ymoslem/whisper-small-ga2en-v7.3-r

Whisper Small GA-EN Speech Translation

This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, SpokenWords, Tatoeba, and Wikimedia dataset.

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: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.03
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Step	Training Loss	Validation Loss	Bleu	Chrf	Wer
100	2.351300	1.908078	6.23	21.98	117.424584
200	1.831200	1.567262	10.80	30.01	109.770374
300	1.494100	1.421244	13.55	33.48	120.306168
400	1.309900	1.326882	19.54	41.06	77.217470
500	1.069600	1.285087	19.10	40.23	95.767672
600	0.889600	1.255216	16.45	38.59	110.445745
700	0.783700	1.204875	30.43	47.94	63.574966
800	0.653300	1.213243	24.28	43.24	72.219721
900	0.585300	1.246309	29.13	44.89	64.295362
1000	0.471100	1.236493	26.55	45.00	70.688879
1100	0.228500	1.271864	26.43	45.95	73.525439
1200	0.213200	1.297864	28.94	45.72	65.736155
1300	0.184800	1.262461	27.33	46.82	72.264746
1400	0.195100	1.283275	29.74	46.85	64.745610
1500	0.172600	1.252396	29.33	46.39	64.655561
1600	0.164400	1.295237	25.88	45.29	74.110761
1700	0.144700	1.291115	24.49	44.95	71.994597
1800	0.148200	1.260603	29.30	46.22	63.935164
1900	0.115800	1.292712	29.48	46.77	63.980189
2000	0.105900	1.293839	29.15	47.87	67.086898
2100	0.086100	1.295070	28.82	47.20	64.700585
2200	0.035700	1.303162	30.17	47.57	64.745610
2300	0.032700	1.324677	27.02	45.37	73.075191
2400	0.031700	1.286437	30.48	48.45	63.619991
2500	0.031200	1.305244	32.51	49.49	61.278703
2600	0.027300	1.306147	31.86	49.90	63.124719
2700	0.027000	1.321022	32.00	49.30	63.124719
2800	0.023100	1.305768	32.19	49.76	61.413778
2900	0.020800	1.319407	32.07	48.47	62.179199
3000	0.022800	1.318355	32.15	48.49	61.638901

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1