All_balanced-lang_tag-whisper-lg-3-Nov28

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

• Loss: 0.2161
• Wer: 22.5266

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: 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: 100
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1372	0.3210	100	0.5984	37.7038
0.4501	0.6421	200	0.4222	30.6371
0.3637	0.9631	300	0.3593	29.2020
0.1957	1.2841	400	0.3426	28.1366
0.2002	1.6051	500	0.2800	27.3755
0.141	1.9262	600	0.2395	25.1794
0.091	2.2472	700	0.2362	25.3316
0.0758	2.5682	800	0.2275	25.3533
0.0806	2.8892	900	0.2129	23.9835
0.0504	3.2103	1000	0.2195	24.8750
0.0523	3.5313	1100	0.2061	25.8317
0.0445	3.8523	1200	0.2196	26.8319
0.0359	4.1734	1300	0.2280	25.0272
0.0288	4.4944	1400	0.2058	24.0487
0.0286	4.8154	1500	0.2035	23.3746
0.0223	5.1364	1600	0.1932	23.4181
0.0181	5.4575	1700	0.2026	23.6138
0.0202	5.7785	1800	0.1983	22.8963
0.0148	6.0995	1900	0.2193	23.7443
0.0125	6.4205	2000	0.2039	23.3312
0.0148	6.7416	2100	0.2183	23.3312
0.0128	7.0626	2200	0.2158	23.2659
0.009	7.3836	2300	0.2022	22.9398
0.008	7.7047	2400	0.2098	23.5269
0.0156	8.0257	2500	0.2145	23.8095
0.0097	8.3467	2600	0.2070	23.2442
0.0102	8.6677	2700	0.2107	23.2007
0.0095	8.9888	2800	0.2157	23.5703
0.0121	9.3098	2900	0.2046	23.1790
0.0081	9.6308	3000	0.2124	24.7445
0.0085	9.9518	3100	0.2044	22.6788
0.006	10.2729	3200	0.2158	25.3968
0.0089	10.5939	3300	0.2131	25.0489
0.0094	10.9149	3400	0.2204	22.7441
0.009	11.2360	3500	0.2160	23.3964
0.007	11.5570	3600	0.2085	26.2883
0.0087	11.8780	3700	0.2008	23.6138
0.013	12.1990	3800	0.2161	22.5266

Framework versions

• Transformers 4.43.4
• Pytorch 2.4.1
• Datasets 3.0.0
• Tokenizers 0.19.1