sudheer997/final-project

uaspeech-foundation-fintuned

• Loss: 2.5324
• Wer: 1.2855

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: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
41.2984	0.7	500	2.8954	1.0
3.0227	1.4	1000	2.8232	1.0042
2.8283	2.11	1500	2.6291	1.0309
2.5552	2.81	2000	2.2593	1.9170
2.1714	3.51	2500	1.9586	1.9142
1.8537	4.21	3000	1.5725	1.8579
1.6087	4.92	3500	1.2772	1.7426
1.3108	5.62	4000	1.2792	1.6751
1.1652	6.32	4500	1.4565	1.6174
1.0113	7.02	5000	1.1906	1.5626
0.925	7.72	5500	1.4491	1.5260
0.8183	8.43	6000	1.3712	1.5387
0.7118	9.13	6500	1.4713	1.4866
0.6959	9.83	7000	1.3336	1.4318
0.6146	10.53	7500	1.3690	1.4177
0.5655	11.24	8000	1.3789	1.4135
0.4969	11.94	8500	1.5476	1.3966
0.4705	12.64	9000	1.9062	1.3797
0.4387	13.34	9500	1.2711	1.3924
0.4115	14.04	10000	1.6318	1.3769
0.3695	14.75	10500	1.5119	1.3755
0.377	15.45	11000	1.6637	1.3812
0.3788	16.15	11500	1.6636	1.3699
0.3396	16.85	12000	1.6572	1.3418
0.3047	17.56	12500	1.4740	1.3361
0.2804	18.26	13000	2.0885	1.3249
0.2995	18.96	13500	1.9536	1.3235
0.2628	19.66	14000	1.7736	1.3179
0.2703	20.37	14500	2.0018	1.3291
0.2335	21.07	15000	1.7962	1.3221
0.2068	21.77	15500	2.3187	1.3136
0.2311	22.47	16000	2.4853	1.3291
0.2491	23.17	16500	2.1901	1.3024
0.1836	23.88	17000	2.4344	1.2911
0.1823	24.58	17500	2.3705	1.3066
0.1575	25.28	18000	2.1864	1.2897
0.1451	25.98	18500	2.4216	1.2883
0.1502	26.69	19000	2.1780	1.2855
0.1392	27.39	19500	2.4009	1.2925
0.1609	28.09	20000	2.4250	1.2982
0.1066	28.79	20500	2.4433	1.2897
0.1514	29.49	21000	2.5063	1.2855

Framework versions

• Transformers 4.23.1
• Pytorch 1.12.1+cu113
• Datasets 1.18.3
• Tokenizers 0.13.2