All_balanced-lang_tag-whisper-lg-3-Nov27

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: 0.2214
• Wer: 13.8661

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: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5388	0.3210	100	0.8188	34.1758
0.6697	0.6421	200	0.5042	29.0756
0.4549	0.9631	300	0.4162	26.4344
0.3066	1.2841	400	0.3520	22.6776
0.2658	1.6051	500	0.3249	21.6758
0.236	1.9262	600	0.2918	20.3552
0.1403	2.2472	700	0.2942	22.1995
0.1338	2.5682	800	0.2535	17.1903
0.1006	2.8892	900	0.2477	16.8033
0.0793	3.2103	1000	0.2554	17.6913
0.0694	3.5313	1100	0.2404	16.2341
0.0576	3.8523	1200	0.2221	14.9590
0.0404	4.1734	1300	0.2349	16.3707
0.0373	4.4944	1400	0.2329	16.0747
0.0363	4.8154	1500	0.2231	15.2322
0.0332	5.1364	1600	0.2249	14.9590
0.0247	5.4575	1700	0.2312	14.4353
0.024	5.7785	1800	0.2257	14.8679
0.0227	6.0995	1900	0.2423	14.7541
0.0197	6.4205	2000	0.2338	14.8452
0.0159	6.7416	2100	0.2231	14.3670
0.0174	7.0626	2200	0.2236	14.5947
0.0156	7.3836	2300	0.2291	14.7313
0.0148	7.7047	2400	0.2323	16.6667
0.016	8.0257	2500	0.2253	14.3215
0.0113	8.3467	2600	0.2345	15.1639
0.012	8.6677	2700	0.2172	13.6384
0.0097	8.9888	2800	0.2316	15.8698
0.0097	9.3098	2900	0.2267	14.5264
0.0093	9.6308	3000	0.2366	16.6894
0.009	9.9518	3100	0.2320	14.9135
0.009	10.2729	3200	0.2385	15.7104
0.0071	10.5939	3300	0.2432	14.7541
0.0103	10.9149	3400	0.2150	15.0501
0.0078	11.2360	3500	0.2382	13.9572
0.0086	11.5570	3600	0.2334	14.1166
0.0102	11.8780	3700	0.2312	13.9572
0.0079	12.1990	3800	0.2306	14.4353
0.0081	12.5201	3900	0.2214	13.8661

Framework versions

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