7643	0.92	1700	1.3250	27.23	45.31	70.3287
0.6971	0.97	1800	1.2795	30.05	48.28	65.5110
0.3055	1.02	1900	1.2994	27.41	47.91	71.1842
0.2801	1.08	2000	1.2998	27.06	47.61	73.4804

Framework versions

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