easycall-whisper-lg-3-Nov29

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: 0.0857
• Wer: 8.1395

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
0.947	0.2151	100	0.2001	26.1628
0.1574	0.4301	200	0.1477	19.4767
0.1189	0.6452	300	0.1046	15.3101
0.098	0.8602	400	0.0869	12.9845
0.0896	1.0753	500	0.0847	12.6938
0.0591	1.2903	600	0.0853	10.6589
0.0699	1.5054	700	0.0784	9.1085
0.0724	1.7204	800	0.0865	11.8217
0.0704	1.9355	900	0.0701	9.1085
0.0508	2.1505	1000	0.0835	9.5930
0.0447	2.3656	1100	0.0760	10.0775
0.0426	2.5806	1200	0.0716	8.8178
0.0535	2.7957	1300	0.0703	10.3682
0.052	3.0108	1400	0.0714	8.6240
0.0336	3.2258	1500	0.0733	22.4806
0.0448	3.4409	1600	0.0616	9.0116
0.0421	3.6559	1700	0.0751	9.1085
0.031	3.8710	1800	0.0723	8.6240
0.0285	4.0860	1900	0.0755	8.3333
0.0233	4.3011	2000	0.0713	7.6550
0.0331	4.5161	2100	0.0880	9.6899
0.0278	4.7312	2200	0.0766	8.4302
0.0342	4.9462	2300	0.0863	10.9496
0.0275	5.1613	2400	0.0929	9.3023
0.0224	5.3763	2500	0.0851	17.7326
0.0232	5.5914	2600	0.0964	10.4651
0.0283	5.8065	2700	0.0766	9.7868
0.0336	6.0215	2800	0.0729	8.5271
0.0202	6.2366	2900	0.0802	8.8178
0.02	6.4516	3000	0.0864	9.2054
0.0203	6.6667	3100	0.0841	10.8527
0.0292	6.8817	3200	0.0811	9.1085
0.0211	7.0968	3300	0.0752	8.7209
0.0161	7.3118	3400	0.0857	8.1395

Framework versions

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